/[pcre]/code/branches/pcre16/pcre_compile.c
ViewVC logotype

Diff of /code/branches/pcre16/pcre_compile.c

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

revision 81 by nigel, Sat Feb 24 21:40:59 2007 UTC revision 613 by ph10, Sat Jul 2 16:59:52 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 PCRE_DEBUG is defined, we need the pcre_printint() function, which is
57    also used by pcretest. PCRE_DEBUG is not defined when building a production
58    library. */
59    
60    #ifdef PCRE_DEBUG
61    #include "pcre_printint.src"
62    #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 BRASTACK_SIZE 200  #define COMPILE_WORK_SIZE (4096)
94    
95    /* 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 63  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 87  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 106  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    #define STRING(a)  # a
313    #define XSTRING(s) STRING(s)
314    
315  static const char *error_texts[] = {  /* The texts of compile-time error messages. These are "char *" because they
316    "no error",  are passed to the outside world. Do not ever re-use any error number, because
317    "\\ at end of pattern",  they are documented. Always add a new error instead. Messages marked DEAD below
318    "\\c at end of pattern",  are no longer used. This used to be a table of strings, but in order to reduce
319    "unrecognized character follows \\",  the number of relocations needed when a shared library is loaded dynamically,
320    "numbers out of order in {} quantifier",  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"
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      ;
413    
414  /* Table to identify digits and hex digits. This is used when compiling  /* Table to identify digits and hex digits. This is used when compiling
415  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 220  For convenience, we use the same bit def Line 427  For convenience, we use the same bit def
427    
428  Then we can use ctype_digit and ctype_xdigit in the code. */  Then we can use ctype_digit and ctype_xdigit in the code. */
429    
430  #if !EBCDIC    /* This is the "normal" case, for ASCII systems */  #ifndef EBCDIC
431    
432    /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
433    UTF-8 mode. */
434    
435  static const unsigned char digitab[] =  static const unsigned char digitab[] =
436    {    {
437    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*   0-  7 */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*   0-  7 */
# Line 256  static const unsigned char digitab[] = Line 467  static const unsigned char digitab[] =
467    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
468    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
469    
470  #else          /* This is the "abnormal" case, for EBCDIC systems */  #else
471    
472    /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
473    
474  static const unsigned char digitab[] =  static const unsigned char digitab[] =
475    {    {
476    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*   0-  7  0 */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*   0-  7  0 */
# Line 270  static const unsigned char digitab[] = Line 484  static const unsigned char digitab[] =
484    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*    - 71 40 */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*    - 71 40 */
485    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  72- |     */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  72- |     */
486    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  & - 87 50 */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  & - 87 50 */
487    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  88-     */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  88- 95    */
488    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  - -103 60 */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  - -103 60 */
489    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ?     */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ?     */
490    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
# Line 304  static const unsigned char ebcdic_charta Line 518  static const unsigned char ebcdic_charta
518    0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*    - 71 */    0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*    - 71 */
519    0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /*  72- |  */    0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /*  72- |  */
520    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  & - 87 */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  & - 87 */
521    0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /*  88-  */    0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /*  88- 95 */
522    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  - -103 */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  - -103 */
523    0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ?  */    0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ?  */
524    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
# Line 331  static const unsigned char ebcdic_charta Line 545  static const unsigned char ebcdic_charta
545  /* Definition to allow mutual recursion */  /* Definition to allow mutual recursion */
546    
547  static BOOL  static BOOL
548    compile_regex(int, int, int *, uschar **, const uschar **, int *, BOOL, int,    compile_regex(int, uschar **, const uschar **, int *, BOOL, BOOL, int, int *,
549      int *, int *, branch_chain *, compile_data *);      int *, branch_chain *, compile_data *, int *);
550    
551    
552    
553    /*************************************************
554    *            Find an error text                  *
555    *************************************************/
556    
557    /* The error texts are now all in one long string, to save on relocations. As
558    some of the text is of unknown length, we can't use a table of offsets.
559    Instead, just count through the strings. This is not a performance issue
560    because it happens only when there has been a compilation error.
561    
562    Argument:   the error number
563    Returns:    pointer to the error string
564    */
565    
566    static const char *
567    find_error_text(int n)
568    {
569    const char *s = error_texts;
570    for (; n > 0; n--)
571      {
572      while (*s++ != 0) {};
573      if (*s == 0) return "Error text not found (please report)";
574      }
575    return s;
576    }
577    
578    
579  /*************************************************  /*************************************************
# Line 342  static BOOL Line 582  static BOOL
582    
583  /* 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
584  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
585  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
586  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
587  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,
588    ptr is pointing at the \. On exit, it is on the final character of the escape
589    sequence.
590    
591  Arguments:  Arguments:
592    ptrptr         points to the pattern position pointer    ptrptr         points to the pattern position pointer
# Line 355  Arguments: Line 597  Arguments:
597    
598  Returns:         zero or positive => a data character  Returns:         zero or positive => a data character
599                   negative => a special escape sequence                   negative => a special escape sequence
600                   on error, errorptr is set                   on error, errorcodeptr is set
601  */  */
602    
603  static int  static int
604  check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,  check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
605    int options, BOOL isclass)    int options, BOOL isclass)
606  {  {
607  const uschar *ptr = *ptrptr;  BOOL utf8 = (options & PCRE_UTF8) != 0;
608    const uschar *ptr = *ptrptr + 1;
609  int c, i;  int c, i;
610    
611    GETCHARINCTEST(c, ptr);           /* Get character value, increment pointer */
612    ptr--;                            /* Set pointer back to the last byte */
613    
614  /* 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. */
615    
 c = *(++ptr);  
616  if (c == 0) *errorcodeptr = ERR1;  if (c == 0) *errorcodeptr = ERR1;
617    
618  /* 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
619  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.
620  Otherwise further processing may be required. */  Otherwise further processing may be required. */
621    
622  #if !EBCDIC    /* ASCII coding */  #ifndef EBCDIC  /* ASCII/UTF-8 coding */
623  else if (c < '0' || c > 'z') {}                           /* Not alphameric */  else if (c < CHAR_0 || c > CHAR_z) {}                     /* Not alphanumeric */
624  else if ((i = escapes[c - '0']) != 0) c = i;  else if ((i = escapes[c - CHAR_0]) != 0) c = i;
625    
626  #else          /* EBCDIC coding */  #else           /* EBCDIC coding */
627  else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {}   /* Not alphameric */  else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {}   /* Not alphanumeric */
628  else if ((i = escapes[c - 0x48]) != 0)  c = i;  else if ((i = escapes[c - 0x48]) != 0)  c = i;
629  #endif  #endif
630    
# Line 388  else if ((i = escapes[c - 0x48]) != 0) Line 633  else if ((i = escapes[c - 0x48]) != 0)
633  else  else
634    {    {
635    const uschar *oldptr;    const uschar *oldptr;
636      BOOL braced, negated;
637    
638    switch (c)    switch (c)
639      {      {
640      /* 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
641      error. */      error. */
642    
643      case 'l':      case CHAR_l:
644      case 'L':      case CHAR_L:
645      case 'N':      case CHAR_u:
646      case 'u':      case CHAR_U:
     case 'U':  
647      *errorcodeptr = ERR37;      *errorcodeptr = ERR37;
648      break;      break;
649    
650        /* \g must be followed by one of a number of specific things:
651    
652        (1) A number, either plain or braced. If positive, it is an absolute
653        backreference. If negative, it is a relative backreference. This is a Perl
654        5.10 feature.
655    
656        (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
657        is part of Perl's movement towards a unified syntax for back references. As
658        this is synonymous with \k{name}, we fudge it up by pretending it really
659        was \k.
