/[pcre]/code/trunk/pcre_compile.c
ViewVC logotype

Diff of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log | View Patch Patch

revision 85 by nigel, Sat Feb 24 21:41:13 2007 UTC revision 638 by ph10, Mon Jul 25 09:41:19 2011 UTC
# Line 6  Line 6 
6  and semantics are as close as possible to those of the Perl 5 language.  and semantics are as close as possible to those of the Perl 5 language.
7    
8                         Written by Philip Hazel                         Written by Philip Hazel
9             Copyright (c) 1997-2005 University of Cambridge             Copyright (c) 1997-2011 University of Cambridge
10    
11  -----------------------------------------------------------------------------  -----------------------------------------------------------------------------
12  Redistribution and use in source and binary forms, with or without  Redistribution and use in source and binary forms, with or without
# Line 42  POSSIBILITY OF SUCH DAMAGE. Line 42  POSSIBILITY OF SUCH DAMAGE.
42  supporting internal functions that are not used by other modules. */  supporting internal functions that are not used by other modules. */
43    
44    
45    #ifdef HAVE_CONFIG_H
46    #include "config.h"
47    #endif
48    
49    #define NLBLOCK cd             /* Block containing newline information */
50    #define PSSTART start_pattern  /* Field containing processed string start */
51    #define PSEND   end_pattern    /* Field containing processed string end */
52    
53  #include "pcre_internal.h"  #include "pcre_internal.h"
54    
55    
56  /* When DEBUG is defined, we need the pcre_printint() function, which is also  /* When PCRE_DEBUG is defined, we need the pcre_printint() function, which is
57  used by pcretest. DEBUG is not defined when building a production library. */  also used by pcretest. PCRE_DEBUG is not defined when building a production
58    library. */
59    
60  #ifdef DEBUG  #ifdef PCRE_DEBUG
61  #include "pcre_printint.src"  #include "pcre_printint.src"
62  #endif  #endif
63    
64    
65    /* Macro for setting individual bits in class bitmaps. */
66    
67    #define SETBIT(a,b) a[b/8] |= (1 << (b%8))
68    
69    /* Maximum length value to check against when making sure that the integer that
70    holds the compiled pattern length does not overflow. We make it a bit less than
71    INT_MAX to allow for adding in group terminating bytes, so that we don't have
72    to check them every time. */
73    
74    #define OFLOW_MAX (INT_MAX - 20)
75    
76    
77  /*************************************************  /*************************************************
78  *      Code parameters and static tables         *  *      Code parameters and static tables         *
79  *************************************************/  *************************************************/
80    
81  /* Maximum number of items on the nested bracket stacks at compile time. This  /* This value specifies the size of stack workspace that is used during the
82  applies to the nesting of all kinds of parentheses. It does not limit  first pre-compile phase that determines how much memory is required. The regex
83  un-nested, non-capturing parentheses. This number can be made bigger if  is partly compiled into this space, but the compiled parts are discarded as
84  necessary - it is used to dimension one int and one unsigned char vector at  soon as they can be, so that hopefully there will never be an overrun. The code
85  compile time. */  does, however, check for an overrun. The largest amount I've seen used is 218,
86    so this number is very generous.
87    
88    The same workspace is used during the second, actual compile phase for
89    remembering forward references to groups so that they can be filled in at the
90    end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
91    is 4 there is plenty of room. */
92    
93    #define COMPILE_WORK_SIZE (4096)
94    
95  #define BRASTACK_SIZE 200  /* The overrun tests check for a slightly smaller size so that they detect the
96    overrun before it actually does run off the end of the data block. */
97    
98    #define WORK_SIZE_CHECK (COMPILE_WORK_SIZE - 100)
99    
100    
101  /* Table for handling escaped characters in the range '0'-'z'. Positive returns  /* Table for handling escaped characters in the range '0'-'z'. Positive returns
# Line 72  are simple data values; negative values Line 103  are simple data values; negative values
103  on. Zero means further processing is needed (for things like \x), or the escape  on. Zero means further processing is needed (for things like \x), or the escape
104  is invalid. */  is invalid. */
105    
106  #if !EBCDIC   /* This is the "normal" table for ASCII systems */  #ifndef EBCDIC
107    
108    /* This is the "normal" table for ASCII systems or for EBCDIC systems running
109    in UTF-8 mode. */
110    
111  static const short int escapes[] = {  static const short int escapes[] = {
112       0,      0,      0,      0,      0,      0,      0,      0,   /* 0 - 7 */       0,                       0,
113       0,      0,    ':',    ';',    '<',    '=',    '>',    '?',   /* 8 - ? */       0,                       0,
114     '@', -ESC_A, -ESC_B, -ESC_C, -ESC_D, -ESC_E,      0, -ESC_G,   /* @ - G */       0,                       0,
115       0,      0,      0,      0,      0,      0,      0,      0,   /* H - O */       0,                       0,
116  -ESC_P, -ESC_Q,      0, -ESC_S,      0,      0,      0, -ESC_W,   /* P - W */       0,                       0,
117  -ESC_X,      0, -ESC_Z,    '[',   '\\',    ']',    '^',    '_',   /* X - _ */       CHAR_COLON,              CHAR_SEMICOLON,
118     '`',      7, -ESC_b,      0, -ESC_d,  ESC_e,  ESC_f,      0,   /* ` - g */       CHAR_LESS_THAN_SIGN,     CHAR_EQUALS_SIGN,
119       0,      0,      0,      0,      0,      0,  ESC_n,      0,   /* h - o */       CHAR_GREATER_THAN_SIGN,  CHAR_QUESTION_MARK,
120  -ESC_p,      0,  ESC_r, -ESC_s,  ESC_tee,    0,      0, -ESC_w,   /* p - w */       CHAR_COMMERCIAL_AT,      -ESC_A,
121       0,      0, -ESC_z                                            /* x - z */       -ESC_B,                  -ESC_C,
122         -ESC_D,                  -ESC_E,
123         0,                       -ESC_G,
124         -ESC_H,                  0,
125         0,                       -ESC_K,
126         0,                       0,
127         -ESC_N,                  0,
128         -ESC_P,                  -ESC_Q,
129         -ESC_R,                  -ESC_S,
130         0,                       0,
131         -ESC_V,                  -ESC_W,
132         -ESC_X,                  0,
133         -ESC_Z,                  CHAR_LEFT_SQUARE_BRACKET,
134         CHAR_BACKSLASH,          CHAR_RIGHT_SQUARE_BRACKET,
135         CHAR_CIRCUMFLEX_ACCENT,  CHAR_UNDERSCORE,
136         CHAR_GRAVE_ACCENT,       7,
137         -ESC_b,                  0,
138         -ESC_d,                  ESC_e,
139         ESC_f,                   0,
140         -ESC_h,                  0,
141         0,                       -ESC_k,
142         0,                       0,
143         ESC_n,                   0,
144         -ESC_p,                  0,
145         ESC_r,                   -ESC_s,
146         ESC_tee,                 0,
147         -ESC_v,                  -ESC_w,
148         0,                       0,
149         -ESC_z
150  };  };
151    
152  #else         /* This is the "abnormal" table for EBCDIC systems */  #else
153    
154    /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
155    
156  static const short int escapes[] = {  static const short int escapes[] = {
157  /*  48 */     0,     0,      0,     '.',    '<',   '(',    '+',    '|',  /*  48 */     0,     0,      0,     '.',    '<',   '(',    '+',    '|',
158  /*  50 */   '&',     0,      0,       0,      0,     0,      0,      0,  /*  50 */   '&',     0,      0,       0,      0,     0,      0,      0,
# Line 96  static const short int escapes[] = { Line 162  static const short int escapes[] = {
162  /*  70 */     0,     0,      0,       0,      0,     0,      0,      0,  /*  70 */     0,     0,      0,       0,      0,     0,      0,      0,
163  /*  78 */     0,   '`',    ':',     '#',    '@',  '\'',    '=',    '"',  /*  78 */     0,   '`',    ':',     '#',    '@',  '\'',    '=',    '"',
164  /*  80 */     0,     7, -ESC_b,       0, -ESC_d, ESC_e,  ESC_f,      0,  /*  80 */     0,     7, -ESC_b,       0, -ESC_d, ESC_e,  ESC_f,      0,
165  /*  88 */     0,     0,      0,     '{',      0,     0,      0,      0,  /*  88 */-ESC_h,     0,      0,     '{',      0,     0,      0,      0,
166  /*  90 */     0,     0,      0,     'l',      0, ESC_n,      0, -ESC_p,  /*  90 */     0,     0, -ESC_k,     'l',      0, ESC_n,      0, -ESC_p,
167  /*  98 */     0, ESC_r,      0,     '}',      0,     0,      0,      0,  /*  98 */     0, ESC_r,      0,     '}',      0,     0,      0,      0,
168  /*  A0 */     0,   '~', -ESC_s, ESC_tee,      0,     0, -ESC_w,      0,  /*  A0 */     0,   '~', -ESC_s, ESC_tee,      0,-ESC_v, -ESC_w,      0,
169  /*  A8 */     0,-ESC_z,      0,       0,      0,   '[',      0,      0,  /*  A8 */     0,-ESC_z,      0,       0,      0,   '[',      0,      0,
170  /*  B0 */     0,     0,      0,       0,      0,     0,      0,      0,  /*  B0 */     0,     0,      0,       0,      0,     0,      0,      0,
171  /*  B8 */     0,     0,      0,       0,      0,   ']',    '=',    '-',  /*  B8 */     0,     0,      0,       0,      0,   ']',    '=',    '-',
172  /*  C0 */   '{',-ESC_A, -ESC_B,  -ESC_C, -ESC_D,-ESC_E,      0, -ESC_G,  /*  C0 */   '{',-ESC_A, -ESC_B,  -ESC_C, -ESC_D,-ESC_E,      0, -ESC_G,
173  /*  C8 */     0,     0,      0,       0,      0,     0,      0,      0,  /*  C8 */-ESC_H,     0,      0,       0,      0,     0,      0,      0,
174  /*  D0 */   '}',     0,      0,       0,      0,     0,      0, -ESC_P,  /*  D0 */   '}',     0, -ESC_K,       0,      0,-ESC_N,      0, -ESC_P,
175  /*  D8 */-ESC_Q,     0,      0,       0,      0,     0,      0,      0,  /*  D8 */-ESC_Q,-ESC_R,      0,       0,      0,     0,      0,      0,
176  /*  E0 */  '\\',     0, -ESC_S,       0,      0,     0, -ESC_W, -ESC_X,  /*  E0 */  '\\',     0, -ESC_S,       0,      0,-ESC_V, -ESC_W, -ESC_X,
177  /*  E8 */     0,-ESC_Z,      0,       0,      0,     0,      0,      0,  /*  E8 */     0,-ESC_Z,      0,       0,      0,     0,      0,      0,
178  /*  F0 */     0,     0,      0,       0,      0,     0,      0,      0,  /*  F0 */     0,     0,      0,       0,      0,     0,      0,      0,
179  /*  F8 */     0,     0,      0,       0,      0,     0,      0,      0  /*  F8 */     0,     0,      0,       0,      0,     0,      0,      0
# Line 115  static const short int escapes[] = { Line 181  static const short int escapes[] = {
181  #endif  #endif
182    
183    
184  /* Tables of names of POSIX character classes and their lengths. The list is  /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
185  terminated by a zero length entry. The first three must be alpha, upper, lower,  searched linearly. Put all the names into a single string, in order to reduce
186  as this is assumed for handling case independence. */  the number of relocations when a shared library is dynamically linked. The
187    string is built from string macros so that it works in UTF-8 mode on EBCDIC
188  static const char *const posix_names[] = {  platforms. */
189    "alpha", "lower", "upper",  
190    "alnum", "ascii", "blank", "cntrl", "digit", "graph",  typedef struct verbitem {
191    "print", "punct", "space", "word",  "xdigit" };    int   len;                 /* Length of verb name */
192      int   op;                  /* Op when no arg, or -1 if arg mandatory */
193      int   op_arg;              /* Op when arg present, or -1 if not allowed */
194    } verbitem;
195    
196    static const char verbnames[] =
197      "\0"                       /* Empty name is a shorthand for MARK */
198      STRING_MARK0
199      STRING_ACCEPT0
200      STRING_COMMIT0
201      STRING_F0
202      STRING_FAIL0
203      STRING_PRUNE0
204      STRING_SKIP0
205      STRING_THEN;
206    
207    static const verbitem verbs[] = {
208      { 0, -1,        OP_MARK },
209      { 4, -1,        OP_MARK },
210      { 6, OP_ACCEPT, -1 },
211      { 6, OP_COMMIT, -1 },
212      { 1, OP_FAIL,   -1 },
213      { 4, OP_FAIL,   -1 },
214      { 5, OP_PRUNE,  OP_PRUNE_ARG },
215      { 4, OP_SKIP,   OP_SKIP_ARG  },
216      { 4, OP_THEN,   OP_THEN_ARG  }
217    };
218    
219    static const int verbcount = sizeof(verbs)/sizeof(verbitem);
220    
221    
222    /* Tables of names of POSIX character classes and their lengths. The names are
223    now all in a single string, to reduce the number of relocations when a shared
224    library is dynamically loaded. The list of lengths is terminated by a zero
225    length entry. The first three must be alpha, lower, upper, as this is assumed
226    for handling case independence. */
227    
228    static const char posix_names[] =
229      STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
230      STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
231      STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
232      STRING_word0  STRING_xdigit;
233    
234  static const uschar posix_name_lengths[] = {  static const uschar posix_name_lengths[] = {
235    5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };    5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
236    
237  /* Table of class bit maps for each POSIX class; up to three may be combined  /* Table of class bit maps for each POSIX class. Each class is formed from a
238  to form the class. The table for [:blank:] is dynamically modified to remove  base map, with an optional addition or removal of another map. Then, for some
239  the vertical space characters. */  classes, there is some additional tweaking: for [:blank:] the vertical space
240    characters are removed, and for [:alpha:] and [:alnum:] the underscore
241    character is removed. The triples in the table consist of the base map offset,
242    second map offset or -1 if no second map, and a non-negative value for map
243    addition or a negative value for map subtraction (if there are two maps). The
244    absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
245    remove vertical space characters, 2 => remove underscore. */
246    
247  static const int posix_class_maps[] = {  static const int posix_class_maps[] = {
248    cbit_lower, cbit_upper, -1,             /* alpha */    cbit_word,  cbit_digit, -2,             /* alpha */
249    cbit_lower, -1,         -1,             /* lower */    cbit_lower, -1,          0,             /* lower */
250    cbit_upper, -1,         -1,             /* upper */    cbit_upper, -1,          0,             /* upper */
251    cbit_digit, cbit_lower, cbit_upper,     /* alnum */    cbit_word,  -1,          2,             /* alnum - word without underscore */
252    cbit_print, cbit_cntrl, -1,             /* ascii */    cbit_print, cbit_cntrl,  0,             /* ascii */
253    cbit_space, -1,         -1,             /* blank - a GNU extension */    cbit_space, -1,          1,             /* blank - a GNU extension */
254    cbit_cntrl, -1,         -1,             /* cntrl */    cbit_cntrl, -1,          0,             /* cntrl */
255    cbit_digit, -1,         -1,             /* digit */    cbit_digit, -1,          0,             /* digit */
256    cbit_graph, -1,         -1,             /* graph */    cbit_graph, -1,          0,             /* graph */
257    cbit_print, -1,         -1,             /* print */    cbit_print, -1,          0,             /* print */
258    cbit_punct, -1,         -1,             /* punct */    cbit_punct, -1,          0,             /* punct */
259    cbit_space, -1,         -1,             /* space */    cbit_space, -1,          0,             /* space */
260    cbit_word,  -1,         -1,             /* word - a Perl extension */    cbit_word,  -1,          0,             /* word - a Perl extension */
261    cbit_xdigit,-1,         -1              /* xdigit */    cbit_xdigit,-1,          0              /* xdigit */
262  };  };
263    
264    /* Table of substitutes for \d etc when PCRE_UCP is set. The POSIX class
265    substitutes must be in the order of the names, defined above, and there are
266    both positive and negative cases. NULL means no substitute. */
267    
268  /* The texts of compile-time error messages. These are "char *" because they  #ifdef SUPPORT_UCP
269  are passed to the outside world. */  static const uschar *substitutes[] = {
270      (uschar *)"\\P{Nd}",    /* \D */
271      (uschar *)"\\p{Nd}",    /* \d */
272      (uschar *)"\\P{Xsp}",   /* \S */       /* NOTE: Xsp is Perl space */
273      (uschar *)"\\p{Xsp}",   /* \s */
274      (uschar *)"\\P{Xwd}",   /* \W */
275      (uschar *)"\\p{Xwd}"    /* \w */
276    };
277    
278    static const uschar *posix_substitutes[] = {
279      (uschar *)"\\p{L}",     /* alpha */
280      (uschar *)"\\p{Ll}",    /* lower */
281      (uschar *)"\\p{Lu}",    /* upper */
282      (uschar *)"\\p{Xan}",   /* alnum */
283      NULL,                   /* ascii */
284      (uschar *)"\\h",        /* blank */
285      NULL,                   /* cntrl */
286      (uschar *)"\\p{Nd}",    /* digit */
287      NULL,                   /* graph */
288      NULL,                   /* print */
289      NULL,                   /* punct */
290      (uschar *)"\\p{Xps}",   /* space */    /* NOTE: Xps is POSIX space */
291      (uschar *)"\\p{Xwd}",   /* word */
292      NULL,                   /* xdigit */
293      /* Negated cases */
294      (uschar *)"\\P{L}",     /* ^alpha */
295      (uschar *)"\\P{Ll}",    /* ^lower */
296      (uschar *)"\\P{Lu}",    /* ^upper */
297      (uschar *)"\\P{Xan}",   /* ^alnum */
298      NULL,                   /* ^ascii */
299      (uschar *)"\\H",        /* ^blank */
300      NULL,                   /* ^cntrl */
301      (uschar *)"\\P{Nd}",    /* ^digit */
302      NULL,                   /* ^graph */
303      NULL,                   /* ^print */
304      NULL,                   /* ^punct */
305      (uschar *)"\\P{Xps}",   /* ^space */   /* NOTE: Xps is POSIX space */
306      (uschar *)"\\P{Xwd}",   /* ^word */
307      NULL                    /* ^xdigit */
308    };
309    #define POSIX_SUBSIZE (sizeof(posix_substitutes)/sizeof(uschar *))
310    #endif
311    
312  static const char *error_texts[] = {  #define STRING(a)  # a
313    "no error",  #define XSTRING(s) STRING(s)
314    "\\ at end of pattern",  
315    "\\c at end of pattern",  /* The texts of compile-time error messages. These are "char *" because they
316    "unrecognized character follows \\",  are passed to the outside world. Do not ever re-use any error number, because
317    "numbers out of order in {} quantifier",  they are documented. Always add a new error instead. Messages marked DEAD below
318    are no longer used. This used to be a table of strings, but in order to reduce
319    the number of relocations needed when a shared library is loaded dynamically,
320    it is now one long string. We cannot use a table of offsets, because the
321    lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
322    simply count through to the one we want - this isn't a performance issue
323    because these strings are used only when there is a compilation error.
324    
325    Each substring ends with \0 to insert a null character. This includes the final
326    substring, so that the whole string ends with \0\0, which can be detected when
327    counting through. */
328    
329    static const char error_texts[] =
330      "no error\0"
331      "\\ at end of pattern\0"
332      "\\c at end of pattern\0"
333      "unrecognized character follows \\\0"
334      "numbers out of order in {} quantifier\0"
335    /* 5 */    /* 5 */
336    "number too big in {} quantifier",    "number too big in {} quantifier\0"
337    "missing terminating ] for character class",    "missing terminating ] for character class\0"
338    "invalid escape sequence in character class",    "invalid escape sequence in character class\0"
339    "range out of order in character class",    "range out of order in character class\0"
340    "nothing to repeat",    "nothing to repeat\0"
341    /* 10 */    /* 10 */
342    "operand of unlimited repeat could match the empty string",    "operand of unlimited repeat could match the empty string\0"  /** DEAD **/
343    "internal error: unexpected repeat",    "internal error: unexpected repeat\0"
344    "unrecognized character after (?",    "unrecognized character after (? or (?-\0"
345    "POSIX named classes are supported only within a class",    "POSIX named classes are supported only within a class\0"
346    "missing )",    "missing )\0"
347    /* 15 */    /* 15 */
348    "reference to non-existent subpattern",    "reference to non-existent subpattern\0"
349    "erroffset passed as NULL",    "erroffset passed as NULL\0"
350    "unknown option bit(s) set",    "unknown option bit(s) set\0"
351    "missing ) after comment",    "missing ) after comment\0"
352    "parentheses nested too deeply",    "parentheses nested too deeply\0"  /** DEAD **/
353    /* 20 */    /* 20 */
354    "regular expression too large",    "regular expression is too large\0"
355    "failed to get memory",    "failed to get memory\0"
356    "unmatched parentheses",    "unmatched parentheses\0"
357    "internal error: code overflow",    "internal error: code overflow\0"
358    "unrecognized character after (?<",    "unrecognized character after (?<\0"
359    /* 25 */    /* 25 */
360    "lookbehind assertion is not fixed length",    "lookbehind assertion is not fixed length\0"
361    "malformed number after (?(",    "malformed number or name after (?(\0"
362    "conditional group contains more than two branches",    "conditional group contains more than two branches\0"
363    "assertion expected after (?(",    "assertion expected after (?(\0"
364    "(?R or (?digits must be followed by )",    "(?R or (?[+-]digits must be followed by )\0"
365    /* 30 */    /* 30 */
366    "unknown POSIX class name",    "unknown POSIX class name\0"
367    "POSIX collating elements are not supported",    "POSIX collating elements are not supported\0"
368    "this version of PCRE is not compiled with PCRE_UTF8 support",    "this version of PCRE is not compiled with PCRE_UTF8 support\0"
369    "spare error",    "spare error\0"  /** DEAD **/
370    "character value in \\x{...} sequence is too large",    "character value in \\x{...} sequence is too large\0"
371    /* 35 */    /* 35 */
372    "invalid condition (?(0)",    "invalid condition (?(0)\0"
373    "\\C not allowed in lookbehind assertion",    "\\C not allowed in lookbehind assertion\0"
374    "PCRE does not support \\L, \\l, \\N, \\U, or \\u",    "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
375    "number after (?C is > 255",    "number after (?C is > 255\0"
376    "closing ) for (?C expected",    "closing ) for (?C expected\0"
377    /* 40 */    /* 40 */
378    "recursive call could loop indefinitely",    "recursive call could loop indefinitely\0"
379    "unrecognized character after (?P",    "unrecognized character after (?P\0"
380    "syntax error after (?P",    "syntax error in subpattern name (missing terminator)\0"
381    "two named groups have the same name",    "two named subpatterns have the same name\0"
382    "invalid UTF-8 string",    "invalid UTF-8 string\0"
383    /* 45 */    /* 45 */
384    "support for \\P, \\p, and \\X has not been compiled",    "support for \\P, \\p, and \\X has not been compiled\0"
385    "malformed \\P or \\p sequence",    "malformed \\P or \\p sequence\0"
386    "unknown property name after \\P or \\p"    "unknown property name after \\P or \\p\0"
387  };    "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
388      "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
389      /* 50 */
390      "repeated subpattern is too long\0"    /** DEAD **/
391      "octal value is greater than \\377 (not in UTF-8 mode)\0"
392      "internal error: overran compiling workspace\0"
393      "internal error: previously-checked referenced subpattern not found\0"
394      "DEFINE group contains more than one branch\0"
395      /* 55 */
396      "repeating a DEFINE group is not allowed\0"  /** DEAD **/
397      "inconsistent NEWLINE options\0"
398      "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
399      "a numbered reference must not be zero\0"
400      "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
401      /* 60 */
402      "(*VERB) not recognized\0"
403      "number is too big\0"
404      "subpattern name expected\0"
405      "digit expected after (?+\0"
406      "] is an invalid data character in JavaScript compatibility mode\0"
407      /* 65 */
408      "different names for subpatterns of the same number are not allowed\0"
409      "(*MARK) must have an argument\0"
410      "this version of PCRE is not compiled with PCRE_UCP support\0"
411      "\\c must be followed by an ASCII character\0"
412      "\\k is not followed by a braced, angle-bracketed, or quoted name\0"
413      ;
414    
415  /* Table to identify digits and hex digits. This is used when compiling  /* Table to identify digits and hex digits. This is used when compiling
416  patterns. Note that the tables in chartables are dependent on the locale, and  patterns. Note that the tables in chartables are dependent on the locale, and
# Line 229  For convenience, we use the same bit def Line 428  For convenience, we use the same bit def
428    
429  Then we can use ctype_digit and ctype_xdigit in the code. */  Then we can use ctype_digit and ctype_xdigit in the code. */
430    
431  #if !EBCDIC    /* This is the "normal" case, for ASCII systems */  #ifndef EBCDIC
432    
433    /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
434    UTF-8 mode. */
435    
436  static const unsigned char digitab[] =  static const unsigned char digitab[] =
437    {    {
438    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*   0-  7 */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*   0-  7 */
# Line 265  static const unsigned char digitab[] = Line 468  static const unsigned char digitab[] =
468    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
469    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
470    
471  #else          /* This is the "abnormal" case, for EBCDIC systems */  #else
472    
473    /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
474    
475  static const unsigned char digitab[] =  static const unsigned char digitab[] =
476    {    {
477    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*   0-  7  0 */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*   0-  7  0 */
# Line 279  static const unsigned char digitab[] = Line 485  static const unsigned char digitab[] =
485    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*    - 71 40 */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*    - 71 40 */
486    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  72- |     */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  72- |     */
487    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  & - 87 50 */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  & - 87 50 */
488    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  88-     */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  88- 95    */
489    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  - -103 60 */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  - -103 60 */
490    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ?     */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ?     */
491    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
# Line 313  static const unsigned char ebcdic_charta Line 519  static const unsigned char ebcdic_charta
519    0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*    - 71 */    0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*    - 71 */
520    0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /*  72- |  */    0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /*  72- |  */
521    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  & - 87 */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  & - 87 */
522    0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /*  88-  */    0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /*  88- 95 */
523    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  - -103 */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  - -103 */
524    0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ?  */    0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ?  */
525    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
# Line 340  static const unsigned char ebcdic_charta Line 546  static const unsigned char ebcdic_charta
546  /* Definition to allow mutual recursion */  /* Definition to allow mutual recursion */
547    
548  static BOOL  static BOOL
549    compile_regex(int, int, int *, uschar **, const uschar **, int *, BOOL, int,    compile_regex(int, uschar **, const uschar **, int *, BOOL, BOOL, int, int *,
550      int *, int *, branch_chain *, compile_data *);      int *, branch_chain *, compile_data *, int *);
551    
552    
553    
554    /*************************************************
555    *            Find an error text                  *
556    *************************************************/
557    
558    /* The error texts are now all in one long string, to save on relocations. As
559    some of the text is of unknown length, we can't use a table of offsets.
560    Instead, just count through the strings. This is not a performance issue
561    because it happens only when there has been a compilation error.
562    
563    Argument:   the error number
564    Returns:    pointer to the error string
565    */
566    
567    static const char *
568    find_error_text(int n)
569    {
570    const char *s = error_texts;
571    for (; n > 0; n--)
572      {
573      while (*s++ != 0) {};
574      if (*s == 0) return "Error text not found (please report)";
575      }
576    return s;
577    }
578    
579    
580  /*************************************************  /*************************************************
# Line 351  static BOOL Line 583  static BOOL
583    
584  /* This function is called when a \ has been encountered. It either returns a  /* This function is called when a \ has been encountered. It either returns a
585  positive value for a simple escape such as \n, or a negative value which  positive value for a simple escape such as \n, or a negative value which
586  encodes one of the more complicated things such as \d. When UTF-8 is enabled,  encodes one of the more complicated things such as \d. A backreference to group
587  a positive value greater than 255 may be returned. On entry, ptr is pointing at  n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
588  the \. On exit, it is on the final character of the escape sequence.  UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
589    ptr is pointing at the \. On exit, it is on the final character of the escape
590    sequence.
591    
592  Arguments:  Arguments:
593    ptrptr         points to the pattern position pointer    ptrptr         points to the pattern position pointer
# Line 364  Arguments: Line 598  Arguments:
598    
599  Returns:         zero or positive => a data character  Returns:         zero or positive => a data character
600                   negative => a special escape sequence                   negative => a special escape sequence
601                   on error, errorptr is set                   on error, errorcodeptr is set
602  */  */
603    
604  static int  static int
605  check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,  check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
606    int options, BOOL isclass)    int options, BOOL isclass)
607  {  {
608  const uschar *ptr = *ptrptr;  BOOL utf8 = (options & PCRE_UTF8) != 0;
609    const uschar *ptr = *ptrptr + 1;
610  int c, i;  int c, i;
611    
612    GETCHARINCTEST(c, ptr);           /* Get character value, increment pointer */
613    ptr--;                            /* Set pointer back to the last byte */
614    
615  /* If backslash is at the end of the pattern, it's an error. */  /* If backslash is at the end of the pattern, it's an error. */
616    
 c = *(++ptr);  
617  if (c == 0) *errorcodeptr = ERR1;  if (c == 0) *errorcodeptr = ERR1;
618    
619  /* Non-alphamerics are literals. For digits or letters, do an initial lookup in  /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
620  a table. A non-zero result is something that can be returned immediately.  in a table. A non-zero result is something that can be returned immediately.
621  Otherwise further processing may be required. */  Otherwise further processing may be required. */
622    
623  #if !EBCDIC    /* ASCII coding */  #ifndef EBCDIC  /* ASCII/UTF-8 coding */
624  else if (c < '0' || c > 'z') {}                           /* Not alphameric */  else if (c < CHAR_0 || c > CHAR_z) {}                     /* Not alphanumeric */
625  else if ((i = escapes[c - '0']) != 0) c = i;  else if ((i = escapes[c - CHAR_0]) != 0) c = i;
626    
627  #else          /* EBCDIC coding */  #else           /* EBCDIC coding */
628  else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {}   /* Not alphameric */  else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {}   /* Not alphanumeric */
629  else if ((i = escapes[c - 0x48]) != 0)  c = i;  else if ((i = escapes[c - 0x48]) != 0)  c = i;
630  #endif  #endif
631    
# Line 397  else if ((i = escapes[c - 0x48]) != 0) Line 634  else if ((i = escapes[c - 0x48]) != 0)
634  else  else
635    {    {
636    const uschar *oldptr;    const uschar *oldptr;
637      BOOL braced, negated;
638    
639    switch (c)    switch (c)
640      {      {
641      /* A number of Perl escapes are not handled by PCRE. We give an explicit      /* A number of Perl escapes are not handled by PCRE. We give an explicit
642      error. */      error. */
643    
644      case 'l':      case CHAR_l:
645      case 'L':      case CHAR_L:
646      case 'N':      case CHAR_u:
647      case 'u':      case CHAR_U:
     case 'U':  
648      *errorcodeptr = ERR37;      *errorcodeptr = ERR37;
649      break;      break;
650    
651        /* \g must be followed by one of a number of specific things:
652    
653        (1) A number, either plain or braced. If positive, it is an absolute
654        backreference. If negative, it is a relative backreference. This is a Perl
655        5.10 feature.
656    
657        (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
658        is part of Perl's movement towards a unified syntax for back references. As
659        this is synonymous with \k{name}, we fudge it up by pretending it really
660        was \k.
661    
662        (3) For Oniguruma compatibility we also support \g followed by a name or a
663        number either in angle brackets or in single quotes. However, these are
664        (possibly recursive) subroutine calls, _not_ backreferences. Just return
665        the -ESC_g code (cf \k). */
666    
667        case CHAR_g:
668        if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
669          {
670          c = -ESC_g;
671          break;
672          }
673    
674        /* Handle the Perl-compatible cases */
675    
676        if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
677          {
678          const uschar *p;
679          for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
680            if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break;
681          if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
682            {
683            c = -ESC_k;
684            break;
685            }
686          braced = TRUE;
687          ptr++;
688          }
689        else braced = FALSE;
690    
691        if (ptr[1] == CHAR_MINUS)
692          {
693          negated = TRUE;
694          ptr++;
695          }
696        else negated = FALSE;
697    
698        c = 0;
699        while ((digitab[ptr[1]] & ctype_digit) != 0)
700          c = c * 10 + *(++ptr) - CHAR_0;
701    
702        if (c < 0)   /* Integer overflow */
703          {
704          *errorcodeptr = ERR61;
705          break;
706          }
707    
708        if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
709          {
710          *errorcodeptr = ERR57;
711          break;
712          }
713    
714        if (c == 0)
715          {
716          *errorcodeptr = ERR58;
717          break;
718          }
719    
720        if (negated)
721          {
722          if (c > bracount)
723            {
724            *errorcodeptr = ERR15;
725            break;
726            }
727          c = bracount - (c - 1);
728          }
729    
730        c = -(ESC_REF + c);
731        break;
732    
733      /* The handling of escape sequences consisting of a string of digits      /* The handling of escape sequences consisting of a string of digits
734      starting with one that is not zero is not straightforward. By experiment,      starting with one that is not zero is not straightforward. By experiment,
735      the way Perl works seems to be as follows:      the way Perl works seems to be as follows:
# Line 422  else Line 742  else
742      value is greater than 377, the least significant 8 bits are taken. Inside a      value is greater than 377, the least significant 8 bits are taken. Inside a
743      character class, \ followed by a digit is always an octal number. */      character class, \ followed by a digit is always an octal number. */
744    
745      case '1': case '2': case '3': case '4': case '5':      case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
746      case '6': case '7': case '8': case '9':      case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
747    
748      if (!isclass)      if (!isclass)
749        {        {
750        oldptr = ptr;        oldptr = ptr;
751        c -= '0';        c -= CHAR_0;
752        while ((digitab[ptr[1]] & ctype_digit) != 0)        while ((digitab[ptr[1]] & ctype_digit) != 0)
753          c = c * 10 + *(++ptr) - '0';          c = c * 10 + *(++ptr) - CHAR_0;
754          if (c < 0)    /* Integer overflow */
755            {
756            *errorcodeptr = ERR61;
757            break;
758            }
759        if (c < 10 || c <= bracount)        if (c < 10 || c <= bracount)
760          {          {
761          c = -(ESC_REF + c);          c = -(ESC_REF + c);
# Line 443  else Line 768  else
768      generates a binary zero byte and treats the digit as a following literal.      generates a binary zero byte and treats the digit as a following literal.
769      Thus we have to pull back the pointer by one. */      Thus we have to pull back the pointer by one. */
770    
771      if ((c = *ptr) >= '8')      if ((c = *ptr) >= CHAR_8)
772        {        {
773        ptr--;        ptr--;
774        c = 0;        c = 0;
# Line 451  else Line 776  else
776        }        }
777    
778      /* \0 always starts an octal number, but we may drop through to here with a      /* \0 always starts an octal number, but we may drop through to here with a
779      larger first octal digit. */      larger first octal digit. The original code used just to take the least
780        significant 8 bits of octal numbers (I think this is what early Perls used
781      case '0':      to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
782      c -= '0';      than 3 octal digits. */
783      while(i++ < 2 && ptr[1] >= '0' && ptr[1] <= '7')  
784          c = c * 8 + *(++ptr) - '0';      case CHAR_0:
785      c &= 255;     /* Take least significant 8 bits */      c -= CHAR_0;
786        while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
787            c = c * 8 + *(++ptr) - CHAR_0;
788        if (!utf8 && c > 255) *errorcodeptr = ERR51;
789      break;      break;
790    
791      /* \x is complicated when UTF-8 is enabled. \x{ddd} is a character number      /* \x is complicated. \x{ddd} is a character number which can be greater
792      which can be greater than 0xff, but only if the ddd are hex digits. */      than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
793        treated as a data character. */
794    
795      case 'x':      case CHAR_x:
796  #ifdef SUPPORT_UTF8      if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
     if (ptr[1] == '{' && (options & PCRE_UTF8) != 0)  
797        {        {
798        const uschar *pt = ptr + 2;        const uschar *pt = ptr + 2;
799        register int count = 0;        int count = 0;
800    
801        c = 0;        c = 0;
802        while ((digitab[*pt] & ctype_xdigit) != 0)        while ((digitab[*pt] & ctype_xdigit) != 0)
803          {          {
804          int cc = *pt++;          register int cc = *pt++;
805            if (c == 0 && cc == CHAR_0) continue;     /* Leading zeroes */
806          count++;          count++;
807  #if !EBCDIC    /* ASCII coding */  
808          if (cc >= 'a') cc -= 32;               /* Convert to upper case */  #ifndef EBCDIC  /* ASCII/UTF-8 coding */
809          c = c * 16 + cc - ((cc < 'A')? '0' : ('A' - 10));          if (cc >= CHAR_a) cc -= 32;               /* Convert to upper case */
810  #else          /* EBCDIC coding */          c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
811          if (cc >= 'a' && cc <= 'z') cc += 64;  /* Convert to upper case */  #else           /* EBCDIC coding */
812          c = c * 16 + cc - ((cc >= '0')? '0' : ('A' - 10));          if (cc >= CHAR_a && cc <= CHAR_z) cc += 64;  /* Convert to upper case */
813            c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
814  #endif  #endif
815          }          }
816        if (*pt == '}')  
817          if (*pt == CHAR_RIGHT_CURLY_BRACKET)
818          {          {
819          if (c < 0 || count > 8) *errorcodeptr = ERR34;          if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
820          ptr = pt;          ptr = pt;
821          break;          break;
822          }          }
823    
824        /* If the sequence of hex digits does not end with '}', then we don't        /* If the sequence of hex digits does not end with '}', then we don't
825        recognize this construct; fall through to the normal \x handling. */        recognize this construct; fall through to the normal \x handling. */
826        }        }
 #endif  
827    
828      /* Read just a single hex char */      /* Read just a single-byte hex-defined char */
829    
830      c = 0;      c = 0;
831      while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)      while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
832        {        {
833        int cc;                               /* Some compilers don't like ++ */        int cc;                                  /* Some compilers don't like */
834        cc = *(++ptr);                        /* in initializers */        cc = *(++ptr);                           /* ++ in initializers */
835  #if !EBCDIC    /* ASCII coding */  #ifndef EBCDIC  /* ASCII/UTF-8 coding */
836        if (cc >= 'a') cc -= 32;              /* Convert to upper case */        if (cc >= CHAR_a) cc -= 32;              /* Convert to upper case */
837        c = c * 16 + cc - ((cc < 'A')? '0' : ('A' - 10));        c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
838  #else          /* EBCDIC coding */  #else           /* EBCDIC coding */
839        if (cc <= 'z') cc += 64;              /* Convert to upper case */        if (cc <= CHAR_z) cc += 64;              /* Convert to upper case */
840        c = c * 16 + cc - ((cc >= '0')? '0' : ('A' - 10));        c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
841  #endif  #endif
842        }        }
843      break;      break;
844    
845      /* Other special escapes not starting with a digit are straightforward */      /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
846        An error is given if the byte following \c is not an ASCII character. This
847        coding is ASCII-specific, but then the whole concept of \cx is
848        ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
849    
850      case 'c':      case CHAR_c:
851      c = *(++ptr);      c = *(++ptr);
852      if (c == 0)      if (c == 0)
853        {        {
854        *errorcodeptr = ERR2;        *errorcodeptr = ERR2;
855        return 0;        break;
856        }        }
857    #ifndef EBCDIC    /* ASCII/UTF-8 coding */
858      /* A letter is upper-cased; then the 0x40 bit is flipped. This coding      if (c > 127)  /* Excludes all non-ASCII in either mode */
859      is ASCII-specific, but then the whole concept of \cx is ASCII-specific.        {
860      (However, an EBCDIC equivalent has now been added.) */        *errorcodeptr = ERR68;
861          break;
862  #if !EBCDIC    /* ASCII coding */        }
863      if (c >= 'a' && c <= 'z') c -= 32;      if (c >= CHAR_a && c <= CHAR_z) c -= 32;
864      c ^= 0x40;      c ^= 0x40;
865  #else          /* EBCDIC coding */  #else             /* EBCDIC coding */
866      if (c >= 'a' && c <= 'z') c += 64;      if (c >= CHAR_a && c <= CHAR_z) c += 64;
867      c ^= 0xC0;      c ^= 0xC0;
868  #endif  #endif
869      break;      break;
870    
871      /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any      /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
872      other alphameric following \ is an error if PCRE_EXTRA was set; otherwise,      other alphanumeric following \ is an error if PCRE_EXTRA was set;
873      for Perl compatibility, it is a literal. This code looks a bit odd, but      otherwise, for Perl compatibility, it is a literal. This code looks a bit
874      there used to be some cases other than the default, and there may be again      odd, but there used to be some cases other than the default, and there may
875      in future, so I haven't "optimized" it. */      be again in future, so I haven't "optimized" it. */
876    
877      default:      default:
878      if ((options & PCRE_EXTRA) != 0) switch(c)      if ((options & PCRE_EXTRA) != 0) switch(c)
# Line 550  else Line 885  else
885      }      }
886    }    }
887    
888    /* Perl supports \N{name} for character names, as well as plain \N for "not
889    newline". PCRE does not support \N{name}. */
890    
891    if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET)
892      *errorcodeptr = ERR37;
893    
894    /* If PCRE_UCP is set, we change the values for \d etc. */
895    
896    if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
897      c -= (ESC_DU - ESC_D);
898    
899    /* Set the pointer to the final character before returning. */
900    
901  *ptrptr = ptr;  *ptrptr = ptr;
902  return c;  return c;
903  }  }
# Line 569  escape sequence. Line 917  escape sequence.
917  Argument:  Argument:
918    ptrptr         points to the pattern position pointer    ptrptr         points to the pattern position pointer
919    negptr         points to a boolean that is set TRUE for negation else FALSE    negptr         points to a boolean that is set TRUE for negation else FALSE
920      dptr           points to an int that is set to the detailed property value
921    errorcodeptr   points to the error code variable    errorcodeptr   points to the error code variable
922    
923  Returns:     value from ucp_type_table, or -1 for an invalid type  Returns:         type value from ucp_type_table, or -1 for an invalid type
924  */  */
925    
926  static int  static int
927  get_ucp(const uschar **ptrptr, BOOL *negptr, int *errorcodeptr)  get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
928  {  {
929  int c, i, bot, top;  int c, i, bot, top;
930  const uschar *ptr = *ptrptr;  const uschar *ptr = *ptrptr;
931  char name[4];  char name[32];
932    
933  c = *(++ptr);  c = *(++ptr);
934  if (c == 0) goto ERROR_RETURN;  if (c == 0) goto ERROR_RETURN;
935    
936  *negptr = FALSE;  *negptr = FALSE;
937    
938  /* \P or \p can be followed by a one- or two-character name in {}, optionally  /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
939  preceded by ^ for negation. */  negation. */
940    
941  if (c == '{')  if (c == CHAR_LEFT_CURLY_BRACKET)
942    {    {
943    if (ptr[1] == '^')    if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
944      {      {
945      *negptr = TRUE;      *negptr = TRUE;
946      ptr++;      ptr++;
947      }      }
948    for (i = 0; i <= 2; i++)    for (i = 0; i < (int)sizeof(name) - 1; i++)
949      {      {
950      c = *(++ptr);      c = *(++ptr);
951      if (c == 0) goto ERROR_RETURN;      if (c == 0) goto ERROR_RETURN;
952      if (c == '}') break;      if (c == CHAR_RIGHT_CURLY_BRACKET) break;
953      name[i] = c;      name[i] = c;
954      }      }
955    if (c !='}')   /* Try to distinguish error cases */    if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
     {  
     while (*(++ptr) != 0 && *ptr != '}');  
     if (*ptr == '}') goto UNKNOWN_RETURN; else goto ERROR_RETURN;  
     }  
956    name[i] = 0;    name[i] = 0;
957    }    }
958    
# Line 628  top = _pcre_utt_size; Line 973  top = _pcre_utt_size;
973    
974  while (bot < top)  while (bot < top)
975    {    {
976    i = (bot + top)/2;    i = (bot + top) >> 1;
977    c = strcmp(name, _pcre_utt[i].name);    c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
978    if (c == 0) return _pcre_utt[i].value;    if (c == 0)
979        {
980        *dptr = _pcre_utt[i].value;
981        return _pcre_utt[i].type;
982        }
983    if (c > 0) bot = i + 1; else top = i;    if (c > 0) bot = i + 1; else top = i;
984    }    }
985    
 UNKNOWN_RETURN:  
986  *errorcodeptr = ERR47;  *errorcodeptr = ERR47;
987  *ptrptr = ptr;  *ptrptr = ptr;
988  return -1;  return -1;
# Line 669  is_counted_repeat(const uschar *p) Line 1017  is_counted_repeat(const uschar *p)
1017  {  {
1018  if ((digitab[*p++] & ctype_digit) == 0) return FALSE;  if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
1019  while ((digitab[*p] & ctype_digit) != 0) p++;  while ((digitab[*p] & ctype_digit) != 0) p++;
1020  if (*p == '}') return TRUE;  if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
1021    
1022  if (*p++ != ',') return FALSE;  if (*p++ != CHAR_COMMA) return FALSE;
1023  if (*p == '}') return TRUE;  if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
1024    
1025  if ((digitab[*p++] & ctype_digit) == 0) return FALSE;  if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
1026  while ((digitab[*p] & ctype_digit) != 0) p++;  while ((digitab[*p] & ctype_digit) != 0) p++;
1027    
1028  return (*p == '}');  return (*p == CHAR_RIGHT_CURLY_BRACKET);
1029  }  }
1030    
1031    
# Line 710  int max = -1; Line 1058  int max = -1;
1058  /* Read the minimum value and do a paranoid check: a negative value indicates  /* Read the minimum value and do a paranoid check: a negative value indicates
1059  an integer overflow. */  an integer overflow. */
1060    
1061  while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - '0';  while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - CHAR_0;
1062  if (min < 0 || min > 65535)  if (min < 0 || min > 65535)
1063    {    {
1064    *errorcodeptr = ERR5;    *errorcodeptr = ERR5;
# Line 720  if (min < 0 || min > 65535) Line 1068  if (min < 0 || min > 65535)
1068  /* Read the maximum value if there is one, and again do a paranoid on its size.  /* Read the maximum value if there is one, and again do a paranoid on its size.
1069  Also, max must not be less than min. */  Also, max must not be less than min. */
1070    
1071  if (*p == '}') max = min; else  if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1072    {    {
1073    if (*(++p) != '}')    if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1074      {      {
1075      max = 0;      max = 0;
1076      while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - '0';      while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0;
1077      if (max < 0 || max > 65535)      if (max < 0 || max > 65535)
1078        {        {
1079        *errorcodeptr = ERR5;        *errorcodeptr = ERR5;
# Line 750  return p; Line 1098  return p;
1098    
1099    
1100  /*************************************************  /*************************************************
1101    *  Subroutine for finding forward reference      *
1102    *************************************************/
1103    
1104    /* This recursive function is called only from find_parens() below. The
1105    top-level call starts at the beginning of the pattern. All other calls must
1106    start at a parenthesis. It scans along a pattern's text looking for capturing
1107    subpatterns, and counting them. If it finds a named pattern that matches the
1108    name it is given, it returns its number. Alternatively, if the name is NULL, it
1109    returns when it reaches a given numbered subpattern. Recursion is used to keep
1110    track of subpatterns that reset the capturing group numbers - the (?| feature.
1111    
1112    This function was originally called only from the second pass, in which we know
1113    that if (?< or (?' or (?P< is encountered, the name will be correctly
1114    terminated because that is checked in the first pass. There is now one call to
1115    this function in the first pass, to check for a recursive back reference by
1116    name (so that we can make the whole group atomic). In this case, we need check
1117    only up to the current position in the pattern, and that is still OK because
1118    and previous occurrences will have been checked. To make this work, the test
1119    for "end of pattern" is a check against cd->end_pattern in the main loop,
1120    instead of looking for a binary zero. This means that the special first-pass
1121    call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1122    processing items within the loop are OK, because afterwards the main loop will
1123    terminate.)
1124    
1125    Arguments:
1126      ptrptr       address of the current character pointer (updated)
1127      cd           compile background data
1128      name         name to seek, or NULL if seeking a numbered subpattern
1129      lorn         name length, or subpattern number if name is NULL
1130      xmode        TRUE if we are in /x mode
1131      utf8         TRUE if we are in UTF-8 mode
1132      count        pointer to the current capturing subpattern number (updated)
1133    
1134    Returns:       the number of the named subpattern, or -1 if not found
1135    */
1136    
1137    static int
1138    find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
1139      BOOL xmode, BOOL utf8, int *count)
1140    {
1141    uschar *ptr = *ptrptr;
1142    int start_count = *count;
1143    int hwm_count = start_count;
1144    BOOL dup_parens = FALSE;
1145    
1146    /* If the first character is a parenthesis, check on the type of group we are
1147    dealing with. The very first call may not start with a parenthesis. */
1148    
1149    if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1150      {
1151      /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1152    
1153      if (ptr[1] == CHAR_ASTERISK) ptr += 2;
1154    
1155      /* Handle a normal, unnamed capturing parenthesis. */
1156    
1157      else if (ptr[1] != CHAR_QUESTION_MARK)
1158        {
1159        *count += 1;
1160        if (name == NULL && *count == lorn) return *count;
1161        ptr++;
1162        }
1163    
1164      /* All cases now have (? at the start. Remember when we are in a group
1165      where the parenthesis numbers are duplicated. */
1166    
1167      else if (ptr[2] == CHAR_VERTICAL_LINE)
1168        {
1169        ptr += 3;
1170        dup_parens = TRUE;
1171        }
1172    
1173      /* Handle comments; all characters are allowed until a ket is reached. */
1174    
1175      else if (ptr[2] == CHAR_NUMBER_SIGN)
1176        {
1177        for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1178        goto FAIL_EXIT;
1179        }
1180    
1181      /* Handle a condition. If it is an assertion, just carry on so that it
1182      is processed as normal. If not, skip to the closing parenthesis of the
1183      condition (there can't be any nested parens). */
1184    
1185      else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1186        {
1187        ptr += 2;
1188        if (ptr[1] != CHAR_QUESTION_MARK)
1189          {
1190          while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1191          if (*ptr != 0) ptr++;
1192          }
1193        }
1194    
1195      /* Start with (? but not a condition. */
1196    
1197      else
1198        {
1199        ptr += 2;
1200        if (*ptr == CHAR_P) ptr++;                      /* Allow optional P */
1201    
1202        /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1203    
1204        if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1205            ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1206          {
1207          int term;
1208          const uschar *thisname;
1209          *count += 1;
1210          if (name == NULL && *count == lorn) return *count;
1211          term = *ptr++;
1212          if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1213          thisname = ptr;
1214          while (*ptr != term) ptr++;
1215          if (name != NULL && lorn == ptr - thisname &&
1216              strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1217            return *count;
1218          term++;
1219          }
1220        }
1221      }
1222    
1223    /* Past any initial parenthesis handling, scan for parentheses or vertical
1224    bars. Stop if we get to cd->end_pattern. Note that this is important for the
1225    first-pass call when this value is temporarily adjusted to stop at the current
1226    position. So DO NOT change this to a test for binary zero. */
1227    
1228    for (; ptr < cd->end_pattern; ptr++)
1229      {
1230      /* Skip over backslashed characters and also entire \Q...\E */
1231    
1232      if (*ptr == CHAR_BACKSLASH)
1233        {
1234        if (*(++ptr) == 0) goto FAIL_EXIT;
1235        if (*ptr == CHAR_Q) for (;;)
1236          {
1237          while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1238          if (*ptr == 0) goto FAIL_EXIT;
1239          if (*(++ptr) == CHAR_E) break;
1240          }
1241        continue;
1242        }
1243    
1244      /* Skip over character classes; this logic must be similar to the way they
1245      are handled for real. If the first character is '^', skip it. Also, if the
1246      first few characters (either before or after ^) are \Q\E or \E we skip them
1247      too. This makes for compatibility with Perl. Note the use of STR macros to
1248      encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1249    
1250      if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1251        {
1252        BOOL negate_class = FALSE;
1253        for (;;)
1254          {
1255          if (ptr[1] == CHAR_BACKSLASH)
1256            {
1257            if (ptr[2] == CHAR_E)
1258              ptr+= 2;
1259            else if (strncmp((const char *)ptr+2,
1260                     STR_Q STR_BACKSLASH STR_E, 3) == 0)
1261              ptr += 4;
1262            else
1263              break;
1264            }
1265          else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1266            {
1267            negate_class = TRUE;
1268            ptr++;
1269            }
1270          else break;
1271          }
1272    
1273        /* If the next character is ']', it is a data character that must be
1274        skipped, except in JavaScript compatibility mode. */
1275    
1276        if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1277            (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1278          ptr++;
1279    
1280        while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1281          {
1282          if (*ptr == 0) return -1;
1283          if (*ptr == CHAR_BACKSLASH)
1284            {
1285            if (*(++ptr) == 0) goto FAIL_EXIT;
1286            if (*ptr == CHAR_Q) for (;;)
1287              {
1288              while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1289              if (*ptr == 0) goto FAIL_EXIT;
1290              if (*(++ptr) == CHAR_E) break;
1291              }
1292            continue;
1293            }
1294          }
1295        continue;
1296        }
1297    
1298      /* Skip comments in /x mode */
1299    
1300      if (xmode && *ptr == CHAR_NUMBER_SIGN)
1301        {
1302        ptr++;
1303        while (*ptr != 0)
1304          {
1305          if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1306          ptr++;
1307    #ifdef SUPPORT_UTF8
1308          if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
1309    #endif
1310          }
1311        if (*ptr == 0) goto FAIL_EXIT;
1312        continue;
1313        }
1314    
1315      /* Check for the special metacharacters */
1316    
1317      if (*ptr == CHAR_LEFT_PARENTHESIS)
1318        {
1319        int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, count);
1320        if (rc > 0) return rc;
1321        if (*ptr == 0) goto FAIL_EXIT;
1322        }
1323    
1324      else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1325        {
1326        if (dup_parens && *count < hwm_count) *count = hwm_count;
1327        goto FAIL_EXIT;
1328        }
1329    
1330      else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1331        {
1332        if (*count > hwm_count) hwm_count = *count;
1333        *count = start_count;
1334        }
1335      }
1336    
1337    FAIL_EXIT:
1338    *ptrptr = ptr;
1339    return -1;
1340    }
1341    
1342    
1343    
1344    
1345    /*************************************************
1346    *       Find forward referenced subpattern       *
1347    *************************************************/
1348    
1349    /* This function scans along a pattern's text looking for capturing
1350    subpatterns, and counting them. If it finds a named pattern that matches the
1351    name it is given, it returns its number. Alternatively, if the name is NULL, it
1352    returns when it reaches a given numbered subpattern. This is used for forward
1353    references to subpatterns. We used to be able to start this scan from the
1354    current compiling point, using the current count value from cd->bracount, and
1355    do it all in a single loop, but the addition of the possibility of duplicate
1356    subpattern numbers means that we have to scan from the very start, in order to
1357    take account of such duplicates, and to use a recursive function to keep track
1358    of the different types of group.
1359    
1360    Arguments:
1361      cd           compile background data
1362      name         name to seek, or NULL if seeking a numbered subpattern
1363      lorn         name length, or subpattern number if name is NULL
1364      xmode        TRUE if we are in /x mode
1365      utf8         TRUE if we are in UTF-8 mode
1366    
1367    Returns:       the number of the found subpattern, or -1 if not found
1368    */
1369    
1370    static int
1371    find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode,
1372      BOOL utf8)
1373    {
1374    uschar *ptr = (uschar *)cd->start_pattern;
1375    int count = 0;
1376    int rc;
1377    
1378    /* If the pattern does not start with an opening parenthesis, the first call
1379    to find_parens_sub() will scan right to the end (if necessary). However, if it
1380    does start with a parenthesis, find_parens_sub() will return when it hits the
1381    matching closing parens. That is why we have to have a loop. */
1382    
1383    for (;;)
1384      {
1385      rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, &count);
1386      if (rc > 0 || *ptr++ == 0) break;
1387      }
1388    
1389    return rc;
1390    }
1391    
1392    
1393    
1394    
1395    /*************************************************
1396  *      Find first significant op code            *  *      Find first significant op code            *
1397  *************************************************/  *************************************************/
1398    
1399  /* This is called by several functions that scan a compiled expression looking  /* This is called by several functions that scan a compiled expression looking
1400  for a fixed first character, or an anchoring op code etc. It skips over things  for a fixed first character, or an anchoring op code etc. It skips over things
1401  that do not influence this. For some calls, a change of option is important.  that do not influence this. For some calls, it makes sense to skip negative
1402  For some calls, it makes sense to skip negative forward and all backward  forward and all backward assertions, and also the \b assertion; for others it
1403  assertions, and also the \b assertion; for others it does not.  does not.
1404    
1405  Arguments:  Arguments:
1406    code         pointer to the start of the group    code         pointer to the start of the group
   options      pointer to external options  
   optbit       the option bit whose changing is significant, or  
                  zero if none are  
1407    skipassert   TRUE if certain assertions are to be skipped    skipassert   TRUE if certain assertions are to be skipped
1408    
1409  Returns:       pointer to the first significant opcode  Returns:       pointer to the first significant opcode
1410  */  */
1411    
1412  static const uschar*  static const uschar*
1413  first_significant_code(const uschar *code, int *options, int optbit,  first_significant_code(const uschar *code, BOOL skipassert)
   BOOL skipassert)  
1414  {  {
1415  for (;;)  for (;;)
1416    {    {
1417    switch ((int)*code)    switch ((int)*code)
1418      {      {
     case OP_OPT:  
     if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))  
       *options = (int)code[1];  
     code += 2;  
     break;  
   