660    
661        (3) For Oniguruma compatibility we also support \g followed by a name or a
662        number either in angle brackets or in single quotes. However, these are
663        (possibly recursive) subroutine calls, _not_ backreferences. Just return
664        the -ESC_g code (cf \k). */
665    
666        case CHAR_g:
667        if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
668          {
669          c = -ESC_g;
670          break;
671          }
672    
673        /* Handle the Perl-compatible cases */
674    
675        if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
676          {
677          const uschar *p;
678          for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
679            if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break;
680          if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
681            {
682            c = -ESC_k;
683            break;
684            }
685          braced = TRUE;
686          ptr++;
687          }
688        else braced = FALSE;
689    
690        if (ptr[1] == CHAR_MINUS)
691          {
692          negated = TRUE;
693          ptr++;
694          }
695        else negated = FALSE;
696    
697        c = 0;
698        while ((digitab[ptr[1]] & ctype_digit) != 0)
699          c = c * 10 + *(++ptr) - CHAR_0;
700    
701        if (c < 0)   /* Integer overflow */
702          {
703          *errorcodeptr = ERR61;
704          break;
705          }
706    
707        if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
708          {
709          *errorcodeptr = ERR57;
710          break;
711          }
712    
713        if (c == 0)
714          {
715          *errorcodeptr = ERR58;
716          break;
717          }
718    
719        if (negated)
720          {
721          if (c > bracount)
722            {
723            *errorcodeptr = ERR15;
724            break;
725            }
726          c = bracount - (c - 1);
727          }
728    
729        c = -(ESC_REF + c);
730        break;
731    
732      /* The handling of escape sequences consisting of a string of digits      /* The handling of escape sequences consisting of a string of digits
733      starting with one that is not zero is not straightforward. By experiment,      starting with one that is not zero is not straightforward. By experiment,
734      the way Perl works seems to be as follows:      the way Perl works seems to be as follows:
# Line 413  else Line 741  else
741      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
742      character class, \ followed by a digit is always an octal number. */      character class, \ followed by a digit is always an octal number. */
743    
744      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:
745      case '6': case '7': case '8': case '9':      case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
746    
747      if (!isclass)      if (!isclass)
748        {        {
749        oldptr = ptr;        oldptr = ptr;
750        c -= '0';        c -= CHAR_0;
751        while ((digitab[ptr[1]] & ctype_digit) != 0)        while ((digitab[ptr[1]] & ctype_digit) != 0)
752          c = c * 10 + *(++ptr) - '0';          c = c * 10 + *(++ptr) - CHAR_0;
753          if (c < 0)    /* Integer overflow */
754            {
755            *errorcodeptr = ERR61;
756            break;
757            }
758        if (c < 10 || c <= bracount)        if (c < 10 || c <= bracount)
759          {          {
760          c = -(ESC_REF + c);          c = -(ESC_REF + c);
# Line 434  else Line 767  else
767      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.
768      Thus we have to pull back the pointer by one. */      Thus we have to pull back the pointer by one. */
769    
770      if ((c = *ptr) >= '8')      if ((c = *ptr) >= CHAR_8)
771        {        {
772        ptr--;        ptr--;
773        c = 0;        c = 0;
# Line 442  else Line 775  else
775        }        }
776    
777      /* \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
778      larger first octal digit. */      larger first octal digit. The original code used just to take the least
779        significant 8 bits of octal numbers (I think this is what early Perls used
780      case '0':      to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
781      c -= '0';      than 3 octal digits. */
782      while(i++ < 2 && ptr[1] >= '0' && ptr[1] <= '7')  
783          c = c * 8 + *(++ptr) - '0';      case CHAR_0:
784      c &= 255;     /* Take least significant 8 bits */      c -= CHAR_0;
785        while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
786            c = c * 8 + *(++ptr) - CHAR_0;
787        if (!utf8 && c > 255) *errorcodeptr = ERR51;
788      break;      break;
789    
790      /* \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
791      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
792        treated as a data character. */
793    
794      case 'x':      case CHAR_x:
795  #ifdef SUPPORT_UTF8      if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
     if (ptr[1] == '{' && (options & PCRE_UTF8) != 0)  
796        {        {
797        const uschar *pt = ptr + 2;        const uschar *pt = ptr + 2;
798        register int count = 0;        int count = 0;
799    
800        c = 0;        c = 0;
801        while ((digitab[*pt] & ctype_xdigit) != 0)        while ((digitab[*pt] & ctype_xdigit) != 0)
802          {          {
803          int cc = *pt++;          register int cc = *pt++;
804            if (c == 0 && cc == CHAR_0) continue;     /* Leading zeroes */
805          count++;          count++;
806  #if !EBCDIC    /* ASCII coding */  
807          if (cc >= 'a') cc -= 32;               /* Convert to upper case */  #ifndef EBCDIC  /* ASCII/UTF-8 coding */
808          c = c * 16 + cc - ((cc < 'A')? '0' : ('A' - 10));          if (cc >= CHAR_a) cc -= 32;               /* Convert to upper case */
809  #else          /* EBCDIC coding */          c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
810          if (cc >= 'a' && cc <= 'z') cc += 64;  /* Convert to upper case */  #else           /* EBCDIC coding */
811          c = c * 16 + cc - ((cc >= '0')? '0' : ('A' - 10));          if (cc >= CHAR_a && cc <= CHAR_z) cc += 64;  /* Convert to upper case */
812            c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
813  #endif  #endif
814          }          }
815        if (*pt == '}')  
816          if (*pt == CHAR_RIGHT_CURLY_BRACKET)
817          {          {
818          if (c < 0 || count > 8) *errorcodeptr = ERR34;          if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
819          ptr = pt;          ptr = pt;
820          break;          break;
821          }          }
822    
823        /* 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
824        recognize this construct; fall through to the normal \x handling. */        recognize this construct; fall through to the normal \x handling. */
825        }        }
 #endif  
826    
827      /* Read just a single hex char */      /* Read just a single-byte hex-defined char */
828    
829      c = 0;      c = 0;
830      while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)      while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
831        {        {
832        int cc;                               /* Some compilers don't like ++ */        int cc;                                  /* Some compilers don't like */
833        cc = *(++ptr);                        /* in initializers */        cc = *(++ptr);                           /* ++ in initializers */
834  #if !EBCDIC    /* ASCII coding */  #ifndef EBCDIC  /* ASCII/UTF-8 coding */
835        if (cc >= 'a') cc -= 32;              /* Convert to upper case */        if (cc >= CHAR_a) cc -= 32;              /* Convert to upper case */
836        c = c * 16 + cc - ((cc < 'A')? '0' : ('A' - 10));        c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
837  #else          /* EBCDIC coding */  #else           /* EBCDIC coding */
838        if (cc <= 'z') cc += 64;              /* Convert to upper case */        if (cc <= CHAR_z) cc += 64;              /* Convert to upper case */
839        c = c * 16 + cc - ((cc >= '0')? '0' : ('A' - 10));        c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
840  #endif  #endif
841        }        }
842      break;      break;
843    
844      /* Other special escapes not starting with a digit are straightforward */      /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
845        An error is given if the byte following \c is not an ASCII character. This
846        coding is ASCII-specific, but then the whole concept of \cx is
847        ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
848    
849      case 'c':      case CHAR_c:
850      c = *(++ptr);      c = *(++ptr);
851      if (c == 0)      if (c == 0)
852        {        {
853        *errorcodeptr = ERR2;        *errorcodeptr = ERR2;
854        return 0;        break;
855        }        }
856    #ifndef EBCDIC    /* ASCII/UTF-8 coding */
857      /* A letter is upper-cased; then the 0x40 bit is flipped. This coding      if (c > 127)  /* Excludes all non-ASCII in either mode */
858      is ASCII-specific, but then the whole concept of \cx is ASCII-specific.        {
859      (However, an EBCDIC equivalent has now been added.) */        *errorcodeptr = ERR68;
860          break;
861  #if !EBCDIC    /* ASCII coding */        }
862      if (c >= 'a' && c <= 'z') c -= 32;      if (c >= CHAR_a && c <= CHAR_z) c -= 32;
863      c ^= 0x40;      c ^= 0x40;
864  #else          /* EBCDIC coding */  #else             /* EBCDIC coding */
865      if (c >= 'a' && c <= 'z') c += 64;      if (c >= CHAR_a && c <= CHAR_z) c += 64;
866      c ^= 0xC0;      c ^= 0xC0;
867  #endif  #endif
868      break;      break;
869    
870      /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any      /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
871      other alphameric following \ is an error if PCRE_EXTRA was set; otherwise,      other alphanumeric following \ is an error if PCRE_EXTRA was set;
872      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
873      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
874      in future, so I haven't "optimized" it. */      be again in future, so I haven't "optimized" it. */
875    
876      default:      default:
877      if ((options & PCRE_EXTRA) != 0) switch(c)      if ((options & PCRE_EXTRA) != 0) switch(c)
# Line 541  else Line 884  else
884      }      }
885    }    }
886    
887    /* Perl supports \N{name} for character names, as well as plain \N for "not
888    newline". PCRE does not support \N{name}. */
889    
890    if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET)
891      *errorcodeptr = ERR37;
892    
893    /* If PCRE_UCP is set, we change the values for \d etc. */
894    
895    if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
896      c -= (ESC_DU - ESC_D);
897    
898    /* Set the pointer to the final character before returning. */
899    
900  *ptrptr = ptr;  *ptrptr = ptr;
901  return c;  return c;