1419      case OP_ASSERT_NOT:      case OP_ASSERT_NOT:
1420      case OP_ASSERTBACK:      case OP_ASSERTBACK:
1421      case OP_ASSERTBACK_NOT:      case OP_ASSERTBACK_NOT:
# Line 798  for (;;) Line 1431  for (;;)
1431    
1432      case OP_CALLOUT:      case OP_CALLOUT:
1433      case OP_CREF:      case OP_CREF:
1434      case OP_BRANUMBER:      case OP_NCREF:
1435        case OP_RREF:
1436        case OP_NRREF:
1437        case OP_DEF:
1438      code += _pcre_OP_lengths[*code];      code += _pcre_OP_lengths[*code];
1439      break;      break;
1440    
# Line 813  for (;;) Line 1449  for (;;)
1449    
1450    
1451  /*************************************************  /*************************************************
1452  *        Find the fixed length of a pattern      *  *        Find the fixed length of a branch       *
1453  *************************************************/  *************************************************/
1454    
1455  /* Scan a pattern and compute the fixed length of subject that will match it,  /* Scan a branch and compute the fixed length of subject that will match it,
1456  if the length is fixed. This is needed for dealing with backward assertions.  if the length is fixed. This is needed for dealing with backward assertions.
1457  In UTF8 mode, the result is in characters rather than bytes.  In UTF8 mode, the result is in characters rather than bytes. The branch is
1458    temporarily terminated with OP_END when this function is called.
1459    
1460    This function is called when a backward assertion is encountered, so that if it
1461    fails, the error message can point to the correct place in the pattern.
1462    However, we cannot do this when the assertion contains subroutine calls,
1463    because they can be forward references. We solve this by remembering this case
1464    and doing the check at the end; a flag specifies which mode we are running in.
1465    
1466  Arguments:  Arguments:
1467    code     points to the start of the pattern (the bracket)    code     points to the start of the pattern (the bracket)
1468    options  the compiling options    utf8     TRUE in UTF-8 mode
1469      atend    TRUE if called when the pattern is complete
1470      cd       the "compile data" structure
1471    
1472  Returns:   the fixed length, or -1 if there is no fixed length,  Returns:   the fixed length,
1473                 or -1 if there is no fixed length,
1474               or -2 if \C was encountered               or -2 if \C was encountered
1475                 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1476  */  */
1477    
1478  static int  static int
1479  find_fixedlength(uschar *code, int options)  find_fixedlength(uschar *code, BOOL utf8, BOOL atend, compile_data *cd)
1480  {  {
1481  int length = -1;  int length = -1;
1482    
# Line 842  branch, check the length against that of Line 1489  branch, check the length against that of
1489  for (;;)  for (;;)
1490    {    {
1491    int d;    int d;
1492      uschar *ce, *cs;
1493    register int op = *cc;    register int op = *cc;
   if (op >= OP_BRA) op = OP_BRA;  
   