902  }  }
# Line 560  escape sequence. Line 916  escape sequence.
916  Argument:  Argument:
917    ptrptr         points to the pattern position pointer    ptrptr         points to the pattern position pointer
918    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
919      dptr           points to an int that is set to the detailed property value
920    errorcodeptr   points to the error code variable    errorcodeptr   points to the error code variable
921    
922  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
923  */  */
924    
925  static int  static int
926  get_ucp(const uschar **ptrptr, BOOL *negptr, int *errorcodeptr)  get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
927  {  {
928  int c, i, bot, top;  int c, i, bot, top;
929  const uschar *ptr = *ptrptr;  const uschar *ptr = *ptrptr;
930  char name[4];  char name[32];
931    
932  c = *(++ptr);  c = *(++ptr);
933  if (c == 0) goto ERROR_RETURN;  if (c == 0) goto ERROR_RETURN;
934    
935  *negptr = FALSE;  *negptr = FALSE;
936    
937  /* \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
938  preceded by ^ for negation. */  negation. */
939    
940  if (c == '{')  if (c == CHAR_LEFT_CURLY_BRACKET)
941    {    {
942    if (ptr[1] == '^')    if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
943      {      {
944      *negptr = TRUE;      *negptr = TRUE;
945      ptr++;      ptr++;
946      }      }
947    for (i = 0; i <= 2; i++)    for (i = 0; i < (int)sizeof(name) - 1; i++)
948      {      {
949      c = *(++ptr);      c = *(++ptr);
950      if (c == 0) goto ERROR_RETURN;      if (c == 0) goto ERROR_RETURN;
951      if (c == '}') break;      if (c == CHAR_RIGHT_CURLY_BRACKET) break;
952      name[i] = c;      name[i] = c;
953      }      }
954    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;  
     }  
955    name[i] = 0;    name[i] = 0;
956    }    }
957    
# Line 619  top = _pcre_utt_size; Line 972  top = _pcre_utt_size;
972    
973  while (bot < top)  while (bot < top)
974    {    {
975    i = (bot + top)/2;    i = (bot + top) >> 1;
976    c = strcmp(name, _pcre_utt[i].name);    c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
977    if (c == 0) return _pcre_utt[i].value;    if (c == 0)
978        {
979        *dptr = _pcre_utt[i].value;
980        return _pcre_utt[i].type;
981        }
982    if (c > 0) bot = i + 1; else top = i;    if (c > 0) bot = i + 1; else top = i;
983    }    }
984    
 UNKNOWN_RETURN:  
985  *errorcodeptr = ERR47;  *errorcodeptr = ERR47;
986  *ptrptr = ptr;  *ptrptr = ptr;
987  return -1;  return -1;
# Line 660  is_counted_repeat(const uschar *p) Line 1016  is_counted_repeat(const uschar *p)
1016  {  {
1017  if ((digitab[*p++] & ctype_digit) == 0) return FALSE;  if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
1018  while ((digitab[*p] & ctype_digit) != 0) p++;  while ((digitab[*p] & ctype_digit) != 0) p++;
1019  if (*p == '}') return TRUE;  if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
1020    
1021  if (*p++ != ',') return FALSE;  if (*p++ != CHAR_COMMA) return FALSE;
1022  if (*p == '}') return TRUE;  if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
1023    
1024  if ((digitab[*p++] & ctype_digit) == 0) return FALSE;  if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
1025  while ((digitab[*p] & ctype_digit) != 0) p++;  while ((digitab[*p] & ctype_digit) != 0) p++;
1026    
1027  return (*p == '}');  return (*p == CHAR_RIGHT_CURLY_BRACKET);
1028  }  }
1029    
1030    
# Line 701  int max = -1; Line 1057  int max = -1;
1057  /* 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
1058  an integer overflow. */  an integer overflow. */
1059    
1060  while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - '0';  while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - CHAR_0;
1061  if (min < 0 || min > 65535)  if (min < 0 || min > 65535)
1062    {    {
1063    *errorcodeptr = ERR5;    *errorcodeptr = ERR5;
# Line 711  if (min < 0 || min > 65535) Line 1067  if (min < 0 || min > 65535)
1067  /* 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.
1068  Also, max must not be less than min. */  Also, max must not be less than min. */
1069    
1070  if (*p == '}') max = min; else  if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1071    {    {
1072    if (*(++p) != '}')    if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1073      {      {
1074      max = 0;      max = 0;
1075      while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - '0';      while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0;
1076      if (max < 0 || max > 65535)      if (max < 0 || max > 65535)
1077        {        {
1078        *errorcodeptr = ERR5;        *errorcodeptr = ERR5;
# Line 741  return p; Line 1097  return p;
1097    
1098    
1099  /*************************************************  /*************************************************
1100    *  Subroutine for finding forward reference      *
1101    *************************************************/
1102    
1103    /* This recursive function is called only from find_parens() below. The
1104    top-level call starts at the beginning of the pattern. All other calls must
1105    start at a parenthesis. It scans along a pattern's text looking for capturing
1106    subpatterns, and counting them. If it finds a named pattern that matches the
1107    name it is given, it returns its number. Alternatively, if the name is NULL, it
1108    returns when it reaches a given numbered subpattern. Recursion is used to keep
1109    track of subpatterns that reset the capturing group numbers - the (?| feature.
1110    
1111    This function was originally called only from the second pass, in which we know
1112    that if (?< or (?' or (?P< is encountered, the name will be correctly
1113    terminated because that is checked in the first pass. There is now one call to
1114    this function in the first pass, to check for a recursive back reference by
1115    name (so that we can make the whole group atomic). In this case, we need check
1116    only up to the current position in the pattern, and that is still OK because
1117    and previous occurrences will have been checked. To make this work, the test
1118    for "end of pattern" is a check against cd->end_pattern in the main loop,
1119    instead of looking for a binary zero. This means that the special first-pass
1120    call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1121    processing items within the loop are OK, because afterwards the main loop will
1122    terminate.)