1494    switch (op)    switch (op)
1495      {      {
1496        /* We only need to continue for OP_CBRA (normal capturing bracket) and
1497        OP_BRA (normal non-capturing bracket) because the other variants of these
1498        opcodes are all concerned with unlimited repeated groups, which of course
1499        are not of fixed length. They will cause a -1 response from the default
1500        case of this switch. */
1501    
1502        case OP_CBRA:
1503      case OP_BRA:      case OP_BRA:
1504      case OP_ONCE:      case OP_ONCE:
1505      case OP_COND:      case OP_COND:
1506      d = find_fixedlength(cc, options);      d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), utf8, atend, cd);
1507      if (d < 0) return d;      if (d < 0) return d;
1508      branchlength += d;      branchlength += d;
1509      do cc += GET(cc, 1); while (*cc == OP_ALT);      do cc += GET(cc, 1); while (*cc == OP_ALT);
# Line 859  for (;;) Line 1512  for (;;)
1512    
1513      /* Reached end of a branch; if it's a ket it is the end of a nested      /* Reached end of a branch; if it's a ket it is the end of a nested
1514      call. If it's ALT it is an alternation in a nested call. If it is      call. If it's ALT it is an alternation in a nested call. If it is
1515      END it's the end of the outer call. All can be handled by the same code. */      END it's the end of the outer call. All can be handled by the same code.
1516        Note that we must not include the OP_KETRxxx opcodes here, because they
1517        all imply an unlimited repeat. */
1518    
1519      case OP_ALT:      case OP_ALT:
1520      case OP_KET:      case OP_KET:
     case OP_KETRMAX:  
     case OP_KETRMIN:  
1521      case OP_END:      case OP_END:
1522      if (length < 0) length = branchlength;      if (length < 0) length = branchlength;
1523        else if (length != branchlength) return -1;        else if (length != branchlength) return -1;
# Line 873  for (;;) Line 1526  for (;;)
1526      branchlength = 0;      branchlength = 0;
1527      break;      break;
1528    
1529      /* Skip over assertive subpatterns */      /* A true recursion implies not fixed length, but a subroutine call may
1530        be OK. If the subroutine is a forward reference, we can't deal with
1531      case OP_ASSERT:      it until the end of the pattern, so return -3. */
1532      case OP_ASSERT_NOT:  
1533        case OP_RECURSE:
1534        if (!atend) return -3;
1535        cs = ce = (uschar *)cd->start_code + GET(cc, 1);  /* Start subpattern */
1536        do ce += GET(ce, 1); while (*ce == OP_ALT);       /* End subpattern */
1537        if (cc > cs && cc < ce) return -1;                /* Recursion */
1538        d = find_fixedlength(cs + 2, utf8, atend, cd);
1539        if (d < 0) return d;
1540        branchlength += d;
1541        cc += 1 + LINK_SIZE;
1542        break;
1543    
1544        /* Skip over assertive subpatterns */
1545    
1546        case OP_ASSERT:
1547        case OP_ASSERT_NOT:
1548      case OP_ASSERTBACK:      case OP_ASSERTBACK:
1549      case OP_ASSERTBACK_NOT:      case OP_ASSERTBACK_NOT:
1550      do cc += GET(cc, 1); while (*cc == OP_ALT);      do cc += GET(cc, 1); while (*cc == OP_ALT);
# Line 885  for (;;) Line 1553  for (;;)
1553      /* Skip over things that don't match chars */      /* Skip over things that don't match chars */
1554    
1555      case OP_REVERSE:      case OP_REVERSE:
     case OP_BRANUMBER:  
1556      case OP_CREF:      case OP_CREF:
1557      case OP_OPT:      case OP_NCREF:
1558        case OP_RREF:
1559        case OP_NRREF:
1560        case OP_DEF:
1561      case OP_CALLOUT:      case OP_CALLOUT:
1562      case OP_SOD:      case OP_SOD:
1563      case OP_SOM:      case OP_SOM:
1564        case OP_SET_SOM:
1565      case OP_EOD:      case OP_EOD:
1566      case OP_EODN:      case OP_EODN:
1567      case OP_CIRC:      case OP_CIRC:
1568        case OP_CIRCM:
1569      case OP_DOLL:      case OP_DOLL:
1570        case OP_DOLLM:
1571      case OP_NOT_WORD_BOUNDARY:      case OP_NOT_WORD_BOUNDARY:
1572      case OP_WORD_BOUNDARY:      case OP_WORD_BOUNDARY:
1573      cc += _pcre_OP_lengths[*cc];      cc += _pcre_OP_lengths[*cc];
# Line 903  for (;;) Line 1576  for (;;)
1576      /* Handle literal characters */      /* Handle literal characters */
1577    
1578      case OP_CHAR:      case OP_CHAR:
1579      case OP_CHARNC:      case OP_CHARI:
1580        case OP_NOT:
1581        case OP_NOTI:
1582      branchlength++;      branchlength++;
1583      cc += 2;      cc += 2;
1584  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
1585      if ((options & PCRE_UTF8) != 0)      if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
       {  
       while ((*cc & 0xc0) == 0x80) cc++;  
       }  
1586  #endif  #endif
1587      break;      break;
1588    
# Line 921  for (;;) Line 1593  for (;;)
1593      branchlength += GET2(cc,1);      branchlength += GET2(cc,1);
1594      cc += 4;      cc += 4;
1595  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
1596      if ((options & PCRE_UTF8) != 0)      if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
       {  
       while((*cc & 0x80) == 0x80) cc++;  
       }  
1597  #endif  #endif
1598      break;      break;
1599    
1600      case OP_TYPEEXACT:      case OP_TYPEEXACT:
1601      branchlength += GET2(cc,1);      branchlength += GET2(cc,1);
1602        if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1603      cc += 4;      cc += 4;
1604      break;      break;
1605    
# Line 937  for (;;) Line 1607  for (;;)
1607    
1608      case OP_PROP:      case OP_PROP:
1609      case OP_NOTPROP:      case OP_NOTPROP:
1610      cc++;      cc += 2;
1611      /* Fall through */      /* Fall through */
1612    
1613      case OP_NOT_DIGIT:      case OP_NOT_DIGIT:
# Line 947  for (;;) Line 1617  for (;;)
1617      case OP_NOT_WORDCHAR:      case OP_NOT_WORDCHAR:
1618      case OP_WORDCHAR:      case OP_WORDCHAR:
1619      case OP_ANY:      case OP_ANY:
1620        case OP_ALLANY:
1621      branchlength++;      branchlength++;
1622      cc++;      cc++;
1623      break;      break;
# Line 1001  for (;;) Line 1672  for (;;)
1672    
1673    
1674  /*************************************************  /*************************************************
1675  *    Scan compiled regex for numbered bracket    *  *    Scan compiled regex for specific bracket    *
1676  *************************************************/  *************************************************/
1677    
1678  /* This little function scans through a compiled pattern until it finds a  /* This little function scans through a compiled pattern until it finds a
1679  capturing bracket with the given number.  capturing bracket with the given number, or, if the number is negative, an
1680    instance of OP_REVERSE for a lookbehind. The function is global in the C sense
1681    so that it can be called from pcre_study() when finding the minimum matching
1682    length.
1683    
1684  Arguments:  Arguments:
1685    code        points to start of expression    code        points to start of expression
1686    utf8        TRUE in UTF-8 mode    utf8        TRUE in UTF-8 mode
1687    number      the required bracket number    number      the required bracket number or negative to find a lookbehind
1688    
1689  Returns:      pointer to the opcode for the bracket, or NULL if not found  Returns:      pointer to the opcode for the bracket, or NULL if not found
1690  */  */
1691    
1692  static const uschar *  const uschar *
1693  find_bracket(const uschar *code, BOOL utf8, int number)  _pcre_find_bracket(const uschar *code, BOOL utf8, int number)
1694  {  {
 #ifndef SUPPORT_UTF8  
 utf8 = utf8;               /* Stop pedantic compilers complaining */  
 #endif  
   
1695  for (;;)  for (;;)
1696    {    {
1697    register int c = *code;    register int c = *code;
1698    
1699    if (c == OP_END) return NULL;    if (c == OP_END) return NULL;
1700    else if (c > OP_BRA)  
1701      /* XCLASS is used for classes that cannot be represented just by a bit
1702      map. This includes negated single high-valued characters. The length in
1703      the table is zero; the actual length is stored in the compiled code. */
1704    
1705      if (c == OP_XCLASS) code += GET(code, 1);
1706    
1707      /* Handle recursion */
1708    
1709      else if (c == OP_REVERSE)
1710      {      {
1711      int n = c - OP_BRA;      if (number < 0) return (uschar *)code;
1712      if (n > EXTRACT_BASIC_MAX) n = GET2(code, 2+LINK_SIZE);      code += _pcre_OP_lengths[c];
1713        }
1714    
1715      /* Handle capturing bracket */
1716    
1717      else if (c == OP_CBRA || c == OP_SCBRA ||
1718               c == OP_CBRAPOS || c == OP_SCBRAPOS)
1719        {
1720        int n = GET2(code, 1+LINK_SIZE);
1721      if (n == number) return (uschar *)code;      if (n == number) return (uschar *)code;
1722      code += _pcre_OP_lengths[OP_BRA];      code += _pcre_OP_lengths[c];
1723      }      }
1724    
1725      /* Otherwise, we can get the item's length from the table, except that for
1726      repeated character types, we have to test for \p and \P, which have an extra
1727      two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1728      must add in its length. */
1729    
1730    else    else
1731      {      {
1732      code += _pcre_OP_lengths[c];      switch(c)
1733          {
1734          case OP_TYPESTAR:
1735          case OP_TYPEMINSTAR:
1736          case OP_TYPEPLUS:
1737          case OP_TYPEMINPLUS:
1738          case OP_TYPEQUERY:
1739          case OP_TYPEMINQUERY:
1740          case OP_TYPEPOSSTAR:
1741          case OP_TYPEPOSPLUS:
1742          case OP_TYPEPOSQUERY:
1743          if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1744          break;
1745    
1746  #ifdef SUPPORT_UTF8        case OP_TYPEUPTO:
1747          case OP_TYPEMINUPTO:
1748          case OP_TYPEEXACT:
1749          case OP_TYPEPOSUPTO:
1750          if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1751          break;
1752    
1753      /* In UTF-8 mode, opcodes that are followed by a character may be followed        case OP_MARK:
1754      by a multi-byte character. The length in the table is a minimum, so we have        case OP_PRUNE_ARG:
1755      to scan along to skip the extra bytes. All opcodes are less than 128, so we        case OP_SKIP_ARG:
1756      can use relatively efficient code. */        code += code[1];
1757          break;
1758    
1759          case OP_THEN_ARG:
1760          code += code[1+LINK_SIZE];
1761          break;
1762          }
1763    
1764        /* Add in the fixed length from the table */
1765    
1766        code += _pcre_OP_lengths[c];
1767    
1768      /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1769      a multi-byte character. The length in the table is a minimum, so we have to
1770      arrange to skip the extra bytes. */
1771    
1772    #ifdef SUPPORT_UTF8
1773      if (utf8) switch(c)      if (utf8) switch(c)
1774        {        {
1775        case OP_CHAR:        case OP_CHAR:
1776        case OP_CHARNC:        case OP_CHARI:
1777        case OP_EXACT:        case OP_EXACT:
1778          case OP_EXACTI:
1779        case OP_UPTO:        case OP_UPTO:
1780          case OP_UPTOI:
1781        case OP_MINUPTO:        case OP_MINUPTO:
1782          case OP_MINUPTOI:
1783          case OP_POSUPTO:
1784          case OP_POSUPTOI:
1785        case OP_STAR:        case OP_STAR:
1786          case OP_STARI:
1787        case OP_MINSTAR:        case OP_MINSTAR:
1788          case OP_MINSTARI:
1789          case OP_POSSTAR:
1790          case OP_POSSTARI:
1791        case OP_PLUS:        case OP_PLUS:
1792          case OP_PLUSI:
1793        case OP_MINPLUS:        case OP_MINPLUS:
1794          case OP_MINPLUSI:
1795          case OP_POSPLUS:
1796          case OP_POSPLUSI:
1797        case OP_QUERY:        case OP_QUERY:
1798          case OP_QUERYI:
1799        case OP_MINQUERY:        case OP_MINQUERY:
1800        while ((*code & 0xc0) == 0x80) code++;        case OP_MINQUERYI:
1801        break;        case OP_POSQUERY:
1802          case OP_POSQUERYI:
1803        /* XCLASS is used for classes that cannot be represented just by a bit        if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
       map. This includes negated single high-valued characters. The length in  
       the table is zero; the actual length is stored in the compiled code. */  
   
       case OP_XCLASS:  
       code += GET(code, 1) + 1;  
1804        break;        break;
1805        }        }
1806    #else
1807        (void)(utf8);  /* Keep compiler happy by referencing function argument */
1808  #endif  #endif
1809      }      }
1810    }    }
# Line 1092  Returns:      pointer to the opcode for Line 1829  Returns:      pointer to the opcode for
1829  static const uschar *  static const uschar *
1830  find_recurse(const uschar *code, BOOL utf8)  find_recurse(const uschar *code, BOOL utf8)
1831  {  {
 #ifndef SUPPORT_UTF8  
 utf8 = utf8;               /* Stop pedantic compilers complaining */  
 #endif  
   
1832  for (;;)  for (;;)
1833    {    {
1834    register int c = *code;    register int c = *code;
1835    if (c == OP_END) return NULL;    if (c == OP_END) return NULL;
1836    else if (c == OP_RECURSE) return code;    if (c == OP_RECURSE) return code;
1837    else if (c > OP_BRA)  
1838      {    /* XCLASS is used for classes that cannot be represented just by a bit
1839      code += _pcre_OP_lengths[OP_BRA];    map. This includes negated single high-valued characters. The length in
1840      }    the table is zero; the actual length is stored in the compiled code. */
1841    
1842      if (c == OP_XCLASS) code += GET(code, 1);
1843    
1844      /* Otherwise, we can get the item's length from the table, except that for
1845      repeated character types, we have to test for \p and \P, which have an extra
1846      two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1847      must add in its length. */
1848    
1849    else    else
1850      {      {
1851      code += _pcre_OP_lengths[c];      switch(c)
1852          {
1853          case OP_TYPESTAR:
1854          case OP_TYPEMINSTAR:
1855          case OP_TYPEPLUS:
1856          case OP_TYPEMINPLUS:
1857          case OP_TYPEQUERY:
1858          case OP_TYPEMINQUERY:
1859          case OP_TYPEPOSSTAR:
1860          case OP_TYPEPOSPLUS:
1861          case OP_TYPEPOSQUERY:
1862          if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1863          break;
1864    
1865  #ifdef SUPPORT_UTF8        case OP_TYPEPOSUPTO:
1866          case OP_TYPEUPTO:
1867          case OP_TYPEMINUPTO:
1868          case OP_TYPEEXACT:
1869          if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1870          break;
1871    
1872          case OP_MARK:
1873          case OP_PRUNE_ARG:
1874          case OP_SKIP_ARG:
1875          code += code[1];
1876          break;
1877    
1878          case OP_THEN_ARG:
1879          code += code[1+LINK_SIZE];
1880          break;
1881          }
1882    
1883        /* Add in the fixed length from the table */
1884    
1885        code += _pcre_OP_lengths[c];
1886    
1887      /* In UTF-8 mode, opcodes that are followed by a character may be followed      /* In UTF-8 mode, opcodes that are followed by a character may be followed
1888      by a multi-byte character. The length in the table is a minimum, so we have      by a multi-byte character. The length in the table is a minimum, so we have
1889      to scan along to skip the extra bytes. All opcodes are less than 128, so we      to arrange to skip the extra bytes. */
     can use relatively efficient code. */  
1890    
1891    #ifdef SUPPORT_UTF8
1892      if (utf8) switch(c)      if (utf8) switch(c)
1893        {        {
1894        case OP_CHAR:        case OP_CHAR:
1895        case OP_CHARNC:        case OP_CHARI:
1896        case OP_EXACT:        case OP_EXACT:
1897          case OP_EXACTI:
1898        case OP_UPTO:        case OP_UPTO:
1899          case OP_UPTOI:
1900        case OP_MINUPTO:        case OP_MINUPTO:
1901          case OP_MINUPTOI:
1902          case OP_POSUPTO:
1903          case OP_POSUPTOI:
1904        case OP_STAR:        case OP_STAR:
1905          case OP_STARI:
1906        case OP_MINSTAR:        case OP_MINSTAR:
1907          case OP_MINSTARI:
1908          case OP_POSSTAR:
1909          case OP_POSSTARI:
1910        case OP_PLUS:        case OP_PLUS:
1911          case OP_PLUSI:
1912        case OP_MINPLUS:        case OP_MINPLUS:
1913          case OP_MINPLUSI:
1914          case OP_POSPLUS:
1915          case OP_POSPLUSI:
1916        case OP_QUERY:        case OP_QUERY:
1917          case OP_QUERYI:
1918        case OP_MINQUERY:        case OP_MINQUERY:
1919        while ((*code & 0xc0) == 0x80) code++;        case OP_MINQUERYI:
1920        break;        case OP_POSQUERY:
1921          case OP_POSQUERYI:
1922        /* XCLASS is used for classes that cannot be represented just by a bit        if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
       map. This includes negated single high-valued characters. The length in  
       the table is zero; the actual length is stored in the compiled code. */  
   
       case OP_XCLASS:  
       code += GET(code, 1) + 1;  
1923        break;        break;
1924        }        }
1925    #else
1926        (void)(utf8);  /* Keep compiler happy by referencing function argument */
1927  #endif  #endif
1928      }      }
1929    }    }
# Line 1152  for (;;) Line 1936  for (;;)
1936  *************************************************/  *************************************************/
1937    
1938  /* This function scans through a branch of a compiled pattern to see whether it  /* This function scans through a branch of a compiled pattern to see whether it
1939  can match the empty string or not. It is called only from could_be_empty()  can match the empty string or not. It is called from could_be_empty()
1940  below. Note that first_significant_code() skips over assertions. If we hit an  below and from compile_branch() when checking for an unlimited repeat of a
1941  unclosed bracket, we return "empty" - this means we've struck an inner bracket  group that can match nothing. Note that first_significant_code() skips over
1942  whose current branch will already have been scanned.  backward and negative forward assertions when its final argument is TRUE. If we
1943    hit an unclosed bracket, we return "empty" - this means we've struck an inner
1944    bracket whose current branch will already have been scanned.
1945    
1946  Arguments:  Arguments:
1947    code        points to start of search    code        points to start of search
1948    endcode     points to where to stop    endcode     points to where to stop
1949    utf8        TRUE if in UTF8 mode    utf8        TRUE if in UTF8 mode
1950      cd          contains pointers to tables etc.
1951    
1952  Returns:      TRUE if what is matched could be empty  Returns:      TRUE if what is matched could be empty
1953  */  */
1954    
1955  static BOOL  static BOOL
1956  could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8)  could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8,
1957      compile_data *cd)
1958  {  {
1959  register int c;  register int c;
1960  for (code = first_significant_code(code + 1 + LINK_SIZE, NULL, 0, TRUE);  for (code = first_significant_code(code + _pcre_OP_lengths[*code], TRUE);
1961       code < endcode;       code < endcode;
1962       code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))       code = first_significant_code(code + _pcre_OP_lengths[c], TRUE))
1963    {    {
1964    const uschar *ccode;    const uschar *ccode;
1965    
1966    c = *code;    c = *code;
1967    
1968    if (c >= OP_BRA)    /* Skip over forward assertions; the other assertions are skipped by
1969      first_significant_code() with a TRUE final argument. */
1970    
1971      if (c == OP_ASSERT)
1972      {      {
1973      BOOL empty_branch;      do code += GET(code, 1); while (*code == OP_ALT);
1974      if (GET(code, 1) == 0) return TRUE;    /* Hit unclosed bracket */      c = *code;
1975        continue;
1976        }
1977    
1978      /* For a recursion/subroutine call, if its end has been reached, which
1979      implies a backward reference subroutine call, we can scan it. If it's a
1980      forward reference subroutine call, we can't. To detect forward reference
1981      we have to scan up the list that is kept in the workspace. This function is
1982      called only when doing the real compile, not during the pre-compile that
1983      measures the size of the compiled pattern. */
1984    
1985      /* Scan a closed bracket */    if (c == OP_RECURSE)
1986        {
1987        const uschar *scode;
1988        BOOL empty_branch;
1989    
1990        /* Test for forward reference */
1991    
1992        for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
1993          if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
1994    
1995        /* Not a forward reference, test for completed backward reference */
1996    
1997      empty_branch = FALSE;      empty_branch = FALSE;
1998        scode = cd->start_code + GET(code, 1);
1999        if (GET(scode, 1) == 0) return TRUE;    /* Unclosed */
2000    
2001        /* Completed backwards reference */
2002    
2003      do      do
2004        {        {
2005        if (!empty_branch && could_be_empty_branch(code, endcode, utf8))        if (could_be_empty_branch(scode, endcode, utf8, cd))
2006            {
2007          empty_branch = TRUE;          empty_branch = TRUE;
2008            break;
2009            }
2010          scode += GET(scode, 1);
2011          }
2012        while (*scode == OP_ALT);
2013    
2014        if (!empty_branch) return FALSE;  /* All branches are non-empty */
2015        continue;
2016        }
2017    
2018      /* Groups with zero repeats can of course be empty; skip them. */
2019    
2020      if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2021          c == OP_BRAPOSZERO)
2022        {
2023        code += _pcre_OP_lengths[c];
2024        do code += GET(code, 1); while (*code == OP_ALT);
2025        c = *code;
2026        continue;
2027        }
2028    
2029      /* A nested group that is already marked as "could be empty" can just be
2030      skipped. */
2031    
2032      if (c == OP_SBRA  || c == OP_SBRAPOS ||
2033          c == OP_SCBRA || c == OP_SCBRAPOS)
2034        {
2035        do code += GET(code, 1); while (*code == OP_ALT);
2036        c = *code;
2037        continue;
2038        }
2039    
2040      /* For other groups, scan the branches. */
2041    
2042      if (c == OP_BRA  || c == OP_BRAPOS ||
2043          c == OP_CBRA || c == OP_CBRAPOS ||
2044          c == OP_ONCE || c == OP_COND)
2045        {
2046        BOOL empty_branch;
2047        if (GET(code, 1) == 0) return TRUE;    /* Hit unclosed bracket */
2048    
2049        /* If a conditional group has only one branch, there is a second, implied,
2050        empty branch, so just skip over the conditional, because it could be empty.
2051        Otherwise, scan the individual branches of the group. */
2052    
2053        if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2054        code += GET(code, 1);        code += GET(code, 1);
2055        else
2056          {
2057          empty_branch = FALSE;
2058          do
2059            {
2060            if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
2061              empty_branch = TRUE;
2062            code += GET(code, 1);
2063            }
2064          while (*code == OP_ALT);
2065          if (!empty_branch) return FALSE;   /* All branches are non-empty */
2066        }        }
2067      while (*code == OP_ALT);  
     if (!empty_branch) return FALSE;   /* All branches are non-empty */  
     code += 1 + LINK_SIZE;  
2068      c = *code;      c = *code;
2069        continue;
2070      }      }
2071    
2072    else switch (c)    /* Handle the other opcodes */
2073    
2074      switch (c)
2075      {      {
2076      /* Check for quantifiers after a class */      /* Check for quantifiers after a class. XCLASS is used for classes that
2077        cannot be represented just by a bit map. This includes negated single
2078        high-valued characters. The length in _pcre_OP_lengths[] is zero; the
2079        actual length is stored in the compiled code, so we must update "code"
2080        here. */
2081    
2082  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
2083      case OP_XCLASS:      case OP_XCLASS:
2084      ccode = code + GET(code, 1);      ccode = code += GET(code, 1);
2085      goto CHECK_CLASS_REPEAT;      goto CHECK_CLASS_REPEAT;
2086  #endif  #endif
2087    
# Line 1247  for (code = first_significant_code(code Line 2125  for (code = first_significant_code(code
2125      case OP_NOT_WORDCHAR:      case OP_NOT_WORDCHAR:
2126      case OP_WORDCHAR:      case OP_WORDCHAR:
2127      case OP_ANY:      case OP_ANY:
2128        case OP_ALLANY:
2129      case OP_ANYBYTE:      case OP_ANYBYTE:
2130      case OP_CHAR:      case OP_CHAR:
2131      case OP_CHARNC:      case OP_CHARI:
2132      case OP_NOT:      case OP_NOT:
2133        case OP_NOTI:
2134      case OP_PLUS:      case OP_PLUS:
2135      case OP_MINPLUS:      case OP_MINPLUS:
2136        case OP_POSPLUS:
2137      case OP_EXACT:      case OP_EXACT:
2138      case OP_NOTPLUS:      case OP_NOTPLUS:
2139      case OP_NOTMINPLUS:      case OP_NOTMINPLUS:
2140        case OP_NOTPOSPLUS:
2141      case OP_NOTEXACT:      case OP_NOTEXACT:
2142      case OP_TYPEPLUS:      case OP_TYPEPLUS:
2143      case OP_TYPEMINPLUS:      case OP_TYPEMINPLUS:
2144        case OP_TYPEPOSPLUS:
2145      case OP_TYPEEXACT:      case OP_TYPEEXACT:
2146      return FALSE;      return FALSE;
2147    
2148        /* These are going to continue, as they may be empty, but we have to
2149        fudge the length for the \p and \P cases. */
2150    
2151        case OP_TYPESTAR:
2152        case OP_TYPEMINSTAR:
2153        case OP_TYPEPOSSTAR:
2154        case OP_TYPEQUERY:
2155        case OP_TYPEMINQUERY:
2156        case OP_TYPEPOSQUERY:
2157        if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2158        break;
2159    
2160        /* Same for these */
2161    
2162        case OP_TYPEUPTO:
2163        case OP_TYPEMINUPTO:
2164        case OP_TYPEPOSUPTO:
2165        if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
2166        break;
2167    
2168      /* End of branch */      /* End of branch */
2169    
2170      case OP_KET:      case OP_KET:
2171      case OP_KETRMAX:      case OP_KETRMAX:
2172      case OP_KETRMIN:      case OP_KETRMIN:
2173        case OP_KETRPOS:
2174      case OP_ALT:      case OP_ALT:
2175      return TRUE;      return TRUE;
2176    
2177      /* In UTF-8 mode, STAR, MINSTAR, QUERY, MINQUERY, UPTO, and MINUPTO  may be      /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2178      followed by a multibyte character */      MINUPTO, and POSUPTO may be followed by a multibyte character */
2179    
2180  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
2181      case OP_STAR:      case OP_STAR:
2182        case OP_STARI:
2183      case OP_MINSTAR:      case OP_MINSTAR:
2184        case OP_MINSTARI:
2185        case OP_POSSTAR:
2186        case OP_POSSTARI:
2187      case OP_QUERY:      case OP_QUERY:
2188        case OP_QUERYI:
2189      case OP_MINQUERY:      case OP_MINQUERY:
2190        case OP_MINQUERYI:
2191        case OP_POSQUERY:
2192        case OP_POSQUERYI:
2193        if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
2194        break;
2195    
2196      case OP_UPTO:      case OP_UPTO:
2197        case OP_UPTOI:
2198      case OP_MINUPTO:      case OP_MINUPTO:
2199      if (utf8) while ((code[2] & 0xc0) == 0x80) code++;      case OP_MINUPTOI:
2200        case OP_POSUPTO:
2201        case OP_POSUPTOI:
2202        if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
2203      break;      break;
2204  #endif  #endif
2205    
2206        /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2207        string. */
2208    
2209        case OP_MARK:
2210        case OP_PRUNE_ARG:
2211        case OP_SKIP_ARG:
2212        code += code[1];
2213        break;
2214    
2215        case OP_THEN_ARG:
2216        code += code[1+LINK_SIZE];
2217        break;
2218    
2219        /* None of the remaining opcodes are required to match a character. */
2220    
2221        default:
2222        break;
2223      }      }
2224    }    }
2225    
# Line 1299  return TRUE; Line 2236  return TRUE;
2236  the current branch of the current pattern to see if it could match the empty  the current branch of the current pattern to see if it could match the empty
2237  string. If it could, we must look outwards for branches at other levels,  string. If it could, we must look outwards for branches at other levels,
2238  stopping when we pass beyond the bracket which is the subject of the recursion.  stopping when we pass beyond the bracket which is the subject of the recursion.
2239    This function is called only during the real compile, not during the
2240    pre-compile.
2241    
2242  Arguments:  Arguments:
2243    code        points to start of the recursion    code        points to start of the recursion
2244    endcode     points to where to stop (current RECURSE item)    endcode     points to where to stop (current RECURSE item)
2245    bcptr       points to the chain of current (unclosed) branch starts    bcptr       points to the chain of current (unclosed) branch starts
2246    utf8        TRUE if in UTF-8 mode    utf8        TRUE if in UTF-8 mode
2247      cd          pointers to tables etc
2248    
2249  Returns:      TRUE if what is matched could be empty  Returns:      TRUE if what is matched could be empty
2250  */  */
2251    
2252  static BOOL  static BOOL
2253  could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,  could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
2254    BOOL utf8)    BOOL utf8, compile_data *cd)
2255  {  {
2256  while (bcptr != NULL && bcptr->current >= code)  while (bcptr != NULL && bcptr->current_branch >= code)
2257    {    {
2258    if (!could_be_empty_branch(bcptr->current, endcode, utf8)) return FALSE;    if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
2259        return FALSE;
2260    bcptr = bcptr->outer;    bcptr = bcptr->outer;
2261    }    }
2262  return TRUE;  return TRUE;
# Line 1328  return TRUE; Line 2269  return TRUE;
2269  *************************************************/  *************************************************/
2270    
2271  /* This function is called when the sequence "[:" or "[." or "[=" is  /* This function is called when the sequence "[:" or "[." or "[=" is
2272  encountered in a character class. It checks whether this is followed by an  encountered in a character class. It checks whether this is followed by a
2273  optional ^ and then a sequence of letters, terminated by a matching ":]" or  sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2274  ".]" or "=]".  reach an unescaped ']' without the special preceding character, return FALSE.
2275    
2276    Originally, this function only recognized a sequence of letters between the
2277    terminators, but it seems that Perl recognizes any sequence of characters,
2278    though of course unknown POSIX names are subsequently rejected. Perl gives an
2279    "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2280    didn't consider this to be a POSIX class. Likewise for [:1234:].
2281    
2282    The problem in trying to be exactly like Perl is in the handling of escapes. We
2283    have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2284    class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2285    below handles the special case of \], but does not try to do any other escape
2286    processing. This makes it different from Perl for cases such as [:l\ower:]
2287    where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2288    "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2289    I think.
2290    
2291  Argument:  Arguments:
2292    ptr      pointer to the initial [    ptr      pointer to the initial [
2293    endptr   where to return the end pointer    endptr   where to return the end pointer
   cd       pointer to compile data  
2294    
2295  Returns:   TRUE or FALSE  Returns:   TRUE or FALSE
2296  */  */
2297    
2298  static BOOL  static BOOL
2299  check_posix_syntax(const uschar *ptr, const uschar **endptr, compile_data *cd)  check_posix_syntax(const uschar *ptr, const uschar **endptr)
2300  {  {
2301  int terminator;          /* Don't combine these lines; the Solaris cc */  int terminator;          /* Don't combine these lines; the Solaris cc */
2302  terminator = *(++ptr);   /* compiler warns about "non-constant" initializer. */  terminator = *(++ptr);   /* compiler warns about "non-constant" initializer. */
2303  if (*(++ptr) == '^') ptr++;  for (++ptr; *ptr != 0; ptr++)
 while ((cd->ctypes[*ptr] & ctype_letter) != 0) ptr++;  
 if (*ptr == terminator && ptr[1] == ']')  
2304    {    {
2305    *endptr = ptr;    if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) ptr++; else
2306    return TRUE;      {
2307        if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2308        if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2309          {
2310          *endptr = ptr;
2311          return TRUE;
2312          }
2313        }
2314    }    }
2315  return FALSE;  return FALSE;
2316  }  }
# Line 1375  Returns:     a value representing the na Line 2335  Returns:     a value representing the na
2335  static int  static int
2336  check_posix_name(const uschar *ptr, int len)  check_posix_name(const uschar *ptr, int len)
2337  {  {
2338    const char *pn = posix_names;
2339  register int yield = 0;  register int yield = 0;
2340  while (posix_name_lengths[yield] != 0)  while (posix_name_lengths[yield] != 0)
2341    {    {
2342    if (len == posix_name_lengths[yield] &&    if (len == posix_name_lengths[yield] &&
2343      strncmp((const char *)ptr, posix_names[yield], len) == 0) return yield;      strncmp((const char *)ptr, pn, len) == 0) return yield;
2344      pn += posix_name_lengths[yield] + 1;
2345    yield++;    yield++;
2346    }    }
2347  return -1;  return -1;
# Line 1394  return -1; Line 2356  return -1;
2356  that is referenced. This means that groups can be replicated for fixed  that is referenced. This means that groups can be replicated for fixed
2357  repetition simply by copying (because the recursion is allowed to refer to  repetition simply by copying (because the recursion is allowed to refer to
2358  earlier groups that are outside the current group). However, when a group is  earlier groups that are outside the current group). However, when a group is
2359  optional (i.e. the minimum quantifier is zero), OP_BRAZERO is inserted before  optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2360  it, after it has been compiled. This means that any OP_RECURSE items within it  inserted before it, after it has been compiled. This means that any OP_RECURSE
2361  that refer to the group itself or any contained groups have to have their  items within it that refer to the group itself or any contained groups have to
2362  offsets adjusted. That is the job of this function. Before it is called, the  have their offsets adjusted. That one of the jobs of this function. Before it
2363  partially compiled regex must be temporarily terminated with OP_END.  is called, the partially compiled regex must be temporarily terminated with
2364    OP_END.
2365    
2366    This function has been extended with the possibility of forward references for
2367    recursions and subroutine calls. It must also check the list of such references
2368    for the group we are dealing with. If it finds that one of the recursions in
2369    the current group is on this list, it adjusts the offset in the list, not the
2370    value in the reference (which is a group number).
2371    
2372  Arguments:  Arguments:
2373    group      points to the start of the group    group      points to the start of the group
2374    adjust     the amount by which the group is to be moved    adjust     the amount by which the group is to be moved
2375    utf8       TRUE in UTF-8 mode    utf8       TRUE in UTF-8 mode
2376    cd         contains pointers to tables etc.    cd         contains pointers to tables etc.
2377      save_hwm   the hwm forward reference pointer at the start of the group
2378    
2379  Returns:     nothing  Returns:     nothing
2380  */  */
2381    
2382  static void  static void
2383  adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd)  adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
2384      uschar *save_hwm)
2385  {  {
2386  uschar *ptr = group;  uschar *ptr = group;
2387    
2388  while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)  while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
2389    {    {
2390    int offset = GET(ptr, 1);    int offset;
2391    if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);    uschar *hc;
2392    
2393      /* See if this recursion is on the forward reference list. If so, adjust the
2394      reference. */
2395    
2396      for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2397        {
2398        offset = GET(hc, 0);
2399        if (cd->start_code + offset == ptr + 1)
2400          {
2401          PUT(hc, 0, offset + adjust);
2402          break;
2403          }
2404        }
2405    
2406      /* Otherwise, adjust the recursion offset if it's after the start of this
2407      group. */
2408    
2409      if (hc >= cd->hwm)
2410        {
2411        offset = GET(ptr, 1);
2412        if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2413        }
2414    
2415    ptr += 1 + LINK_SIZE;    ptr += 1 + LINK_SIZE;
2416    }    }
2417  }  }
# Line 1443  auto_callout(uschar *code, const uschar Line 2438  auto_callout(uschar *code, const uschar
2438  {  {
2439  *code++ = OP_CALLOUT;  *code++ = OP_CALLOUT;
2440  *code++ = 255;  *code++ = 255;
2441  PUT(code, 0, ptr - cd->start_pattern);  /* Pattern offset */  PUT(code, 0, (int)(ptr - cd->start_pattern));  /* Pattern offset */
2442  PUT(code, LINK_SIZE, 0);                /* Default length */  PUT(code, LINK_SIZE, 0);                       /* Default length */
2443  return code + 2*LINK_SIZE;  return code + 2*LINK_SIZE;
2444  }  }
2445    
# Line 1469  Returns:             nothing Line 2464  Returns:             nothing
2464  static void  static void
2465  complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)  complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2466  {  {
2467  int length = ptr - cd->start_pattern - GET(previous_callout, 2);  int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2468  PUT(previous_callout, 2 + LINK_SIZE, length);  PUT(previous_callout, 2 + LINK_SIZE, length);
2469  }  }
2470    
# Line 1495  Yield:        TRUE when range returned; Line 2490  Yield:        TRUE when range returned;
2490  */  */
2491    
2492  static BOOL  static BOOL
2493  get_othercase_range(int *cptr, int d, int *ocptr, int *odptr)  get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2494      unsigned int *odptr)
2495  {  {
2496  int c, chartype, othercase, next;  unsigned int c, othercase, next;
2497    
2498  for (c = *cptr; c <= d; c++)  for (c = *cptr; c <= d; c++)
2499    {    { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
   if (_pcre_ucp_findchar(c, &chartype, &othercase) == ucp_L && othercase != 0)  
     break;  
   }  
2500    
2501  if (c > d) return FALSE;  if (c > d) return FALSE;
2502    
# Line 1512  next = othercase + 1; Line 2505  next = othercase + 1;
2505    
2506  for (++c; c <= d; c++)  for (++c; c <= d; c++)
2507    {    {
2508    if (_pcre_ucp_findchar(c, &chartype, &othercase) != ucp_L ||    if (UCD_OTHERCASE(c) != next) break;
         othercase != next)  
     break;  
2509    next++;    next++;
2510    }    }
2511    
# Line 1523  for (++c; c <= d; c++) Line 2514  for (++c; c <= d; c++)
2514    
2515  return TRUE;  return TRUE;
2516  }  }
2517  #endif  /* SUPPORT_UCP */  
2518    
2519    
2520  /*************************************************  /*************************************************
2521  *           Compile one branch                   *  *        Check a character and a property        *
2522  *************************************************/  *************************************************/
2523    
2524  /* Scan the pattern, compiling it into the code vector. If the options are  /* This function is called by check_auto_possessive() when a property item
2525  changed during the branch, the pointer is used to change the external options  is adjacent to a fixed character.
 bits.  
2526    
2527  Arguments:  Arguments:
2528    optionsptr     pointer to the option bits    c            the character
2529    brackets       points to number of extracting brackets used    ptype        the property type
2530    codeptr        points to the pointer to the current code point    pdata        the data for the type
2531    ptrptr         points to the current pattern pointer    negated      TRUE if it's a negated property (\P or \p{^)
   errorcodeptr   points to error code variable  
   firstbyteptr   set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)  
   reqbyteptr     set to the last literal character required, else < 0  
   bcptr          points to current branch chain  
   cd             contains pointers to tables etc.  
2532    
2533  Returns:         TRUE on success  Returns:       TRUE if auto-possessifying is OK
                  FALSE, with *errorcodeptr set non-zero on error  
2534  */  */
2535    
2536  static BOOL  static BOOL
2537  compile_branch(int *optionsptr, int *brackets, uschar **codeptr,  check_char_prop(int c, int ptype, int pdata, BOOL negated)
   const uschar **ptrptr, int *errorcodeptr, int *firstbyteptr,  
   int *reqbyteptr, branch_chain *bcptr, compile_data *cd)  
2538  {  {
2539  int repeat_type, op_type;  const ucd_record *prop = GET_UCD(c);
2540  int repeat_min = 0, repeat_max = 0;      /* To please picky compilers */  switch(ptype)
2541  int bravalue = 0;    {
2542  int greedy_default, greedy_non_default;    case PT_LAMP:
2543  int firstbyte, reqbyte;    return (prop->chartype == ucp_Lu ||
2544  int zeroreqbyte, zerofirstbyte;            prop->chartype == ucp_Ll ||
2545  int req_caseopt, reqvary, tempreqvary;            prop->chartype == ucp_Lt) == negated;
2546  int condcount = 0;  
2547  int options = *optionsptr;    case PT_GC:
2548  int after_manual_callout = 0;    return (pdata == _pcre_ucp_gentype[prop->chartype]) == negated;
2549  register int c;  
2550  register uschar *code = *codeptr;    case PT_PC:
2551  uschar *tempcode;    return (pdata == prop->chartype) == negated;
2552  BOOL inescq = FALSE;  
2553  BOOL groupsetfirstbyte = FALSE;    case PT_SC:
2554  const uschar *ptr = *ptrptr;    return (pdata == prop->script) == negated;
2555  const uschar *tempptr;  
2556  uschar *previous = NULL;    /* These are specials */
2557  uschar *previous_callout = NULL;  
2558  uschar classbits[32];    case PT_ALNUM:
2559      return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2560  #ifdef SUPPORT_UTF8            _pcre_ucp_gentype[prop->chartype] == ucp_N) == negated;
2561  BOOL class_utf8;  
2562  BOOL utf8 = (options & PCRE_UTF8) != 0;    case PT_SPACE:    /* Perl space */
2563  uschar *class_utf8data;    return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2564  uschar utf8_char[6];            c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2565  #else            == negated;
2566  BOOL utf8 = FALSE;  
2567  #endif    case PT_PXSPACE:  /* POSIX space */
2568      return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2569              c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2570              c == CHAR_FF || c == CHAR_CR)
2571              == negated;
2572    
2573      case PT_WORD:
2574      return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2575              _pcre_ucp_gentype[prop->chartype] == ucp_N ||
2576              c == CHAR_UNDERSCORE) == negated;
2577      }
2578    return FALSE;
2579    }
2580    #endif  /* SUPPORT_UCP */
2581    
 /* Set up the default and non-default settings for greediness */  
2582    
 greedy_default = ((options & PCRE_UNGREEDY) != 0);  
 greedy_non_default = greedy_default ^ 1;  
2583    
2584  /* Initialize no first byte, no required byte. REQ_UNSET means "no char  /*************************************************
2585  matching encountered yet". It gets changed to REQ_NONE if we hit something that  *     Check if auto-possessifying is possible    *
2586  matches a non-fixed char first char; reqbyte just remains unset if we never  *************************************************/
 find one.  
2587    
2588  When we hit a repeat whose minimum is zero, we may have to adjust these values  /* This function is called for unlimited repeats of certain items, to see
2589  to take the zero repeat into account. This is implemented by setting them to  whether the next thing could possibly match the repeated item. If not, it makes
2590  zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual  sense to automatically possessify the repeated item.
 item types that can be repeated set these backoff variables appropriately. */  
2591    
2592  firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;  Arguments:
2593      previous      pointer to the repeated opcode
2594      utf8          TRUE in UTF-8 mode
2595      ptr           next character in pattern
2596      options       options bits
2597      cd            contains pointers to tables etc.
2598    
2599  /* The variable req_caseopt contains either the REQ_CASELESS value or zero,  Returns:        TRUE if possessifying is wanted
2600  according to the current setting of the caseless flag. REQ_CASELESS is a bit  */
 value > 255. It is added into the firstbyte or reqbyte variables to record the  
 case status of the value. This is used only for ASCII characters. */  
2601    
2602  req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;  static BOOL
2603    check_auto_possessive(const uschar *previous, BOOL utf8, const uschar *ptr,
2604      int options, compile_data *cd)
2605    {
2606    int c, next;
2607    int op_code = *previous++;
2608    
2609  /* Switch on next character until the end of the branch */  /* Skip whitespace and comments in extended mode */
2610    
2611  for (;; ptr++)  if ((options & PCRE_EXTENDED) != 0)
2612    {    {
2613    BOOL negate_class;    for (;;)
   BOOL possessive_quantifier;  
   BOOL is_quantifier;  
   int class_charcount;  
   int class_lastchar;  
   int newoptions;  
   int recno;  
   int skipbytes;  
   int subreqbyte;  
   int subfirstbyte;  
   int mclength;  
   uschar mcbuffer[8];  
   
   /* Next byte in the pattern */  
   
   c = *ptr;  
   