1123    
1124    Arguments:
1125      ptrptr       address of the current character pointer (updated)
1126      cd           compile background data
1127      name         name to seek, or NULL if seeking a numbered subpattern
1128      lorn         name length, or subpattern number if name is NULL
1129      xmode        TRUE if we are in /x mode
1130      utf8         TRUE if we are in UTF-8 mode
1131      count        pointer to the current capturing subpattern number (updated)
1132    
1133    Returns:       the number of the named subpattern, or -1 if not found
1134    */
1135    
1136    static int
1137    find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
1138      BOOL xmode, BOOL utf8, int *count)
1139    {
1140    uschar *ptr = *ptrptr;
1141    int start_count = *count;
1142    int hwm_count = start_count;
1143    BOOL dup_parens = FALSE;
1144    
1145    /* If the first character is a parenthesis, check on the type of group we are
1146    dealing with. The very first call may not start with a parenthesis. */
1147    
1148    if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1149      {
1150      /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1151    
1152      if (ptr[1] == CHAR_ASTERISK) ptr += 2;
1153    
1154      /* Handle a normal, unnamed capturing parenthesis. */
1155    
1156      else if (ptr[1] != CHAR_QUESTION_MARK)
1157        {
1158        *count += 1;
1159        if (name == NULL && *count == lorn) return *count;
1160        ptr++;
1161        }
1162    
1163      /* All cases now have (? at the start. Remember when we are in a group
1164      where the parenthesis numbers are duplicated. */
1165    
1166      else if (ptr[2] == CHAR_VERTICAL_LINE)
1167        {
1168        ptr += 3;
1169        dup_parens = TRUE;
1170        }
1171    
1172      /* Handle comments; all characters are allowed until a ket is reached. */
1173    
1174      else if (ptr[2] == CHAR_NUMBER_SIGN)
1175        {
1176        for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1177        goto FAIL_EXIT;
1178        }
1179    
1180      /* Handle a condition. If it is an assertion, just carry on so that it
1181      is processed as normal. If not, skip to the closing parenthesis of the
1182      condition (there can't be any nested parens). */
1183    
1184      else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1185        {
1186        ptr += 2;
1187        if (ptr[1] != CHAR_QUESTION_MARK)
1188          {
1189          while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1190          if (*ptr != 0) ptr++;
1191          }
1192        }
1193    
1194      /* Start with (? but not a condition. */
1195    
1196      else
1197        {
1198        ptr += 2;
1199        if (*ptr == CHAR_P) ptr++;                      /* Allow optional P */
1200    
1201        /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1202    
1203        if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1204            ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1205          {
1206          int term;
1207          const uschar *thisname;
1208          *count += 1;
1209          if (name == NULL && *count == lorn) return *count;
1210          term = *ptr++;
1211          if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1212          thisname = ptr;
1213          while (*ptr != term) ptr++;
1214          if (name != NULL && lorn == ptr - thisname &&
1215              strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1216            return *count;
1217          term++;
1218          }
1219        }
1220      }
1221    
1222    /* Past any initial parenthesis handling, scan for parentheses or vertical
1223    bars. Stop if we get to cd->end_pattern. Note that this is important for the
1224    first-pass call when this value is temporarily adjusted to stop at the current
1225    position. So DO NOT change this to a test for binary zero. */
1226    
1227    for (; ptr < cd->end_pattern; ptr++)
1228      {
1229      /* Skip over backslashed characters and also entire \Q...\E */
1230    
1231      if (*ptr == CHAR_BACKSLASH)
1232        {
1233        if (*(++ptr) == 0) goto FAIL_EXIT;
1234        if (*ptr == CHAR_Q) for (;;)
1235          {
1236          while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1237          if (*ptr == 0) goto FAIL_EXIT;
1238          if (*(++ptr) == CHAR_E) break;
1239          }
1240        continue;
1241        }
1242    
1243      /* Skip over character classes; this logic must be similar to the way they
1244      are handled for real. If the first character is '^', skip it. Also, if the
1245      first few characters (either before or after ^) are \Q\E or \E we skip them
1246      too. This makes for compatibility with Perl. Note the use of STR macros to
1247      encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1248    
1249      if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1250        {
1251        BOOL negate_class = FALSE;
1252        for (;;)
1253          {
1254          if (ptr[1] == CHAR_BACKSLASH)
1255            {
1256            if (ptr[2] == CHAR_E)
1257              ptr+= 2;
1258            else if (strncmp((const char *)ptr+2,
1259                     STR_Q STR_BACKSLASH STR_E, 3) == 0)
1260              ptr += 4;
1261            else
1262              break;
1263            }
1264          else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1265            {
1266            negate_class = TRUE;
1267            ptr++;
1268            }
1269          else break;
1270          }
1271    
1272        /* If the next character is ']', it is a data character that must be
1273        skipped, except in JavaScript compatibility mode. */
1274    
1275        if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1276            (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1277          ptr++;
1278    
1279        while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1280          {
1281          if (*ptr == 0) return -1;
1282          if (*ptr == CHAR_BACKSLASH)
1283            {
1284            if (*(++ptr) == 0) goto FAIL_EXIT;
1285            if (*ptr == CHAR_Q) for (;;)
1286              {
1287              while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1288              if (*ptr == 0) goto FAIL_EXIT;
1289              if (*(++ptr) == CHAR_E) break;
1290              }
1291            continue;
1292            }
1293          }
1294        continue;
1295        }
1296    
1297      /* Skip comments in /x mode */
1298    
1299      if (xmode && *ptr == CHAR_NUMBER_SIGN)
1300        {
1301        ptr++;
1302        while (*ptr != 0)
1303          {
1304          if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1305          ptr++;
1306    #ifdef SUPPORT_UTF8
1307          if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
1308    #endif
1309          }
1310        if (*ptr == 0) goto FAIL_EXIT;
1311        continue;
1312        }
1313    
1314      /* Check for the special metacharacters */
1315    
1316      if (*ptr == CHAR_LEFT_PARENTHESIS)
1317        {
1318        int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, count);
1319        if (rc > 0) return rc;
1320        if (*ptr == 0) goto FAIL_EXIT;
1321        }
1322    
1323      else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1324        {
1325        if (dup_parens && *count < hwm_count) *count = hwm_count;
1326        goto FAIL_EXIT;
1327        }
1328    
1329      else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1330        {
1331        if (*count > hwm_count) hwm_count = *count;
1332        *count = start_count;
1333        }
1334      }
1335    
1336    FAIL_EXIT:
1337    *ptrptr = ptr;
1338    return -1;
1339    }
1340    
1341    
1342    
1343    
1344    /*************************************************
1345    *       Find forward referenced subpattern       *
1346    *************************************************/
1347    
1348    /* This function scans along a pattern's text looking for capturing
1349    subpatterns, and counting them. If it finds a named pattern that matches the
1350    name it is given, it returns its number. Alternatively, if the name is NULL, it
1351    returns when it reaches a given numbered subpattern. This is used for forward
1352    references to subpatterns. We used to be able to start this scan from the
1353    current compiling point, using the current count value from cd->bracount, and
1354    do it all in a single loop, but the addition of the possibility of duplicate
1355    subpattern numbers means that we have to scan from the very start, in order to
1356    take account of such duplicates, and to use a recursive function to keep track
1357    of the different types of group.
1358    
1359    Arguments:
1360      cd           compile background data
1361      name         name to seek, or NULL if seeking a numbered subpattern
1362      lorn         name length, or subpattern number if name is NULL
1363      xmode        TRUE if we are in /x mode
1364      utf8         TRUE if we are in UTF-8 mode
1365    
1366    Returns:       the number of the found subpattern, or -1 if not found
1367    */
1368    
1369    static int
1370    find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode,
1371      BOOL utf8)
1372    {
1373    uschar *ptr = (uschar *)cd->start_pattern;
1374    int count = 0;
1375    int rc;
1376    
1377    /* If the pattern does not start with an opening parenthesis, the first call
1378    to find_parens_sub() will scan right to the end (if necessary). However, if it
1379    does start with a parenthesis, find_parens_sub() will return when it hits the
1380    matching closing parens. That is why we have to have a loop. */
1381    
1382    for (;;)
1383      {
1384      rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, &count);
1385      if (rc > 0 || *ptr++ == 0) break;
1386      }
1387    
1388    return rc;
1389    }
1390    
1391    
1392    
1393    
1394    /*************************************************
1395  *      Find first significant op code            *  *      Find first significant op code            *
1396  *************************************************/  *************************************************/
1397    
1398  /* This is called by several functions that scan a compiled expression looking  /* This is called by several functions that scan a compiled expression looking
1399  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
1400  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
1401  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
1402  assertions, and also the \b assertion; for others it does not.  does not.
1403    
1404  Arguments:  Arguments:
1405    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  
1406    skipassert   TRUE if certain assertions are to be skipped    skipassert   TRUE if certain assertions are to be skipped
1407    
1408  Returns:       pointer to the first significant opcode  Returns:       pointer to the first significant opcode
1409  */  */
1410    
1411  static const uschar*  static const uschar*
1412  first_significant_code(const uschar *code, int *options, int optbit,  first_significant_code(const uschar *code, BOOL skipassert)
   BOOL skipassert)  
1413  {  {
1414  for (;;)  for (;;)
1415    {    {
1416    switch ((int)*code)    switch ((int)*code)
1417      {      {
     case OP_OPT:  
     if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))  
       *options = (int)code[1];  
     code += 2;  
     break;  
   
1418      case OP_ASSERT_NOT:      case OP_ASSERT_NOT:
1419      case OP_ASSERTBACK:      case OP_ASSERTBACK:
1420      case OP_ASSERTBACK_NOT:      case OP_ASSERTBACK_NOT:
# Line 789  for (;;) Line 1430  for (;;)
1430    
1431      case OP_CALLOUT:      case OP_CALLOUT:
1432      case OP_CREF:      case OP_CREF:
1433      case OP_BRANUMBER:      case OP_NCREF:
1434        case OP_RREF:
1435        case OP_NRREF:
1436        case OP_DEF:
1437      code += _pcre_OP_lengths[*code];      code += _pcre_OP_lengths[*code];
1438      break;      break;
1439    
# Line 804  for (;;) Line 1448  for (;;)
1448    
1449    
1450  /*************************************************  /*************************************************
1451  *        Find the fixed length of a pattern      *  *        Find the fixed length of a branch       *
1452  *************************************************/  *************************************************/
1453    
1454  /* 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,
1455  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.
1456  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
1457    temporarily terminated with OP_END when this function is called.
1458    
1459    This function is called when a backward assertion is encountered, so that if it
1460    fails, the error message can point to the correct place in the pattern.
1461    However, we cannot do this when the assertion contains subroutine calls,
1462    because they can be forward references. We solve this by remembering this case
1463    and doing the check at the end; a flag specifies which mode we are running in.
1464    
1465  Arguments:  Arguments:
1466    code     points to the start of the pattern (the bracket)    code     points to the start of the pattern (the bracket)
1467    options  the compiling options    utf8     TRUE in UTF-8 mode
1468      atend    TRUE if called when the pattern is complete
1469      cd       the "compile data" structure
1470    
1471  Returns:   the fixed length, or -1 if there is no fixed length,  Returns:   the fixed length,
1472                 or -1 if there is no fixed length,
1473               or -2 if \C was encountered               or -2 if \C was encountered
1474                 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1475  */  */
1476    
1477  static int  static int
1478  find_fixedlength(uschar *code, int options)  find_fixedlength(uschar *code, BOOL utf8, BOOL atend, compile_data *cd)
1479  {  {
1480  int length = -1;  int length = -1;
1481    
# Line 833  branch, check the length against that of Line 1488  branch, check the length against that of
1488  for (;;)  for (;;)
1489    {    {
1490    int d;    int d;
1491      uschar *ce, *cs;
1492    register int op = *cc;    register int op = *cc;
   if (op >= OP_BRA) op = OP_BRA;  
   
1493    switch (op)    switch (op)
1494      {      {
1495        /* We only need to continue for OP_CBRA (normal capturing bracket) and
1496        OP_BRA (normal non-capturing bracket) because the other variants of these
1497        opcodes are all concerned with unlimited repeated groups, which of course
1498        are not of fixed length. They will cause a -1 response from the default
1499        case of this switch. */
1500    
1501        case OP_CBRA:
1502      case OP_BRA:      case OP_BRA:
1503      case OP_ONCE:      case OP_ONCE:
1504      case OP_COND:      case OP_COND:
1505      d = find_fixedlength(cc, options);      d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), utf8, atend, cd);
1506      if (d < 0) return d;      if (d < 0) return d;
1507      branchlength += d;      branchlength += d;
1508      do cc += GET(cc, 1); while (*cc == OP_ALT);      do cc += GET(cc, 1); while (*cc == OP_ALT);
# Line 850  for (;;) Line 1511  for (;;)
1511    
1512      /* 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
1513      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
1514      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.
1515        Note that we must not include the OP_KETRxxx opcodes here, because they
1516        all imply an unlimited repeat. */
1517    
1518      case OP_ALT:      case OP_ALT:
1519      case OP_KET:      case OP_KET:
     case OP_KETRMAX:  
     case OP_KETRMIN:  
1520      case OP_END:      case OP_END:
1521      if (length < 0) length = branchlength;      if (length < 0) length = branchlength;
1522        else if (length != branchlength) return -1;        else if (length != branchlength) return -1;
# Line 864  for (;;) Line 1525  for (;;)
1525      branchlength = 0;      branchlength = 0;
1526      break;      break;
1527    
1528        /* A true recursion implies not fixed length, but a subroutine call may
1529        be OK. If the subroutine is a forward reference, we can't deal with
1530        it until the end of the pattern, so return -3. */
1531    
1532        case OP_RECURSE:
1533        if (!atend) return -3;
1534        cs = ce = (uschar *)cd->start_code + GET(cc, 1);  /* Start subpattern */
1535        do ce += GET(ce, 1); while (*ce == OP_ALT);       /* End subpattern */
1536        if (cc > cs && cc < ce) return -1;                /* Recursion */
1537        d = find_fixedlength(cs + 2, utf8, atend, cd);
1538        if (d < 0) return d;
1539        branchlength += d;
1540        cc += 1 + LINK_SIZE;
1541        break;
1542    
1543      /* Skip over assertive subpatterns */      /* Skip over assertive subpatterns */
1544    
1545      case OP_ASSERT:      case OP_ASSERT:
# Line 876  for (;;) Line 1552  for (;;)
1552      /* Skip over things that don't match chars */      /* Skip over things that don't match chars */
1553    
1554      case OP_REVERSE:      case OP_REVERSE:
     case OP_BRANUMBER:  
1555      case OP_CREF:      case OP_CREF:
1556      case OP_OPT:      case OP_NCREF:
1557        case OP_RREF:
1558        case OP_NRREF:
1559        case OP_DEF:
1560      case OP_CALLOUT:      case OP_CALLOUT:
1561      case OP_SOD:      case OP_SOD:
1562      case OP_SOM:      case OP_SOM:
1563        case OP_SET_SOM:
1564      case OP_EOD:      case OP_EOD:
1565      case OP_EODN:      case OP_EODN:
1566      case OP_CIRC:      case OP_CIRC:
1567        case OP_CIRCM:
1568      case OP_DOLL:      case OP_DOLL:
1569        case OP_DOLLM:
1570      case OP_NOT_WORD_BOUNDARY:      case OP_NOT_WORD_BOUNDARY:
1571      case OP_WORD_BOUNDARY:      case OP_WORD_BOUNDARY:
1572      cc += _pcre_OP_lengths[*cc];      cc += _pcre_OP_lengths[*cc];
# Line 894  for (;;) Line 1575  for (;;)
1575      /* Handle literal characters */      /* Handle literal characters */
1576    
1577      case OP_CHAR:      case OP_CHAR:
1578      case OP_CHARNC:      case OP_CHARI:
1579        case OP_NOT:
1580        case OP_NOTI:
1581      branchlength++;      branchlength++;
1582      cc += 2;      cc += 2;
1583  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
1584      if ((options & PCRE_UTF8) != 0)      if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
       {  
       while ((*cc & 0xc0) == 0x80) cc++;  
       }  
1585  #endif  #endif
1586      break;      break;
1587    
# Line 912  for (;;) Line 1592  for (;;)
1592      branchlength += GET2(cc,1);      branchlength += GET2(cc,1);
1593      cc += 4;      cc += 4;
1594  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
1595      if ((options & PCRE_UTF8) != 0)      if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
       {  
       while((*cc & 0x80) == 0x80) cc++;  
       }  
1596  #endif  #endif
1597      break;      break;
1598    
1599      case OP_TYPEEXACT:      case OP_TYPEEXACT:
1600      branchlength += GET2(cc,1);      branchlength += GET2(cc,1);
1601        if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1602      cc += 4;      cc += 4;
1603      break;      break;
1604    
# Line 928  for (;;) Line 1606  for (;;)
1606    
1607      case OP_PROP:      case OP_PROP:
1608      case OP_NOTPROP:      case OP_NOTPROP:
1609      cc++;      cc += 2;
1610      /* Fall through */      /* Fall through */
1611    
1612      case OP_NOT_DIGIT:      case OP_NOT_DIGIT:
# Line 938  for (;;) Line 1616  for (;;)
1616      case OP_NOT_WORDCHAR:      case OP_NOT_WORDCHAR:
1617      case OP_WORDCHAR:      case OP_WORDCHAR:
1618      case OP_ANY:      case OP_ANY:
1619        case OP_ALLANY:
1620      branchlength++;      branchlength++;
1621      cc++;      cc++;
1622      break;      break;
# Line 992  for (;;) Line 1671  for (;;)
1671    
1672    
1673  /*************************************************  /*************************************************
1674  *    Scan compiled regex for numbered bracket    *  *    Scan compiled regex for specific bracket    *
1675  *************************************************/  *************************************************/
1676    
1677  /* This little function scans through a compiled pattern until it finds a  /* This little function scans through a compiled pattern until it finds a
1678  capturing bracket with the given number.  capturing bracket with the given number, or, if the number is negative, an
1679    instance of OP_REVERSE for a lookbehind. The function is global in the C sense
1680    so that it can be called from pcre_study() when finding the minimum matching
1681    length.