   /* If in \Q...\E, check for the end; if not, we have a literal */  
   
   if (inescq && c != 0)  
2614      {      {
2615      if (c == '\\' && ptr[1] == 'E')      while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2616        if (*ptr == CHAR_NUMBER_SIGN)
2617        {        {
       inescq = FALSE;  
2618        ptr++;        ptr++;
2619        continue;        while (*ptr != 0)
       }  
     else  
       {  
       if (previous_callout != NULL)  
         {  
         complete_callout(previous_callout, ptr, cd);  
         previous_callout = NULL;  
         }  
       if ((options & PCRE_AUTO_CALLOUT) != 0)  
2620          {          {
2621          previous_callout = code;          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2622          code = auto_callout(code, ptr, cd);          ptr++;
2623    #ifdef SUPPORT_UTF8
2624            if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2625    #endif
2626          }          }
       goto NORMAL_CHAR;  
2627        }        }
2628        else break;
2629      }      }
2630      }
2631    
2632    /* Fill in length of a previous callout, except when the next thing is  /* If the next item is one that we can handle, get its value. A non-negative
2633    a quantifier. */  value is a character, a negative value is an escape value. */
2634    
2635    is_quantifier = c == '*' || c == '+' || c == '?' ||  if (*ptr == CHAR_BACKSLASH)
2636      (c == '{' && is_counted_repeat(ptr+1));    {
2637      int temperrorcode = 0;
2638      next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2639      if (temperrorcode != 0) return FALSE;
2640      ptr++;    /* Point after the escape sequence */
2641      }
2642    
2643    if (!is_quantifier && previous_callout != NULL &&  else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2644         after_manual_callout-- <= 0)    {
2645      {  #ifdef SUPPORT_UTF8
2646      complete_callout(previous_callout, ptr, cd);    if (utf8) { GETCHARINC(next, ptr); } else
2647      previous_callout = NULL;  #endif
2648      }    next = *ptr++;
2649      }
2650    
2651    /* In extended mode, skip white space and comments */  else return FALSE;
2652    
2653    if ((options & PCRE_EXTENDED) != 0)  /* Skip whitespace and comments in extended mode */
2654    
2655    if ((options & PCRE_EXTENDED) != 0)
2656      {
2657      for (;;)
2658      {      {
2659      if ((cd->ctypes[c] & ctype_space) != 0) continue;      while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2660      if (c == '#')      if (*ptr == CHAR_NUMBER_SIGN)
2661        {        {
2662        /* The space before the ; is to avoid a warning on a silly compiler        ptr++;
2663        on the Macintosh. */        while (*ptr != 0)
2664        while ((c = *(++ptr)) != 0 && c != NEWLINE) ;          {
2665        if (c != 0) continue;   /* Else fall through to handle end of string */          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2666            ptr++;
2667    #ifdef SUPPORT_UTF8
2668            if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2669    #endif
2670            }
2671        }        }
2672        else break;
2673      }      }
2674      }
2675    
2676    /* No auto callout for quantifiers. */  /* If the next thing is itself optional, we have to give up. */
2677    
2678    if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)  if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2679      {    strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2680      previous_callout = code;      return FALSE;
     code = auto_callout(code, ptr, cd);  
     }  
2681    
2682    switch(c)  /* Now compare the next item with the previous opcode. First, handle cases when
2683      {  the next item is a character. */
     /* The branch terminates at end of string, |, or ). */  
2684    
2685      case 0:  if (next >= 0) switch(op_code)
2686      case '|':    {
2687      case ')':    case OP_CHAR:
2688      *firstbyteptr = firstbyte;  #ifdef SUPPORT_UTF8
2689      *reqbyteptr = reqbyte;    GETCHARTEST(c, previous);
2690      *codeptr = code;  #else
2691      *ptrptr = ptr;    c = *previous;
2692      return TRUE;  #endif
2693      return c != next;
2694    
2695      /* Handle single-character metacharacters. In multiline mode, ^ disables    /* For CHARI (caseless character) we must check the other case. If we have
2696      the setting of any following char as a first character. */    Unicode property support, we can use it to test the other case of
2697      high-valued characters. */
2698    
2699      case '^':    case OP_CHARI:
2700      if ((options & PCRE_MULTILINE) != 0)  #ifdef SUPPORT_UTF8
2701        {    GETCHARTEST(c, previous);
2702        if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;  #else
2703        }    c = *previous;
2704      previous = NULL;  #endif
2705      *code++ = OP_CIRC;    if (c == next) return FALSE;
2706      break;  #ifdef SUPPORT_UTF8
2707      if (utf8)
2708        {
2709        unsigned int othercase;
2710        if (next < 128) othercase = cd->fcc[next]; else
2711    #ifdef SUPPORT_UCP
2712        othercase = UCD_OTHERCASE((unsigned int)next);
2713    #else
2714        othercase = NOTACHAR;
2715    #endif
2716        return (unsigned int)c != othercase;
2717        }
2718      else
2719    #endif  /* SUPPORT_UTF8 */
2720      return (c != cd->fcc[next]);  /* Non-UTF-8 mode */
2721    
2722      case '$':    /* For OP_NOT and OP_NOTI, the data is always a single-byte character. These
2723      previous = NULL;    opcodes are not used for multi-byte characters, because they are coded using
2724      *code++ = OP_DOLL;    an XCLASS instead. */
     break;  
2725    
2726      /* There can never be a first char if '.' is first, whatever happens about    case OP_NOT:
2727      repeats. The value of reqbyte doesn't change either. */    return (c = *previous) == next;
2728    
2729      case '.':    case OP_NOTI:
2730      if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;    if ((c = *previous) == next) return TRUE;
2731      zerofirstbyte = firstbyte;  #ifdef SUPPORT_UTF8
2732      zeroreqbyte = reqbyte;    if (utf8)
2733      previous = code;      {
2734      *code++ = OP_ANY;      unsigned int othercase;
2735      break;      if (next < 128) othercase = cd->fcc[next]; else
2736    #ifdef SUPPORT_UCP
2737        othercase = UCD_OTHERCASE(next);
2738    #else
2739        othercase = NOTACHAR;
2740    #endif
2741        return (unsigned int)c == othercase;
2742        }
2743      else
2744    #endif  /* SUPPORT_UTF8 */
2745      return (c == cd->fcc[next]);  /* Non-UTF-8 mode */
2746    
2747      /* Character classes. If the included characters are all < 255 in value, we    /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
2748      build a 32-byte bitmap of the permitted characters, except in the special    When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
     case where there is only one such character. For negated classes, we build  
     the map as usual, then invert it at the end. However, we use a different  
     opcode so that data characters > 255 can be handled correctly.  
2749    
2750      If the class contains characters outside the 0-255 range, a different    case OP_DIGIT:
2751      opcode is compiled. It may optionally have a bit map for characters < 256,    return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
     but those above are are explicitly listed afterwards. A flag byte tells  
     whether the bitmap is present, and whether this is a negated class or not.  
     */  
2752    
2753      case '[':    case OP_NOT_DIGIT:
2754      previous = code;    return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2755    
2756      /* PCRE supports POSIX class stuff inside a class. Perl gives an error if    case OP_WHITESPACE:
2757      they are encountered at the top level, so we'll do that too. */    return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2758    
2759      if ((ptr[1] == ':' || ptr[1] == '.' || ptr[1] == '=') &&    case OP_NOT_WHITESPACE:
2760          check_posix_syntax(ptr, &tempptr, cd))    return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
       {  
       *errorcodeptr = (ptr[1] == ':')? ERR13 : ERR31;  
       goto FAILED;  
       }  
2761    
2762      /* If the first character is '^', set the negation flag and skip it. */    case OP_WORDCHAR:
2763      return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2764    
2765      if ((c = *(++ptr)) == '^')    case OP_NOT_WORDCHAR:
2766        {    return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
       negate_class = TRUE;  
       c = *(++ptr);  
       }  
     else  
       {  
       negate_class = FALSE;  
       }  
2767    
2768      /* Keep a count of chars with values < 256 so that we can optimize the case    case OP_HSPACE:
2769      of just a single character (as long as it's < 256). For higher valued UTF-8    case OP_NOT_HSPACE:
2770      characters, we don't yet do any optimization. */    switch(next)
2771        {
2772        case 0x09:
2773        case 0x20:
2774        case 0xa0:
2775        case 0x1680:
2776        case 0x180e:
2777        case 0x2000:
2778        case 0x2001:
2779        case 0x2002:
2780        case 0x2003:
2781        case 0x2004:
2782        case 0x2005:
2783        case 0x2006:
2784        case 0x2007:
2785        case 0x2008:
2786        case 0x2009:
2787        case 0x200A:
2788        case 0x202f:
2789        case 0x205f:
2790        case 0x3000:
2791        return op_code == OP_NOT_HSPACE;
2792        default:
2793        return op_code != OP_NOT_HSPACE;
2794        }
2795    
2796      class_charcount = 0;    case OP_ANYNL:
2797      class_lastchar = -1;    case OP_VSPACE:
2798      case OP_NOT_VSPACE:
2799      switch(next)
2800        {
2801        case 0x0a:
2802        case 0x0b:
2803        case 0x0c:
2804        case 0x0d:
2805        case 0x85:
2806        case 0x2028:
2807        case 0x2029:
2808        return op_code == OP_NOT_VSPACE;
2809        default:
2810        return op_code != OP_NOT_VSPACE;
2811        }
2812    
2813  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UCP
2814      class_utf8 = FALSE;                       /* No chars >= 256 */    case OP_PROP:
2815      class_utf8data = code + LINK_SIZE + 34;   /* For UTF-8 items */    return check_char_prop(next, previous[0], previous[1], FALSE);
2816    
2817      case OP_NOTPROP:
2818      return check_char_prop(next, previous[0], previous[1], TRUE);
2819  #endif  #endif
2820    
2821      /* Initialize the 32-char bit map to all zeros. We have to build the    default:
2822      map in a temporary bit of store, in case the class contains only 1    return FALSE;
2823      character (< 256), because in that case the compiled code doesn't use the    }
     bit map. */  
2824    
     memset(classbits, 0, 32 * sizeof(uschar));  
2825    
2826      /* Process characters until ] is reached. By writing this as a "do" it  /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
2827      means that an initial ] is taken as a data character. The first pass  is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
2828      through the regex checked the overall syntax, so we don't need to be very  generated only when PCRE_UCP is *not* set, that is, when only ASCII
2829      strict here. At the start of the loop, c contains the first byte of the  characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
2830      character. */  replaced by OP_PROP codes when PCRE_UCP is set. */
2831    
2832      do  switch(op_code)
2833        {    {
2834      case OP_CHAR:
2835      case OP_CHARI:
2836  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
2837        if (utf8 && c > 127)    GETCHARTEST(c, previous);
2838          {                           /* Braces are required because the */  #else
2839          GETCHARLEN(c, ptr, ptr);    /* macro generates multiple statements */    c = *previous;
         }  
2840  #endif  #endif
2841      switch(-next)
2842        {
2843        case ESC_d:
2844        return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
2845    
2846        /* Inside \Q...\E everything is literal except \E */      case ESC_D:
2847        return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
2848    
2849        if (inescq)      case ESC_s:
2850          {      return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
         if (c == '\\' && ptr[1] == 'E')  
           {  
           inescq = FALSE;  
           ptr++;  
           continue;  
           }  
         else goto LONE_SINGLE_CHARACTER;  
         }  
2851    
2852        /* Handle POSIX class names. Perl allows a negation extension of the      case ESC_S:
2853        form [:^name:]. A square bracket that doesn't match the syntax is      return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
2854        treated as a literal. We also recognize the POSIX constructions  
2855        [.ch.] and [=ch=] ("collating elements") and fault them, as Perl      case ESC_w:
2856        5.6 and 5.8 do. */      return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
2857    
2858        case ESC_W:
2859        return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
2860    
2861        case ESC_h:
2862        case ESC_H:
2863        switch(c)
2864          {
2865          case 0x09:
2866          case 0x20:
2867          case 0xa0:
2868          case 0x1680:
2869          case 0x180e:
2870          case 0x2000:
2871          case 0x2001:
2872          case 0x2002:
2873          case 0x2003:
2874          case 0x2004:
2875          case 0x2005:
2876          case 0x2006:
2877          case 0x2007:
2878          case 0x2008:
2879          case 0x2009:
2880          case 0x200A:
2881          case 0x202f:
2882          case 0x205f:
2883          case 0x3000:
2884          return -next != ESC_h;
2885          default:
2886          return -next == ESC_h;
2887          }
2888    
2889        if (c == '[' &&      case ESC_v:
2890            (ptr[1] == ':' || ptr[1] == '.' || ptr[1] == '=') &&      case ESC_V:
2891            check_posix_syntax(ptr, &tempptr, cd))      switch(c)
2892          {        {
2893          BOOL local_negate = FALSE;        case 0x0a:
2894          int posix_class, i;        case 0x0b:
2895          register const uschar *cbits = cd->cbits;        case 0x0c:
2896          case 0x0d:
2897          case 0x85:
2898          case 0x2028:
2899          case 0x2029:
2900          return -next != ESC_v;
2901          default:
2902          return -next == ESC_v;
2903          }
2904    
2905          if (ptr[1] != ':')      /* When PCRE_UCP is set, these values get generated for \d etc. Find
2906            {      their substitutions and process them. The result will always be either
2907            *errorcodeptr = ERR31;      -ESC_p or -ESC_P. Then fall through to process those values. */
           goto FAILED;  
           }  
2908    
2909          ptr += 2;  #ifdef SUPPORT_UCP
2910          if (*ptr == '^')      case ESC_du:
2911            {      case ESC_DU:
2912            local_negate = TRUE;      case ESC_wu:
2913            ptr++;      case ESC_WU:
2914            }      case ESC_su:
2915        case ESC_SU:
2916          {
2917          int temperrorcode = 0;
2918          ptr = substitutes[-next - ESC_DU];
2919          next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
2920          if (temperrorcode != 0) return FALSE;
2921          ptr++;    /* For compatibility */
2922          }
2923        /* Fall through */
2924    
2925          posix_class = check_posix_name(ptr, tempptr - ptr);      case ESC_p:
2926          if (posix_class < 0)      case ESC_P:
2927            {        {
2928            *errorcodeptr = ERR30;        int ptype, pdata, errorcodeptr;
2929            goto FAILED;        BOOL negated;
           }  
2930    
2931          /* If matching is caseless, upper and lower are converted to        ptr--;      /* Make ptr point at the p or P */
2932          alpha. This relies on the fact that the class table starts with        ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
2933          alpha, lower, upper as the first 3 entries. */        if (ptype < 0) return FALSE;
2934          ptr++;      /* Point past the final curly ket */
2935    
2936          /* If the property item is optional, we have to give up. (When generated
2937          from \d etc by PCRE_UCP, this test will have been applied much earlier,
2938          to the original \d etc. At this point, ptr will point to a zero byte. */
2939    
2940          if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2941            strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2942              return FALSE;
2943    
2944          if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)        /* Do the property check. */
           posix_class = 0;  
2945    
2946          /* Or into the map we are building up to 3 of the static class        return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
2947          tables, or their negations. The [:blank:] class sets up the same        }
2948          chars as the [:space:] class (all white space). We remove the vertical  #endif
         white space chars afterwards. */  
2949    
2950          posix_class *= 3;      default:
2951          for (i = 0; i < 3; i++)      return FALSE;
2952            {      }
           BOOL blankclass = strncmp((char *)ptr, "blank", 5) == 0;  
           int taboffset = posix_class_maps[posix_class + i];  
           if (taboffset < 0) break;  
           if (local_negate)  
             {  
             if (i == 0)  
               for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+taboffset];  
             else  
               for (c = 0; c < 32; c++) classbits[c] &= ~cbits[c+taboffset];  
             if (blankclass) classbits[1] |= 0x3c;  
             }  
           else  
             {  
             for (c = 0; c < 32; c++) classbits[c] |= cbits[c+taboffset];  
             if (blankclass) classbits[1] &= ~0x3c;  
             }  
           }  
2953    
2954          ptr = tempptr + 1;    /* In principle, support for Unicode properties should be integrated here as
2955          class_charcount = 10;  /* Set > 1; assumes more than 1 per class */    well. It means re-organizing the above code so as to get hold of the property
2956          continue;    /* End of POSIX syntax handling */    values before switching on the op-code. However, I wonder how many patterns
2957          }    combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
2958      these op-codes are never generated.) */
2959    
2960        /* Backslash may introduce a single character, or it may introduce one    case OP_DIGIT:
2961        of the specials, which just set a flag. Escaped items are checked for    return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
2962        validity in the pre-compiling pass. The sequence \b is a special case.           next == -ESC_h || next == -ESC_v || next == -ESC_R;
       Inside a class (and only there) it is treated as backspace. Elsewhere  
       it marks a word boundary. Other escapes have preset maps ready to  
       or into the one we are building. We assume they have more than one  
       character in them, so set class_charcount bigger than one. */  
2963    
2964        if (c == '\\')    case OP_NOT_DIGIT:
2965          {    return next == -ESC_d;
         c = check_escape(&ptr, errorcodeptr, *brackets, options, TRUE);  
2966    
2967          if (-c == ESC_b) c = '\b';       /* \b is backslash in a class */    case OP_WHITESPACE:
2968          else if (-c == ESC_X) c = 'X';   /* \X is literal X in a class */    return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
         else if (-c == ESC_Q)            /* Handle start of quoted string */  
           {  
           if (ptr[1] == '\\' && ptr[2] == 'E')  
             {  
             ptr += 2; /* avoid empty string */  
             }  
           else inescq = TRUE;  
           continue;  
           }  
2969    
2970          if (c < 0)    case OP_NOT_WHITESPACE:
2971            {    return next == -ESC_s || next == -ESC_h || next == -ESC_v;
           register const uschar *cbits = cd->cbits;  
           class_charcount += 2;     /* Greater than 1 is what matters */  
           switch (-c)  
             {  
             case ESC_d:  
             for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];  
             continue;  
2972    
2973              case ESC_D:    case OP_HSPACE:
2974              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];    return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
2975              continue;           next == -ESC_w || next == -ESC_v || next == -ESC_R;
2976    
2977              case ESC_w:    case OP_NOT_HSPACE:
2978              for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];    return next == -ESC_h;
             continue;  
2979    
2980              case ESC_W:    /* Can't have \S in here because VT matches \S (Perl anomaly) */
2981              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];    case OP_ANYNL:
2982              continue;    case OP_VSPACE:
2983      return next == -ESC_V || next == -ESC_d || next == -ESC_w;
2984    
2985              case ESC_s:    case OP_NOT_VSPACE:
2986              for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];    return next == -ESC_v || next == -ESC_R;
             classbits[1] &= ~0x08;   /* Perl 5.004 onwards omits VT from \s */  
             continue;  
2987    
2988              case ESC_S:    case OP_WORDCHAR:
2989              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];    return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
2990              classbits[1] |= 0x08;    /* Perl 5.004 onwards omits VT from \s */           next == -ESC_v || next == -ESC_R;
             continue;  
2991    
2992  #ifdef SUPPORT_UCP    case OP_NOT_WORDCHAR:
2993              case ESC_p:    return next == -ESC_w || next == -ESC_d;
             case ESC_P:  
               {  
               BOOL negated;  
               int property = get_ucp(&ptr, &negated, errorcodeptr);  
               if (property < 0) goto FAILED;  
               class_utf8 = TRUE;  
               *class_utf8data++ = ((-c == ESC_p) != negated)?  
                 XCL_PROP : XCL_NOTPROP;  
               *class_utf8data++ = property;  
               class_charcount -= 2;   /* Not a < 256 character */  
               }  
             continue;  
 #endif  
2994    
2995              /* Unrecognized escapes are faulted if PCRE is running in its    default:
2996              strict mode. By default, for compatibility with Perl, they are    return FALSE;
2997              treated as literals. */    }
2998    
2999              default:  /* Control does not reach here */
3000              if ((options & PCRE_EXTRA) != 0)  }
               {  
               *errorcodeptr = ERR7;  
               goto FAILED;  
               }  
             c = *ptr;              /* The final character */  
             class_charcount -= 2;  /* Undo the default count from above */  
             }  
           }  
3001    
         /* Fall through if we have a single character (c >= 0). This may be  
         > 256 in UTF-8 mode. */  
3002    
         }   /* End of backslash handling */  
3003    
3004        /* A single character may be followed by '-' to form a range. However,  /*************************************************
3005        Perl does not permit ']' to be the end of the range. A '-' character  *           Compile one branch                   *
3006        here is treated as a literal. */  *************************************************/
3007    
3008        if (ptr[1] == '-' && ptr[2] != ']')  /* Scan the pattern, compiling it into the a vector. If the options are
3009          {  changed during the branch, the pointer is used to change the external options
3010          int d;  bits. This function is used during the pre-compile phase when we are trying
3011          ptr += 2;  to find out the amount of memory needed, as well as during the real compile
3012    phase. The value of lengthptr distinguishes the two phases.
3013    
3014    Arguments:
3015      optionsptr     pointer to the option bits
3016      codeptr        points to the pointer to the current code point
3017      ptrptr         points to the current pattern pointer
3018      errorcodeptr   points to error code variable
3019      firstbyteptr   set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3020      reqbyteptr     set to the last literal character required, else < 0
3021      bcptr          points to current branch chain
3022      cd             contains pointers to tables etc.
3023      lengthptr      NULL during the real compile phase
3024                     points to length accumulator during pre-compile phase
3025    
3026    Returns:         TRUE on success
3027                     FALSE, with *errorcodeptr set non-zero on error
3028    */
3029    
3030    static BOOL
3031    compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
3032      int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
3033      compile_data *cd, int *lengthptr)
3034    {
3035    int repeat_type, op_type;
3036    int repeat_min = 0, repeat_max = 0;      /* To please picky compilers */
3037    int bravalue = 0;
3038    int greedy_default, greedy_non_default;
3039    int firstbyte, reqbyte;
3040    int zeroreqbyte, zerofirstbyte;
3041    int req_caseopt, reqvary, tempreqvary;
3042    int options = *optionsptr;               /* May change dynamically */
3043    int after_manual_callout = 0;
3044    int length_prevgroup = 0;
3045    register int c;
3046    register uschar *code = *codeptr;
3047    uschar *last_code = code;
3048    uschar *orig_code = code;
3049    uschar *tempcode;
3050    BOOL inescq = FALSE;
3051    BOOL groupsetfirstbyte = FALSE;
3052    const uschar *ptr = *ptrptr;
3053    const uschar *tempptr;
3054    const uschar *nestptr = NULL;
3055    uschar *previous = NULL;
3056    uschar *previous_callout = NULL;
3057    uschar *save_hwm = NULL;
3058    uschar classbits[32];
3059    
3060    /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3061    must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3062    dynamically as we process the pattern. */
3063    
3064  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
3065          if (utf8)  BOOL class_utf8;
3066            {                           /* Braces are required because the */  BOOL utf8 = (options & PCRE_UTF8) != 0;
3067            GETCHARLEN(d, ptr, ptr);    /* macro generates multiple statements */  uschar *class_utf8data;
3068            }  uschar *class_utf8data_base;
3069          else  uschar utf8_char[6];
3070    #else
3071    BOOL utf8 = FALSE;
3072    uschar *utf8_char = NULL;
3073  #endif  #endif
         d = *ptr;  /* Not UTF-8 mode */  
3074    
3075          /* The second part of a range can be a single-character escape, but  #ifdef PCRE_DEBUG
3076          not any of the other escapes. Perl 5.6 treats a hyphen as a literal  if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3077          in such circumstances. */  #endif
3078    
3079          if (d == '\\')  /* Set up the default and non-default settings for greediness */
           {  
           const uschar *oldptr = ptr;  
           d = check_escape(&ptr, errorcodeptr, *brackets, options, TRUE);  
3080    
3081            /* \b is backslash; \X is literal X; any other special means the '-'  greedy_default = ((options & PCRE_UNGREEDY) != 0);
3082            was literal */  greedy_non_default = greedy_default ^ 1;
3083    
3084            if (d < 0)  /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3085              {  matching encountered yet". It gets changed to REQ_NONE if we hit something that
3086              if (d == -ESC_b) d = '\b';  matches a non-fixed char first char; reqbyte just remains unset if we never
3087              else if (d == -ESC_X) d = 'X'; else  find one.
               {  
               ptr = oldptr - 2;  
               goto LONE_SINGLE_CHARACTER;  /* A few lines below */  
               }  
             }  
           }  
3088    
3089          /* The check that the two values are in the correct order happens in  When we hit a repeat whose minimum is zero, we may have to adjust these values
3090          the pre-pass. Optimize one-character ranges */  to take the zero repeat into account. This is implemented by setting them to
3091    zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
3092    item types that can be repeated set these backoff variables appropriately. */
3093    
3094          if (d == c) goto LONE_SINGLE_CHARACTER;  /* A few lines below */  firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
3095    
3096          /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless  /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
3097          matching, we have to use an XCLASS with extra data items. Caseless  according to the current setting of the caseless flag. REQ_CASELESS is a bit
3098          matching for characters > 127 is available only if UCP support is  value > 255. It is added into the firstbyte or reqbyte variables to record the
3099          available. */  case status of the value. This is used only for ASCII characters. */
3100    
3101  #ifdef SUPPORT_UTF8  req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
         if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))  
           {  
           class_utf8 = TRUE;  
3102    
3103            /* With UCP support, we can find the other case equivalents of  /* Switch on next character until the end of the branch */
           the relevant characters. There may be several ranges. Optimize how  
           they fit with the basic range. */  
3104    
3105  #ifdef SUPPORT_UCP  for (;; ptr++)
3106            if ((options & PCRE_CASELESS) != 0)    {
3107              {    BOOL negate_class;
3108              int occ, ocd;    BOOL should_flip_negation;
3109              int cc = c;    BOOL possessive_quantifier;
3110              int origd = d;    BOOL is_quantifier;
3111              while (get_othercase_range(&cc, origd, &occ, &ocd))    BOOL is_recurse;
3112                {    BOOL reset_bracount;
3113                if (occ >= c && ocd <= d) continue;  /* Skip embedded ranges */    int class_charcount;
3114      int class_lastchar;
3115      int newoptions;
3116      int recno;
3117      int refsign;
3118      int skipbytes;
3119      int subreqbyte;
3120      int subfirstbyte;
3121      int terminator;
3122      int mclength;
3123      uschar mcbuffer[8];
3124    
3125                if (occ < c  && ocd >= c - 1)        /* Extend the basic range */    /* Get next byte in the pattern */
                 {                                  /* if there is overlap,   */  
                 c = occ;                           /* noting that if occ < c */  
                 continue;                          /* we can't have ocd > d  */  
                 }                                  /* because a subrange is  */  
               if (ocd > d && occ <= d + 1)         /* always shorter than    */  
                 {                                  /* the basic range.       */  
                 d = ocd;  
                 continue;  
                 }  
3126    
3127                if (occ == ocd)    c = *ptr;
                 {  
                 *class_utf8data++ = XCL_SINGLE;  
                 }  
               else  
                 {  
                 *class_utf8data++ = XCL_RANGE;  
                 class_utf8data += _pcre_ord2utf8(occ, class_utf8data);  
                 }  
               class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);  
               }  
             }  
 #endif  /* SUPPORT_UCP */  
3128    
3129            /* Now record the original range, possibly modified for UCP caseless    /* If we are at the end of a nested substitution, revert to the outer level
3130            overlapping ranges. */    string. Nesting only happens one level deep. */
3131    
3132            *class_utf8data++ = XCL_RANGE;    if (c == 0 && nestptr != NULL)
3133            class_utf8data += _pcre_ord2utf8(c, class_utf8data);      {
3134            class_utf8data += _pcre_ord2utf8(d, class_utf8data);      ptr = nestptr;
3135        nestptr = NULL;
3136        c = *ptr;
3137        }
3138    
3139            /* With UCP support, we are done. Without UCP support, there is no    /* If we are in the pre-compile phase, accumulate the length used for the
3140            caseless matching for UTF-8 characters > 127; we can use the bit map    previous cycle of this loop. */
           for the smaller ones. */  
3141    
3142  #ifdef SUPPORT_UCP    if (lengthptr != NULL)
3143            continue;    /* With next character in the class */      {
3144  #else  #ifdef PCRE_DEBUG
3145            if ((options & PCRE_CASELESS) == 0 || c > 127) continue;      if (code > cd->hwm) cd->hwm = code;                 /* High water info */
3146    #endif
3147        if (code > cd->start_workspace + WORK_SIZE_CHECK)   /* Check for overrun */
3148          {
3149          *errorcodeptr = ERR52;
3150          goto FAILED;
3151          }
3152    
3153            /* Adjust upper limit and fall through to set up the map */      /* There is at least one situation where code goes backwards: this is the
3154        case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3155        the class is simply eliminated. However, it is created first, so we have to
3156        allow memory for it. Therefore, don't ever reduce the length at this point.
3157        */
3158    
3159            d = 127;      if (code < last_code) code = last_code;
3160    
3161  #endif  /* SUPPORT_UCP */      /* Paranoid check for integer overflow */
           }  
 #endif  /* SUPPORT_UTF8 */  
3162    
3163          /* We use the bit map for all cases when not in UTF-8 mode; else      if (OFLOW_MAX - *lengthptr < code - last_code)
3164          ranges that lie entirely within 0-127 when there is UCP support; else        {
3165          for partial ranges without UCP support. */        *errorcodeptr = ERR20;
3166          goto FAILED;
3167          }
3168    
3169          for (; c <= d; c++)      *lengthptr += (int)(code - last_code);
3170            {      DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
           classbits[c/8] |= (1 << (c&7));  
           if ((options & PCRE_CASELESS) != 0)  
             {  
             int uc = cd->fcc[c];           /* flip case */  
             classbits[uc/8] |= (1 << (uc&7));  
             }  
           class_charcount++;                /* in case a one-char range */  
           class_lastchar = c;  
           }  
3171    
3172          continue;   /* Go get the next char in the class */      /* If "previous" is set and it is not at the start of the work space, move
3173        it back to there, in order to avoid filling up the work space. Otherwise,
3174        if "previous" is NULL, reset the current code pointer to the start. */
3175    
3176        if (previous != NULL)
3177          {
3178          if (previous > orig_code)
3179            {
3180            memmove(orig_code, previous, code - previous);
3181            code -= previous - orig_code;
3182            previous = orig_code;
3183          }          }
3184          }
3185        else code = orig_code;
3186    
3187        /* Handle a lone single character - we can get here for a normal      /* Remember where this code item starts so we can pick up the length
3188        non-escape char, or after \ that introduces a single character or for an      next time round. */
       apparent range that isn't. */  
3189    
3190        LONE_SINGLE_CHARACTER:      last_code = code;
3191        }
3192    
3193        /* Handle a character that cannot go in the bit map */    /* In the real compile phase, just check the workspace used by the forward
3194      reference list. */
3195    
3196  #ifdef SUPPORT_UTF8    else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
3197        if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))      {
3198          {      *errorcodeptr = ERR52;
3199          class_utf8 = TRUE;      goto FAILED;
3200          *class_utf8data++ = XCL_SINGLE;      }
         class_utf8data += _pcre_ord2utf8(c, class_utf8data);  
3201    
3202  #ifdef SUPPORT_UCP    /* If in \Q...\E, check for the end; if not, we have a literal */
         if ((options & PCRE_CASELESS) != 0)  
           {  
           int chartype;  
           int othercase;  
           if (_pcre_ucp_findchar(c, &chartype, &othercase) >= 0 &&  
                othercase > 0)  
             {  
             *class_utf8data++ = XCL_SINGLE;  
             class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);  
             }  
           }  
 #endif  /* SUPPORT_UCP */  
3203    
3204      if (inescq && c != 0)
3205        {
3206        if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3207          {
3208          inescq = FALSE;
3209          ptr++;
3210          continue;
3211          }
3212        else
3213          {
3214          if (previous_callout != NULL)
3215            {
3216            if (lengthptr == NULL)  /* Don't attempt in pre-compile phase */
3217              complete_callout(previous_callout, ptr, cd);
3218            previous_callout = NULL;
3219          }          }
3220        else        if ((options & PCRE_AUTO_CALLOUT) != 0)
 #endif  /* SUPPORT_UTF8 */  
   