1682    
1683  Arguments:  Arguments:
1684    code        points to start of expression    code        points to start of expression
1685    utf8        TRUE in UTF-8 mode    utf8        TRUE in UTF-8 mode
1686    number      the required bracket number    number      the required bracket number or negative to find a lookbehind
1687    
1688  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
1689  */  */
1690    
1691  static const uschar *  const uschar *
1692  find_bracket(const uschar *code, BOOL utf8, int number)  _pcre_find_bracket(const uschar *code, BOOL utf8, int number)
1693  {  {
 #ifndef SUPPORT_UTF8  
 utf8 = utf8;               /* Stop pedantic compilers complaining */  
 #endif  
   
1694  for (;;)  for (;;)
1695    {    {
1696    register int c = *code;    register int c = *code;
1697    if (c == OP_END) return NULL;    if (c == OP_END) return NULL;
1698    else if (c > OP_BRA)  
1699      /* XCLASS is used for classes that cannot be represented just by a bit
1700      map. This includes negated single high-valued characters. The length in
1701      the table is zero; the actual length is stored in the compiled code. */
1702    
1703      if (c == OP_XCLASS) code += GET(code, 1);
1704    
1705      /* Handle recursion */
1706    
1707      else if (c == OP_REVERSE)
1708        {
1709        if (number < 0) return (uschar *)code;
1710        code += _pcre_OP_lengths[c];
1711        }
1712    
1713      /* Handle capturing bracket */
1714    
1715      else if (c == OP_CBRA || c == OP_SCBRA ||
1716               c == OP_CBRAPOS || c == OP_SCBRAPOS)
1717      {      {
1718      int n = c - OP_BRA;      int n = GET2(code, 1+LINK_SIZE);
     if (n > EXTRACT_BASIC_MAX) n = GET2(code, 2+LINK_SIZE);  
1719      if (n == number) return (uschar *)code;      if (n == number) return (uschar *)code;
1720      code += _pcre_OP_lengths[OP_BRA];      code += _pcre_OP_lengths[c];
1721      }      }
1722    
1723      /* Otherwise, we can get the item's length from the table, except that for
1724      repeated character types, we have to test for \p and \P, which have an extra
1725      two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1726      must add in its length. */
1727    
1728    else    else
1729      {      {
1730      code += _pcre_OP_lengths[c];      switch(c)
1731          {
1732          case OP_TYPESTAR:
1733          case OP_TYPEMINSTAR:
1734          case OP_TYPEPLUS:
1735          case OP_TYPEMINPLUS:
1736          case OP_TYPEQUERY:
1737          case OP_TYPEMINQUERY:
1738          case OP_TYPEPOSSTAR:
1739          case OP_TYPEPOSPLUS:
1740          case OP_TYPEPOSQUERY:
1741          if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1742          break;
1743    
1744  #ifdef SUPPORT_UTF8        case OP_TYPEUPTO:
1745          case OP_TYPEMINUPTO:
1746          case OP_TYPEEXACT:
1747          case OP_TYPEPOSUPTO:
1748          if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1749          break;
1750    
1751      /* In UTF-8 mode, opcodes that are followed by a character may be followed        case OP_MARK:
1752      by a multi-byte character. The length in the table is a minimum, so we have        case OP_PRUNE_ARG:
1753      to scan along to skip the extra bytes. All opcodes are less than 128, so we        case OP_SKIP_ARG:
1754      can use relatively efficient code. */        code += code[1];
1755          break;
1756    
1757          case OP_THEN_ARG:
1758          code += code[1+LINK_SIZE];
1759          break;
1760          }
1761    
1762        /* Add in the fixed length from the table */
1763    
1764        code += _pcre_OP_lengths[c];
1765    
1766      /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1767      a multi-byte character. The length in the table is a minimum, so we have to
1768      arrange to skip the extra bytes. */
1769    
1770    #ifdef SUPPORT_UTF8
1771      if (utf8) switch(c)      if (utf8) switch(c)
1772        {        {
1773        case OP_CHAR:        case OP_CHAR:
1774        case OP_CHARNC:        case OP_CHARI:
1775        case OP_EXACT:        case OP_EXACT:
1776          case OP_EXACTI:
1777        case OP_UPTO:        case OP_UPTO:
1778          case OP_UPTOI:
1779        case OP_MINUPTO:        case OP_MINUPTO:
1780          case OP_MINUPTOI:
1781          case OP_POSUPTO:
1782          case OP_POSUPTOI:
1783        case OP_STAR:        case OP_STAR:
1784          case OP_STARI:
1785        case OP_MINSTAR:        case OP_MINSTAR:
1786          case OP_MINSTARI:
1787          case OP_POSSTAR:
1788          case OP_POSSTARI:
1789        case OP_PLUS:        case OP_PLUS:
1790          case OP_PLUSI:
1791        case OP_MINPLUS:        case OP_MINPLUS:
1792          case OP_MINPLUSI:
1793          case OP_POSPLUS:
1794          case OP_POSPLUSI:
1795        case OP_QUERY:        case OP_QUERY:
1796          case OP_QUERYI:
1797        case OP_MINQUERY:        case OP_MINQUERY:
1798        while ((*code & 0xc0) == 0x80) code++;        case OP_MINQUERYI:
1799        break;        case OP_POSQUERY:
1800          case OP_POSQUERYI:
1801        /* 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;  
1802        break;        break;
1803        }        }
1804    #else
1805        (void)(utf8);  /* Keep compiler happy by referencing function argument */
1806  #endif  #endif
1807      }      }
1808    }    }
# Line 1083  Returns:      pointer to the opcode for Line 1827  Returns:      pointer to the opcode for
1827  static const uschar *  static const uschar *
1828  find_recurse(const uschar *code, BOOL utf8)  find_recurse(const uschar *code, BOOL utf8)
1829  {  {
 #ifndef SUPPORT_UTF8  
 utf8 = utf8;               /* Stop pedantic compilers complaining */  
 #endif  
   
1830  for (;;)  for (;;)
1831    {    {
1832    register int c = *code;    register int c = *code;
1833    if (c == OP_END) return NULL;    if (c == OP_END) return NULL;
1834    else if (c == OP_RECURSE) return code;    if (c == OP_RECURSE) return code;
1835    else if (c > OP_BRA)  
1836      {    /* XCLASS is used for classes that cannot be represented just by a bit
1837      code += _pcre_OP_lengths[OP_BRA];    map. This includes negated single high-valued characters. The length in
1838      }    the table is zero; the actual length is stored in the compiled code. */
1839    
1840      if (c == OP_XCLASS) code += GET(code, 1);
1841    
1842      /* Otherwise, we can get the item's length from the table, except that for
1843      repeated character types, we have to test for \p and \P, which have an extra
1844      two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1845      must add in its length. */
1846    
1847    else    else
1848      {      {
1849      code += _pcre_OP_lengths[c];      switch(c)
1850          {
1851          case OP_TYPESTAR:
1852          case OP_TYPEMINSTAR:
1853          case OP_TYPEPLUS:
1854          case OP_TYPEMINPLUS:
1855          case OP_TYPEQUERY:
1856          case OP_TYPEMINQUERY:
1857          case OP_TYPEPOSSTAR:
1858          case OP_TYPEPOSPLUS:
1859          case OP_TYPEPOSQUERY:
1860          if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1861          break;
1862    
1863  #ifdef SUPPORT_UTF8        case OP_TYPEPOSUPTO:
1864          case OP_TYPEUPTO:
1865          case OP_TYPEMINUPTO:
1866          case OP_TYPEEXACT:
1867          if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1868          break;
1869    
1870          case OP_MARK:
1871          case OP_PRUNE_ARG:
1872          case OP_SKIP_ARG:
1873          code += code[1];
1874          break;
1875    
1876          case OP_THEN_ARG:
1877          code += code[1+LINK_SIZE];
1878          break;
1879          }
1880    
1881        /* Add in the fixed length from the table */
1882    
1883        code += _pcre_OP_lengths[c];
1884    
1885      /* 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
1886      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
1887      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. */  
1888    
1889    #ifdef SUPPORT_UTF8
1890      if (utf8) switch(c)      if (utf8) switch(c)
1891        {        {
1892        case OP_CHAR:        case OP_CHAR:
1893        case OP_CHARNC:        case OP_CHARI:
1894        case OP_EXACT:        case OP_EXACT:
1895          case OP_EXACTI:
1896        case OP_UPTO:        case OP_UPTO:
1897          case OP_UPTOI:
1898        case OP_MINUPTO:        case OP_MINUPTO:
1899          case OP_MINUPTOI:
1900          case OP_POSUPTO:
1901          case OP_POSUPTOI:
1902        case OP_STAR:        case OP_STAR:
1903          case OP_STARI:
1904        case OP_MINSTAR:        case OP_MINSTAR:
1905          case OP_MINSTARI:
1906          case OP_POSSTAR:
1907          case OP_POSSTARI:
1908        case OP_PLUS:        case OP_PLUS:
1909          case OP_PLUSI:
1910        case OP_MINPLUS:        case OP_MINPLUS:
1911          case OP_MINPLUSI:
1912          case OP_POSPLUS:
1913          case OP_POSPLUSI:
1914        case OP_QUERY:        case OP_QUERY:
1915          case OP_QUERYI:
1916        case OP_MINQUERY:        case OP_MINQUERY:
1917        while ((*code & 0xc0) == 0x80) code++;        case OP_MINQUERYI:
1918        break;        case OP_POSQUERY:
1919          case OP_POSQUERYI:
1920        /* 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;  
1921        break;        break;
1922        }        }
1923    #else
1924        (void)(utf8);  /* Keep compiler happy by referencing function argument */
1925  #endif  #endif
1926      }      }
1927    }    }
# Line 1143  for (;;) Line 1934  for (;;)
1934  *************************************************/  *************************************************/
1935    
1936  /* 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
1937  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()
1938  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
1939  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
1940  whose current branch will already have been scanned.  backward and negative forward assertions when its final argument is TRUE. If we
1941    hit an unclosed bracket, we return "empty" - this means we've struck an inner
1942    bracket whose current branch will already have been scanned.