       /* Handle a single-byte character */  
3221          {          {
3222          classbits[c/8] |= (1 << (c&7));          previous_callout = code;
3223          if ((options & PCRE_CASELESS) != 0)          code = auto_callout(code, ptr, cd);
           {  
           c = cd->fcc[c];   /* flip case */  
           classbits[c/8] |= (1 << (c&7));  
           }  
         class_charcount++;  
         class_lastchar = c;  
3224          }          }
3225          goto NORMAL_CHAR;
3226        }        }
3227        }
3228    
3229      /* Loop until ']' reached; the check for end of string happens inside the    /* Fill in length of a previous callout, except when the next thing is
3230      loop. This "while" is the end of the "do" above. */    a quantifier. */
   
     while ((c = *(++ptr)) != ']' || inescq);  
3231    
3232      /* If class_charcount is 1, we saw precisely one character whose value is    is_quantifier =
3233      less than 256. In non-UTF-8 mode we can always optimize. In UTF-8 mode, we      c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3234      can optimize the negative case only if there were no characters >= 128      (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
     because OP_NOT and the related opcodes like OP_NOTSTAR operate on  
     single-bytes only. This is an historical hangover. Maybe one day we can  
     tidy these opcodes to handle multi-byte characters.  
3235    
3236      The optimization throws away the bit map. We turn the item into a    if (!is_quantifier && previous_callout != NULL &&
3237      1-character OP_CHAR[NC] if it's positive, or OP_NOT if it's negative. Note         after_manual_callout-- <= 0)
3238      that OP_NOT does not support multibyte characters. In the positive case, it      {
3239      can cause firstbyte to be set. Otherwise, there can be no first char if      if (lengthptr == NULL)      /* Don't attempt in pre-compile phase */
3240      this item is first, whatever repeat count may follow. In the case of        complete_callout(previous_callout, ptr, cd);
3241      reqbyte, save the previous value for reinstating. */      previous_callout = NULL;
3242        }
3243    
3244  #ifdef SUPPORT_UTF8    /* In extended mode, skip white space and comments. */
     if (class_charcount == 1 &&  
           (!utf8 ||  
           (!class_utf8 && (!negate_class || class_lastchar < 128))))  
3245    
3246  #else    if ((options & PCRE_EXTENDED) != 0)
3247      if (class_charcount == 1)      {
3248  #endif      if ((cd->ctypes[c] & ctype_space) != 0) continue;
3249        if (c == CHAR_NUMBER_SIGN)
3250        {        {
3251        zeroreqbyte = reqbyte;        ptr++;
3252          while (*ptr != 0)
       /* The OP_NOT opcode works on one-byte characters only. */  
   
       if (negate_class)  
3253          {          {
3254          if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;          if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3255          zerofirstbyte = firstbyte;          ptr++;
         *code++ = OP_NOT;  
         *code++ = class_lastchar;  
         break;  
         }  
   
       /* For a single, positive character, get the value into mcbuffer, and  
       then we can handle this with the normal one-character code. */  
   
3256  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
3257        if (utf8 && class_lastchar > 127)          if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
         mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);  
       else  
3258  #endif  #endif
         {  
         mcbuffer[0] = class_lastchar;  
         mclength = 1;  
3259          }          }
3260        goto ONE_CHAR;        if (*ptr != 0) continue;
       }       /* End of 1-char optimization */  
3261    
3262      /* The general case - not the one-char optimization. If this is the first        /* Else fall through to handle end of string */
3263      thing in the branch, there can be no first char setting, whatever the        c = 0;
3264      repeat count. Any reqbyte setting must remain unchanged after any kind of        }
3265      repeat. */      }
3266    
3267      if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;    /* No auto callout for quantifiers. */
     zerofirstbyte = firstbyte;  
     zeroreqbyte = reqbyte;  
3268    
3269      /* If there are characters with values > 255, we have to compile an    if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3270      extended class, with its own opcode. If there are no characters < 256,      {
3271      we can omit the bitmap. */      previous_callout = code;
3272        code = auto_callout(code, ptr, cd);
3273        }
3274    
3275  #ifdef SUPPORT_UTF8    switch(c)
3276      if (class_utf8)      {
3277        /* ===================================================================*/
3278        case 0:                        /* The branch terminates at string end */
3279        case CHAR_VERTICAL_LINE:       /* or | or ) */
3280        case CHAR_RIGHT_PARENTHESIS:
3281        *firstbyteptr = firstbyte;
3282        *reqbyteptr = reqbyte;
3283        *codeptr = code;
3284        *ptrptr = ptr;
3285        if (lengthptr != NULL)
3286        {        {
3287        *class_utf8data++ = XCL_END;    /* Marks the end of extra data */        if (OFLOW_MAX - *lengthptr < code - last_code)
       *code++ = OP_XCLASS;  
       code += LINK_SIZE;  
       *code = negate_class? XCL_NOT : 0;  
   
       /* If the map is required, install it, and move on to the end of  
       the extra data */  
   
       if (class_charcount > 0)  
         {  
         *code++ |= XCL_MAP;  
         memcpy(code, classbits, 32);  
         code = class_utf8data;  
         }  
   
       /* If the map is not required, slide down the extra data. */  
   
       else  
3288          {          {
3289          int len = class_utf8data - (code + 33);          *errorcodeptr = ERR20;
3290          memmove(code + 1, code + 33, len);          goto FAILED;
         code += len + 1;  
3291          }          }
3292          *lengthptr += (int)(code - last_code);   /* To include callout length */
3293        /* Now fill in the complete length of the item */        DPRINTF((">> end branch\n"));
   
       PUT(previous, 1, code - previous);  
       break;   /* End of class handling */  
       }  
 #endif  
   
     /* If there are no characters > 255, negate the 32-byte map if necessary,  
     and copy it into the code vector. If this is the first thing in the branch,  
     there can be no first char setting, whatever the repeat count. Any reqbyte  
     setting must remain unchanged after any kind of repeat. */  
   
     if (negate_class)  
       {  
       *code++ = OP_NCLASS;  
       for (c = 0; c < 32; c++) code[c] = ~classbits[c];  
       }  
     else  
       {  
       *code++ = OP_CLASS;  
       memcpy(code, classbits, 32);  
3294        }        }
3295      code += 32;      return TRUE;
     break;  
   
     /* Various kinds of repeat; '{' is not necessarily a quantifier, but this  
     has been tested above. */  
   
     case '{':  
     if (!is_quantifier) goto NORMAL_CHAR;  
     ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);  
     if (*errorcodeptr != 0) goto FAILED;  
     goto REPEAT;  
   
     case '*':  
     repeat_min = 0;  
     repeat_max = -1;  
     goto REPEAT;  
3296    
     case '+':  
     repeat_min = 1;  
     repeat_max = -1;  
     goto REPEAT;  
3297    
3298      case '?':      /* ===================================================================*/
3299      repeat_min = 0;      /* Handle single-character metacharacters. In multiline mode, ^ disables
3300      repeat_max = 1;      the setting of any following char as a first character. */
3301    
3302      REPEAT:      case CHAR_CIRCUMFLEX_ACCENT:
3303      if (previous == NULL)      previous = NULL;
3304        if ((options & PCRE_MULTILINE) != 0)
3305        {        {
3306        *errorcodeptr = ERR9;        if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3307        goto FAILED;        *code++ = OP_CIRCM;
3308        }        }
3309        else *code++ = OP_CIRC;
3310        break;
3311    
3312      if (repeat_min == 0)      case CHAR_DOLLAR_SIGN:
3313        {      previous = NULL;
3314        firstbyte = zerofirstbyte;    /* Adjust for zero repeat */      *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3315        reqbyte = zeroreqbyte;        /* Ditto */      break;
       }  
3316    
3317      /* Remember whether this is a variable length repeat */      /* There can never be a first char if '.' is first, whatever happens about
3318        repeats. The value of reqbyte doesn't change either. */
3319    
3320      reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;      case CHAR_DOT:
3321        if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3322        zerofirstbyte = firstbyte;
3323        zeroreqbyte = reqbyte;
3324        previous = code;
3325        *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3326        break;
3327    
     op_type = 0;                    /* Default single-char op codes */  
     possessive_quantifier = FALSE;  /* Default not possessive quantifier */  
3328    
3329      /* Save start of previous item, in case we have to move it up to make space      /* ===================================================================*/
3330      for an inserted OP_ONCE for the additional '+' extension. */      /* Character classes. If the included characters are all < 256, we build a
3331        32-byte bitmap of the permitted characters, except in the special case
3332        where there is only one such character. For negated classes, we build the
3333        map as usual, then invert it at the end. However, we use a different opcode
3334        so that data characters > 255 can be handled correctly.
3335    
3336      tempcode = previous;      If the class contains characters outside the 0-255 range, a different
3337        opcode is compiled. It may optionally have a bit map for characters < 256,
3338        but those above are are explicitly listed afterwards. A flag byte tells
3339        whether the bitmap is present, and whether this is a negated class or not.
3340    
3341      /* If the next character is '+', we have a possessive quantifier. This      In JavaScript compatibility mode, an isolated ']' causes an error. In
3342      implies greediness, whatever the setting of the PCRE_UNGREEDY option.      default (Perl) mode, it is treated as a data character. */
     If the next character is '?' this is a minimizing repeat, by default,  
     but if PCRE_UNGREEDY is set, it works the other way round. We change the  
     repeat type to the non-default. */  
3343    
3344      if (ptr[1] == '+')      case CHAR_RIGHT_SQUARE_BRACKET:
3345        {      if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
       repeat_type = 0;                  /* Force greedy */  
       possessive_quantifier = TRUE;  
       ptr++;  
       }  
     else if (ptr[1] == '?')  
3346        {        {
3347        repeat_type = greedy_non_default;        *errorcodeptr = ERR64;
3348        ptr++;        goto FAILED;
3349        }        }
3350      else repeat_type = greedy_default;      goto NORMAL_CHAR;
3351    
3352      /* If previous was a recursion, we need to wrap it inside brackets so that      case CHAR_LEFT_SQUARE_BRACKET:
3353      it can be replicated if necessary. */      previous = code;
3354    
3355      if (*previous == OP_RECURSE)      /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3356        they are encountered at the top level, so we'll do that too. */
3357    
3358        if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3359             ptr[1] == CHAR_EQUALS_SIGN) &&
3360            check_posix_syntax(ptr, &tempptr))
3361        {        {
3362        memmove(previous + 1 + LINK_SIZE, previous, 1 + LINK_SIZE);        *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3363        code += 1 + LINK_SIZE;        goto FAILED;
       *previous = OP_BRA;  
       PUT(previous, 1, code - previous);  
       *code = OP_KET;  
       PUT(code, 1, code - previous);  
       code += 1 + LINK_SIZE;  
3364        }        }
3365    
3366      /* If previous was a character match, abolish the item and generate a      /* If the first character is '^', set the negation flag and skip it. Also,
3367      repeat item instead. If a char item has a minumum of more than one, ensure      if the first few characters (either before or after ^) are \Q\E or \E we
3368      that it is set in reqbyte - it might not be if a sequence such as x{3} is      skip them too. This makes for compatibility with Perl. */
     the first thing in a branch because the x will have gone into firstbyte  
     instead.  */  
3369    
3370      if (*previous == OP_CHAR || *previous == OP_CHARNC)      negate_class = FALSE;
3371        for (;;)
3372        {        {
3373        /* Deal with UTF-8 characters that take up more than one byte. It's        c = *(++ptr);
3374        easier to write this out separately than try to macrify it. Use c to        if (c == CHAR_BACKSLASH)
       hold the length of the character in bytes, plus 0x80 to flag that it's a  
       length rather than a small character. */  
   
 #ifdef SUPPORT_UTF8  
       if (utf8 && (code[-1] & 0x80) != 0)  
         {  
         uschar *lastchar = code - 1;  
         while((*lastchar & 0xc0) == 0x80) lastchar--;  
         c = code - lastchar;            /* Length of UTF-8 character */  
         memcpy(utf8_char, lastchar, c); /* Save the char */  
         c |= 0x80;                      /* Flag c as a length */  
         }  
       else  
 #endif  
   
       /* Handle the case of a single byte - either with no UTF8 support, or  
       with UTF-8 disabled, or for a UTF-8 character < 128. */  
   
3375          {          {
3376          c = code[-1];          if (ptr[1] == CHAR_E)
3377          if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;            ptr++;
3378            else if (strncmp((const char *)ptr+1,
3379                              STR_Q STR_BACKSLASH STR_E, 3) == 0)
3380              ptr += 3;
3381            else
3382              break;
3383          }          }
3384          else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3385        goto OUTPUT_SINGLE_REPEAT;   /* Code shared with single character types */          negate_class = TRUE;
3386          else break;
3387        }        }
3388    
3389      /* If previous was a single negated character ([^a] or similar), we use      /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3390      one of the special opcodes, replacing it. The code is shared with single-      an initial ']' is taken as a data character -- the code below handles
3391      character repeats by setting opt_type to add a suitable offset into      that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3392      repeat_type. OP_NOT is currently used only for single-byte chars. */      [^] must match any character, so generate OP_ALLANY. */
3393    
3394      else if (*previous == OP_NOT)      if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3395            (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3396        {        {
3397        op_type = OP_NOTSTAR - OP_STAR;  /* Use "not" opcodes */        *code++ = negate_class? OP_ALLANY : OP_FAIL;
3398        c = previous[1];        if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3399        goto OUTPUT_SINGLE_REPEAT;        zerofirstbyte = firstbyte;
3400          break;
3401        }        }
3402    
3403      /* If previous was a character type match (\d or similar), abolish it and      /* If a class contains a negative special such as \S, we need to flip the
3404      create a suitable repeat item. The code is shared with single-character      negation flag at the end, so that support for characters > 255 works
3405      repeats by setting op_type to add a suitable offset into repeat_type. Note      correctly (they are all included in the class). */
     the the Unicode property types will be present only when SUPPORT_UCP is  
     defined, but we don't wrap the little bits of code here because it just  
     makes it horribly messy. */  
3406    
3407      else if (*previous < OP_EODN)      should_flip_negation = FALSE;
       {  
       uschar *oldcode;  
       int prop_type;  
       op_type = OP_TYPESTAR - OP_STAR;  /* Use type opcodes */  
       c = *previous;  
3408    
3409        OUTPUT_SINGLE_REPEAT:      /* Keep a count of chars with values < 256 so that we can optimize the case
3410        prop_type = (*previous == OP_PROP || *previous == OP_NOTPROP)?      of just a single character (as long as it's < 256). However, For higher
3411          previous[1] : -1;      valued UTF-8 characters, we don't yet do any optimization. */
3412    
3413        oldcode = code;      class_charcount = 0;
3414        code = previous;                  /* Usually overwrite previous item */      class_lastchar = -1;
3415    
3416        /* If the maximum is zero then the minimum must also be zero; Perl allows      /* Initialize the 32-char bit map to all zeros. We build the map in a
3417        this case, so we do too - by simply omitting the item altogether. */      temporary bit of memory, in case the class contains only 1 character (less
3418        than 256), because in that case the compiled code doesn't use the bit map.
3419        */
3420    
3421        if (repeat_max == 0) goto END_REPEAT;      memset(classbits, 0, 32 * sizeof(uschar));
3422    
3423        /* All real repeats make it impossible to handle partial matching (maybe  #ifdef SUPPORT_UTF8
3424        one day we will be able to remove this restriction). */      class_utf8 = FALSE;                       /* No chars >= 256 */
3425        class_utf8data = code + LINK_SIZE + 2;    /* For UTF-8 items */
3426        class_utf8data_base = class_utf8data;     /* For resetting in pass 1 */
3427    #endif
3428    
3429        if (repeat_max != 1) cd->nopartial = TRUE;      /* Process characters until ] is reached. By writing this as a "do" it
3430        means that an initial ] is taken as a data character. At the start of the
3431        loop, c contains the first byte of the character. */
3432    
3433        /* Combine the op_type with the repeat_type */      if (c != 0) do
3434          {
3435          const uschar *oldptr;
3436    
3437        repeat_type += op_type;  #ifdef SUPPORT_UTF8
3438          if (utf8 && c > 127)
3439            {                           /* Braces are required because the */
3440            GETCHARLEN(c, ptr, ptr);    /* macro generates multiple statements */
3441            }
3442    
3443        /* A minimum of zero is handled either as the special case * or ?, or as        /* In the pre-compile phase, accumulate the length of any UTF-8 extra
3444        an UPTO, with the maximum given. */        data and reset the pointer. This is so that very large classes that
3445          contain a zillion UTF-8 characters no longer overwrite the work space
3446          (which is on the stack). */
3447    
3448        if (repeat_min == 0)        if (lengthptr != NULL)
3449          {          {
3450          if (repeat_max == -1) *code++ = OP_STAR + repeat_type;          *lengthptr += class_utf8data - class_utf8data_base;
3451            else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;          class_utf8data = class_utf8data_base;
         else  
           {  
           *code++ = OP_UPTO + repeat_type;  
           PUT2INC(code, 0, repeat_max);  
           }  
3452          }          }
3453    
3454        /* A repeat minimum of 1 is optimized into some special cases. If the  #endif
       maximum is unlimited, we use OP_PLUS. Otherwise, the original item it  
       left in place and, if the maximum is greater than 1, we use OP_UPTO with  
       one less than the maximum. */  
3455    
3456        else if (repeat_min == 1)        /* Inside \Q...\E everything is literal except \E */
3457    
3458          if (inescq)
3459          {          {
3460          if (repeat_max == -1)          if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)  /* If we are at \E */
           *code++ = OP_PLUS + repeat_type;  
         else  
3461            {            {
3462            code = oldcode;                 /* leave previous item in place */            inescq = FALSE;                   /* Reset literal state */
3463            if (repeat_max == 1) goto END_REPEAT;            ptr++;                            /* Skip the 'E' */
3464            *code++ = OP_UPTO + repeat_type;            continue;                         /* Carry on with next */
           PUT2INC(code, 0, repeat_max - 1);  
3465            }            }
3466            goto CHECK_RANGE;                   /* Could be range if \E follows */
3467          }          }
3468    
3469        /* The case {n,n} is just an EXACT, while the general case {n,m} is        /* Handle POSIX class names. Perl allows a negation extension of the
3470        handled as an EXACT followed by an UPTO. */        form [:^name:]. A square bracket that doesn't match the syntax is
3471          treated as a literal. We also recognize the POSIX constructions
3472          [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3473          5.6 and 5.8 do. */
3474    
3475        else        if (c == CHAR_LEFT_SQUARE_BRACKET &&
3476              (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3477               ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3478          {          {
3479          *code++ = OP_EXACT + op_type;  /* NB EXACT doesn't have repeat_type */          BOOL local_negate = FALSE;
3480          PUT2INC(code, 0, repeat_min);          int posix_class, taboffset, tabopt;
3481            register const uschar *cbits = cd->cbits;
3482          /* If the maximum is unlimited, insert an OP_STAR. Before doing so,          uschar pbits[32];
         we have to insert the character for the previous code. For a repeated  
         Unicode property match, there is an extra byte that defines the  
         required property. In UTF-8 mode, long characters have their length in  
         c, with the 0x80 bit as a flag. */  
3483    
3484          if (repeat_max < 0)          if (ptr[1] != CHAR_COLON)
3485            {            {
3486  #ifdef SUPPORT_UTF8            *errorcodeptr = ERR31;
3487            if (utf8 && c >= 128)            goto FAILED;
             {  
             memcpy(code, utf8_char, c & 7);  
             code += c & 7;  
             }  
           else  
 #endif  
             {  
             *code++ = c;  
             if (prop_type >= 0) *code++ = prop_type;  
             }  
           *code++ = OP_STAR + repeat_type;  
3488            }            }
3489    
3490          /* Else insert an UPTO if the max is greater than the min, again          ptr += 2;
3491          preceded by the character, for the previously inserted code. */          if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3492              {
3493              local_negate = TRUE;
3494              should_flip_negation = TRUE;  /* Note negative special */
3495              ptr++;
3496              }
3497    
3498          else if (repeat_max != repeat_min)          posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3499            if (posix_class < 0)
3500            {            {
3501  #ifdef SUPPORT_UTF8            *errorcodeptr = ERR30;
3502            if (utf8 && c >= 128)            goto FAILED;
             {  
             memcpy(code, utf8_char, c & 7);  
             code += c & 7;  
             }  
           else  
 #endif  
           *code++ = c;  
           if (prop_type >= 0) *code++ = prop_type;  
           repeat_max -= repeat_min;  
           *code++ = OP_UPTO + repeat_type;  
           PUT2INC(code, 0, repeat_max);  
3503            }            }
         }  
3504    
3505        /* The character or character type itself comes last in all cases. */          /* If matching is caseless, upper and lower are converted to
3506            alpha. This relies on the fact that the class table starts with
3507            alpha, lower, upper as the first 3 entries. */
3508    
3509  #ifdef SUPPORT_UTF8          if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3510        if (utf8 && c >= 128)            posix_class = 0;
         {  
         memcpy(code, utf8_char, c & 7);  
         code += c & 7;  
         }  
       else  
 #endif  
       *code++ = c;  
3511    
3512        /* For a repeated Unicode property match, there is an extra byte that          /* When PCRE_UCP is set, some of the POSIX classes are converted to
3513        defines the required property. */          different escape sequences that use Unicode properties. */
3514    
3515  #ifdef SUPPORT_UCP  #ifdef SUPPORT_UCP
3516        if (prop_type >= 0) *code++ = prop_type;          if ((options & PCRE_UCP) != 0)
3517  #endif            {
3518        }            int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3519              if (posix_substitutes[pc] != NULL)
3520      /* If previous was a character class or a back reference, we put the repeat              {
3521      stuff after it, but just skip the item if the repeat was {0,0}. */              nestptr = tempptr + 1;
3522                ptr = posix_substitutes[pc] - 1;
3523      else if (*previous == OP_CLASS ||              continue;
3524               *previous == OP_NCLASS ||              }
3525  #ifdef SUPPORT_UTF8            }
              *previous == OP_XCLASS ||  
3526  #endif  #endif
3527               *previous == OP_REF)          /* In the non-UCP case, we build the bit map for the POSIX class in a
3528        {          chunk of local store because we may be adding and subtracting from it,
3529        if (repeat_max == 0)          and we don't want to subtract bits that may be in the main map already.
3530          {          At the end we or the result into the bit map that is being built. */
         code = previous;  
         goto END_REPEAT;  
         }  
   
       /* All real repeats make it impossible to handle partial matching (maybe  
       one day we will be able to remove this restriction). */  
   
       if (repeat_max != 1) cd->nopartial = TRUE;  
   
       if (repeat_min == 0 && repeat_max == -1)  
         *code++ = OP_CRSTAR + repeat_type;  
       else if (repeat_min == 1 && repeat_max == -1)  
         *code++ = OP_CRPLUS + repeat_type;  
       else if (repeat_min == 0 && repeat_max == 1)  
         *code++ = OP_CRQUERY + repeat_type;  
       else  
         {  
         *code++ = OP_CRRANGE + repeat_type;  
         PUT2INC(code, 0, repeat_min);  
         if (repeat_max == -1) repeat_max = 0;  /* 2-byte encoding for max */  
         PUT2INC(code, 0, repeat_max);  
         }  
       }  
3531    
3532      /* If previous was a bracket group, we may have to replicate it in certain          posix_class *= 3;
     cases. */  
3533    
3534      else if (*previous >= OP_BRA || *previous == OP_ONCE ||          /* Copy in the first table (always present) */
              *previous == OP_COND)  
       {  
       register int i;  
       int ketoffset = 0;  
       int len = code - previous;  
       uschar *bralink = NULL;  
3535    
3536        /* If the maximum repeat count is unlimited, find the end of the bracket          memcpy(pbits, cbits + posix_class_maps[posix_class],
3537        by scanning through from the start, and compute the offset back to it            32 * sizeof(uschar));
       from the current code pointer. There may be an OP_OPT setting following  
       the final KET, so we can't find the end just by going back from the code  
       pointer. */  
   
       if (repeat_max == -1)  
         {  
         register uschar *ket = previous;  
         do ket += GET(ket, 1); while (*ket != OP_KET);  
         ketoffset = code - ket;  
         }  
3538    
3539        /* The case of a zero minimum is special because of the need to stick          /* If there is a second table, add or remove it as required. */
       OP_BRAZERO in front of it, and because the group appears once in the  
       data, whereas in other cases it appears the minimum number of times. For  
       this reason, it is simplest to treat this case separately, as otherwise  
       the code gets far too messy. There are several special subcases when the  
       minimum is zero. */  
3540    
3541        if (repeat_min == 0)          taboffset = posix_class_maps[posix_class + 1];
3542          {          tabopt = posix_class_maps[posix_class + 2];
         /* If the maximum is also zero, we just omit the group from the output  
         altogether. */  
3543    
3544          if (repeat_max == 0)          if (taboffset >= 0)
3545            {            {
3546            code = previous;            if (tabopt >= 0)
3547            goto END_REPEAT;              for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3548              else
3549                for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3550            }            }
3551    
3552          /* If the maximum is 1 or unlimited, we just have to stick in the          /* Not see if we need to remove any special characters. An option
3553          BRAZERO and do no more at this point. However, we do need to adjust          value of 1 removes vertical space and 2 removes underscore. */
         any OP_RECURSE calls inside the group that refer to the group itself or  
         any internal group, because the offset is from the start of the whole  
         regex. Temporarily terminate the pattern while doing this. */  
3554    
3555          if (repeat_max <= 1)          if (tabopt < 0) tabopt = -tabopt;
3556            {          if (tabopt == 1) pbits[1] &= ~0x3c;
3557            *code = OP_END;            else if (tabopt == 2) pbits[11] &= 0x7f;
           adjust_recurse(previous, 1, utf8, cd);  
           memmove(previous+1, previous, len);  
           code++;  
           *previous++ = OP_BRAZERO + repeat_type;  
           }  
3558    
3559          /* If the maximum is greater than 1 and limited, we have to replicate          /* Add the POSIX table or its complement into the main table that is
3560          in a nested fashion, sticking OP_BRAZERO before each set of brackets.          being built and we are done. */
         The first one has to be handled carefully because it's the original  
         copy, which has to be moved up. The remainder can be handled by code  
         that is common with the non-zero minimum case below. We have to  
         adjust the value or repeat_max, since one less copy is required. Once  
         again, we may have to adjust any OP_RECURSE calls inside the group. */  
3561    
3562            if (local_negate)
3563              for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3564          else          else
3565            {            for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
           int offset;  
           *code = OP_END;  
           adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd);  
           memmove(previous + 2 + LINK_SIZE, previous, len);  
           code += 2 + LINK_SIZE;  
           *previous++ = OP_BRAZERO + repeat_type;  
           *previous++ = OP_BRA;  
   
           /* We chain together the bracket offset fields that have to be  
           filled in later when the ends of the brackets are reached. */  
   
           offset = (bralink == NULL)? 0 : previous - bralink;  
           bralink = previous;  
           PUTINC(previous, 0, offset);  
           }  
3566    
3567          repeat_max--;          ptr = tempptr + 1;
3568            class_charcount = 10;  /* Set > 1; assumes more than 1 per class */
3569            continue;    /* End of POSIX syntax handling */
3570          }          }
3571    
3572        /* If the minimum is greater than zero, replicate the group as many        /* Backslash may introduce a single character, or it may introduce one
3573        times as necessary, and adjust the maximum to the number of subsequent        of the specials, which just set a flag. The sequence \b is a special
3574        copies that we need. If we set a first char from the group, and didn't        case. Inside a class (and only there) it is treated as backspace. We
3575        set a required char, copy the latter from the former. */        assume that other escapes have more than one character in them, so set
3576          class_charcount bigger than one. Unrecognized escapes fall through and
3577          are either treated as literal characters (by default), or are faulted if
3578          PCRE_EXTRA is set. */
3579    
3580        else        if (c == CHAR_BACKSLASH)
3581          {          {
3582          if (repeat_min > 1)          c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3583            if (*errorcodeptr != 0) goto FAILED;
3584    
3585            if (-c == ESC_b) c = CHAR_BS;    /* \b is backspace in a class */
3586            else if (-c == ESC_Q)            /* Handle start of quoted string */
3587            {            {
3588            if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;            if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
           for (i = 1; i < repeat_min; i++)  
3589              {              {
3590              memcpy(code, previous, len);              ptr += 2; /* avoid empty string */
             code += len;  
3591              }              }
3592              else inescq = TRUE;
3593              continue;
3594            }            }
3595          if (repeat_max > 0) repeat_max -= repeat_min;          else if (-c == ESC_E) continue;  /* Ignore orphan \E */
         }  
   
       /* This code is common to both the zero and non-zero minimum cases. If  
       the maximum is limited, it replicates the group in a nested fashion,  
       remembering the bracket starts on a stack. In the case of a zero minimum,  
       the first one was set up above. In all cases the repeat_max now specifies  
       the number of additional copies needed. */  
3596    
3597        if (repeat_max >= 0)          if (c < 0)
         {  
         for (i = repeat_max - 1; i >= 0; i--)  
3598            {            {
3599            *code++ = OP_BRAZERO + repeat_type;            register const uschar *cbits = cd->cbits;
3600              class_charcount += 2;     /* Greater than 1 is what matters */
           /* All but the final copy start a new nesting, maintaining the  
           chain of brackets outstanding. */  
3601    
3602            if (i != 0)            switch (-c)
3603              {              {
3604              int offset;  #ifdef SUPPORT_UCP
3605              *code++ = OP_BRA;              case ESC_du:     /* These are the values given for \d etc */
3606              offset = (bralink == NULL)? 0 : code - bralink;              case ESC_DU:     /* when PCRE_UCP is set. We replace the */
3607              bralink = code;              case ESC_wu:     /* escape sequence with an appropriate \p */
3608              PUTINC(code, 0, offset);              case ESC_WU:     /* or \P to test Unicode properties instead */
3609              }              case ESC_su:     /* of the default ASCII testing. */
3610                case ESC_SU:
3611                nestptr = ptr;
3612                ptr = substitutes[-c - ESC_DU] - 1;  /* Just before substitute */
3613                class_charcount -= 2;                /* Undo! */
3614                continue;
3615    #endif
3616                case ESC_d:
3617                for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3618                continue;
3619    
3620            memcpy(code, previous, len);              case ESC_D:
3621            code += len;              should_flip_negation = TRUE;
3622            }              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3623                continue;
3624    
3625          /* Now chain through the pending brackets, and fill in their length              case ESC_w:
3626          fields (which are holding the chain links pro tem). */              for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3627                continue;
3628    
3629          while (bralink != NULL)              case ESC_W:
3630            {              should_flip_negation = TRUE;
3631            int oldlinkoffset;              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3632            int offset = code - bralink + 1;              continue;
           uschar *bra = code - offset;  
           oldlinkoffset = GET(bra, 1);  
           bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;  
           *code++ = OP_KET;  
           PUTINC(code, 0, offset);  
           PUT(bra, 1, offset);  
           }  
         }  
3633    
3634        /* If the maximum is unlimited, set a repeater in the final copy. We              /* Perl 5.004 onwards omits VT from \s, but we must preserve it
3635        can't just offset backwards from the current code point, because we              if it was previously set by something earlier in the character