1943    
1944  Arguments:  Arguments:
1945    code        points to start of search    code        points to start of search
1946    endcode     points to where to stop    endcode     points to where to stop
1947    utf8        TRUE if in UTF8 mode    utf8        TRUE if in UTF8 mode
1948      cd          contains pointers to tables etc.
1949    
1950  Returns:      TRUE if what is matched could be empty  Returns:      TRUE if what is matched could be empty
1951  */  */
1952    
1953  static BOOL  static BOOL
1954  could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8)  could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8,
1955      compile_data *cd)
1956  {  {
1957  register int c;  register int c;
1958  for (code = first_significant_code(code + 1 + LINK_SIZE, NULL, 0, TRUE);  for (code = first_significant_code(code + _pcre_OP_lengths[*code], TRUE);
1959       code < endcode;       code < endcode;
1960       code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))       code = first_significant_code(code + _pcre_OP_lengths[c], TRUE))
1961    {    {
1962    const uschar *ccode;    const uschar *ccode;
1963    
1964    c = *code;    c = *code;
1965    
1966    if (c >= OP_BRA)    /* Skip over forward assertions; the other assertions are skipped by
1967      first_significant_code() with a TRUE final argument. */
1968    
1969      if (c == OP_ASSERT)
1970      {      {
1971      BOOL empty_branch;      do code += GET(code, 1); while (*code == OP_ALT);
1972      if (GET(code, 1) == 0) return TRUE;    /* Hit unclosed bracket */      c = *code;
1973        continue;
1974        }
1975    
1976      /* Scan a closed bracket */    /* For a recursion/subroutine call, if its end has been reached, which
1977      implies a subroutine call, we can scan it. */
1978    
1979      empty_branch = FALSE;    if (c == OP_RECURSE)
1980        {
1981        BOOL empty_branch = FALSE;
1982        const uschar *scode = cd->start_code + GET(code, 1);
1983        if (GET(scode, 1) == 0) return TRUE;    /* Unclosed */
1984      do      do
1985        {        {
1986        if (!empty_branch && could_be_empty_branch(code, endcode, utf8))        if (could_be_empty_branch(scode, endcode, utf8, cd))
1987            {
1988          empty_branch = TRUE;          empty_branch = TRUE;
1989            break;
1990            }
1991          scode += GET(scode, 1);
1992          }
1993        while (*scode == OP_ALT);
1994        if (!empty_branch) return FALSE;  /* All branches are non-empty */
1995        continue;
1996        }
1997    
1998      /* Groups with zero repeats can of course be empty; skip them. */
1999    
2000      if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2001          c == OP_BRAPOSZERO)
2002        {
2003        code += _pcre_OP_lengths[c];
2004        do code += GET(code, 1); while (*code == OP_ALT);
2005        c = *code;
2006        continue;
2007        }
2008    
2009      /* A nested group that is already marked as "could be empty" can just be
2010      skipped. */
2011    
2012      if (c == OP_SBRA  || c == OP_SBRAPOS ||
2013          c == OP_SCBRA || c == OP_SCBRAPOS)
2014        {
2015        do code += GET(code, 1); while (*code == OP_ALT);
2016        c = *code;
2017        continue;
2018        }
2019    
2020      /* For other groups, scan the branches. */
2021    
2022      if (c == OP_BRA  || c == OP_BRAPOS ||
2023          c == OP_CBRA || c == OP_CBRAPOS ||
2024          c == OP_ONCE || c == OP_COND)
2025        {
2026        BOOL empty_branch;
2027        if (GET(code, 1) == 0) return TRUE;    /* Hit unclosed bracket */
2028    
2029        /* If a conditional group has only one branch, there is a second, implied,
2030        empty branch, so just skip over the conditional, because it could be empty.
2031        Otherwise, scan the individual branches of the group. */
2032    
2033        if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2034        code += GET(code, 1);        code += GET(code, 1);
2035        else
2036          {
2037          empty_branch = FALSE;
2038          do
2039            {
2040            if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
2041              empty_branch = TRUE;
2042            code += GET(code, 1);
2043            }
2044          while (*code == OP_ALT);
2045          if (!empty_branch) return FALSE;   /* All branches are non-empty */
2046        }        }
2047      while (*code == OP_ALT);  
     if (!empty_branch) return FALSE;   /* All branches are non-empty */  
     code += 1 + LINK_SIZE;  
2048      c = *code;      c = *code;
2049        continue;
2050      }      }
2051    
2052    else switch (c)    /* Handle the other opcodes */
2053    
2054      switch (c)
2055      {      {
2056      /* Check for quantifiers after a class */      /* Check for quantifiers after a class. XCLASS is used for classes that
2057        cannot be represented just by a bit map. This includes negated single
2058        high-valued characters. The length in _pcre_OP_lengths[] is zero; the
2059        actual length is stored in the compiled code, so we must update "code"
2060        here. */
2061    
2062  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
2063      case OP_XCLASS:      case OP_XCLASS:
2064      ccode = code + GET(code, 1);      ccode = code += GET(code, 1);
2065      goto CHECK_CLASS_REPEAT;      goto CHECK_CLASS_REPEAT;
2066  #endif  #endif
2067    
# Line 1238  for (code = first_significant_code(code Line 2105  for (code = first_significant_code(code
2105      case OP_NOT_WORDCHAR:      case OP_NOT_WORDCHAR:
2106      case OP_WORDCHAR:      case OP_WORDCHAR:
2107      case OP_ANY:      case OP_ANY:
2108        case OP_ALLANY:
2109      case OP_ANYBYTE:      case OP_ANYBYTE:
2110      case OP_CHAR:      case OP_CHAR:
2111      case OP_CHARNC:      case OP_CHARI:
2112      case OP_NOT:      case OP_NOT:
2113        case OP_NOTI:
2114      case OP_PLUS:      case OP_PLUS:
2115      case OP_MINPLUS:      case OP_MINPLUS:
2116        case OP_POSPLUS:
2117      case OP_EXACT:      case OP_EXACT:
2118      case OP_NOTPLUS:      case OP_NOTPLUS:
2119      case OP_NOTMINPLUS:      case OP_NOTMINPLUS:
2120        case OP_NOTPOSPLUS:
2121      case OP_NOTEXACT:      case OP_NOTEXACT:
2122      case OP_TYPEPLUS:      case OP_TYPEPLUS:
2123      case OP_TYPEMINPLUS:      case OP_TYPEMINPLUS:
2124        case OP_TYPEPOSPLUS:
2125      case OP_TYPEEXACT:      case OP_TYPEEXACT:
2126      return FALSE;      return FALSE;
2127    
2128        /* These are going to continue, as they may be empty, but we have to
2129        fudge the length for the \p and \P cases. */
2130    
2131        case OP_TYPESTAR:
2132        case OP_TYPEMINSTAR:
2133        case OP_TYPEPOSSTAR:
2134        case OP_TYPEQUERY:
2135        case OP_TYPEMINQUERY:
2136        case OP_TYPEPOSQUERY:
2137        if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2138        break;
2139    
2140        /* Same for these */
2141    
2142        case OP_TYPEUPTO:
2143        case OP_TYPEMINUPTO:
2144        case OP_TYPEPOSUPTO:
2145        if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
2146        break;
2147    
2148      /* End of branch */      /* End of branch */
2149    
2150      case OP_KET:      case OP_KET:
2151      case OP_KETRMAX:      case OP_KETRMAX:
2152      case OP_KETRMIN:      case OP_KETRMIN:
2153        case OP_KETRPOS:
2154      case OP_ALT:      case OP_ALT:
2155      return TRUE;      return TRUE;
2156    
2157      /* In UTF-8 mode, STAR, MINSTAR, QUERY, MINQUERY, UPTO, and MINUPTO  may be      /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2158      followed by a multibyte character */      MINUPTO, and POSUPTO may be followed by a multibyte character */
2159    
2160  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
2161      case OP_STAR:      case OP_STAR:
2162        case OP_STARI:
2163      case OP_MINSTAR:      case OP_MINSTAR:
2164        case OP_MINSTARI:
2165        case OP_POSSTAR:
2166        case OP_POSSTARI:
2167      case OP_QUERY:      case OP_QUERY:
2168        case OP_QUERYI:
2169      case OP_MINQUERY:      case OP_MINQUERY:
2170        case OP_MINQUERYI:
2171        case OP_POSQUERY:
2172        case OP_POSQUERYI:
2173        if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
2174        break;
2175    
2176      case OP_UPTO:      case OP_UPTO:
2177        case OP_UPTOI:
2178      case OP_MINUPTO:      case OP_MINUPTO:
2179      if (utf8) while ((code[2] & 0xc0) == 0x80) code++;      case OP_MINUPTOI:
2180        case OP_POSUPTO:
2181        case OP_POSUPTOI:
2182        if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
2183      break;      break;
2184  #endif  #endif
2185    
2186        /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2187        string. */
2188    
2189        case OP_MARK:
2190        case OP_PRUNE_ARG:
2191        case OP_SKIP_ARG:
2192        code += code[1];
2193        break;
2194    
2195        case OP_THEN_ARG:
2196        code += code[1+LINK_SIZE];
2197        break;
2198    
2199        /* None of the remaining opcodes are required to match a character. */
2200    
2201        default:
2202        break;
2203      }      }
2204    }    }
2205    
# Line 1296  Arguments: Line 2222  Arguments:
2222    endcode     points to where to stop (current RECURSE item)    endcode     points to where to stop (current RECURSE item)
2223    bcptr       points to the chain of current (unclosed) branch starts    bcptr       points to the chain of current (unclosed) branch starts
2224    utf8        TRUE if in UTF-8 mode    utf8        TRUE if in UTF-8 mode
2225      cd          pointers to tables etc
2226    
2227  Returns:      TRUE if what is matched could be empty  Returns:      TRUE if what is matched could be empty
2228  */  */
2229    
2230  static BOOL  static BOOL
2231  could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,  could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
2232    BOOL utf8)    BOOL utf8, compile_data *cd)
2233  {  {
2234  while (bcptr != NULL && bcptr->current >= code)  while (bcptr != NULL && bcptr->current_branch >= code)
2235    {    {
2236    if (!could_be_empty_branch(bcptr->current, endcode, utf8)) return FALSE;    if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
2237        return FALSE;
2238    bcptr = bcptr->outer;    bcptr = bcptr->outer;
2239    }    }
2240  return TRUE;  return TRUE;
# Line 1319  return TRUE; Line 2247  return TRUE;
2247  *************************************************/  *************************************************/
2248    
2249  /* This function is called when the sequence "[:" or "[." or "[=" is  /* This function is called when the sequence "[:" or "[." or "[=" is
2250  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
2251  optional ^ and then a sequence of letters, terminated by a matching ":]" or  sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2252  ".]" or "=]".  reach an unescaped ']' without the special preceding character, return FALSE.
2253    
2254    Originally, this function only recognized a sequence of letters between the
2255    terminators, but it seems that Perl recognizes any sequence of characters,
2256    though of course unknown POSIX names are subsequently rejected. Perl gives an
2257    "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2258    didn't consider this to be a POSIX class. Likewise for [:1234:].
2259    
2260    The problem in trying to be exactly like Perl is in the handling of escapes. We
2261    have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2262    class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2263    below handles the special case of \], but does not try to do any other escape
2264    processing. This makes it different from Perl for cases such as [:l\ower:]
2265    where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2266    "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2267    I think.
2268    
2269  Argument:  Arguments:
2270    ptr      pointer to the initial [    ptr      pointer to the initial [
2271    endptr   where to return the end pointer    endptr   where to return the end pointer
   cd       pointer to compile data  
2272    
2273  Returns:   TRUE or FALSE  Returns:   TRUE or FALSE
2274  */  */
2275    
2276  static BOOL  static BOOL
2277  check_posix_syntax(const uschar *ptr, const uschar **endptr, compile_data *cd)  check_posix_syntax(const uschar *ptr, const uschar **endptr)
2278  {  {
2279  int terminator;          /* Don't combine these lines; the Solaris cc */  int terminator;          /* Don't combine these lines; the Solaris cc */
2280  terminator = *(++ptr);   /* compiler warns about "non-constant" initializer. */  terminator = *(++ptr);   /* compiler warns about "non-constant" initializer. */
2281  if (*(++ptr) == '^') ptr++;  for (++ptr; *ptr != 0; ptr++)
 while ((cd->ctypes[*ptr] & ctype_letter) != 0) ptr++;  
 if (*ptr == terminator && ptr[1] == ']')  
2282    {    {
2283    *endptr = ptr;    if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) ptr++; else
2284    return TRUE;      {
2285        if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2286        if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2287          {
2288          *endptr = ptr;
2289          return TRUE;
2290          }
2291        }
2292    }    }
2293  return FALSE;  return FALSE;
2294  }  }
# Line 1366  Returns:     a value representing the na Line 2313  Returns:     a value representing the na
2313  static int  static int
2314  check_posix_name(const uschar *ptr, int len)  check_posix_name(const uschar *ptr, int len)
2315  {  {
2316    const char *pn = posix_names;
2317  register int yield = 0;  register int yield = 0;
2318  while (posix_name_lengths[yield] != 0)  while (posix_name_lengths[yield] != 0)
2319    {    {
2320    if (len == posix_name_lengths[yield] &&    if (len == posix_name_lengths[yield] &&
2321      strncmp((const char *)ptr, posix_names[yield], len) == 0) return yield;      strncmp((const char *)ptr, pn, len) == 0) return yield;
2322      pn += posix_name_lengths[yield] + 1;
2323    yield++;    yield++;
2324    }    }
2325  return -1;  return -1;
# Line 1385  return -1; Line 2334  return -1;
2334  that is referenced. This means that groups can be replicated for fixed  that is referenced. This means that groups can be replicated for fixed
2335  repetition simply by copying (because the recursion is allowed to refer to  repetition simply by copying (because the recursion is allowed to refer to
2336  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
2337  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
2338  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
2339  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
2340  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
2341  partially compiled regex must be temporarily terminated with OP_END.  is called, the partially compiled regex must be temporarily terminated with
2342    OP_END.
2343    
2344    This function has been extended with the possibility of forward references for
2345    recursions and subroutine calls. It must also check the list of such references
2346    for the group we are dealing with. If it finds that one of the recursions in
2347    the current group is on this list, it adjusts the offset in the list, not the
2348    value in the reference (which is a group number).
2349    
2350  Arguments:  Arguments:
2351    group      points to the start of the group    group      points to the start of the group
2352    adjust     the amount by which the group is to be moved    adjust     the amount by which the group is to be moved
2353    utf8       TRUE in UTF-8 mode    utf8       TRUE in UTF-8 mode
2354    cd         contains pointers to tables etc.    cd         contains pointers to tables etc.
2355      save_hwm   the hwm forward reference pointer at the start of the group
2356    
2357  Returns:     nothing  Returns:     nothing
2358  */  */
2359    
2360  static void  static void
2361  adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd)  adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
2362      uschar *save_hwm)
2363  {  {
2364  uschar *ptr = group;  uschar *ptr = group;
2365    
2366  while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)  while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
2367    {    {
2368    int offset = GET(ptr, 1);    int offset;
2369    if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);    uschar *hc;
2370    
2371      /* See if this recursion is on the forward reference list. If so, adjust the
2372      reference. */
2373    
2374      for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2375        {
2376        offset = GET(hc, 0);
2377        if (cd->start_code + offset == ptr + 1)
2378          {
2379          PUT(hc, 0, offset + adjust);
2380          break;
2381          }
2382        }
2383    
2384      /* Otherwise, adjust the recursion offset if it's after the start of this
2385      group. */
2386    
2387      if (hc >= cd->hwm)
2388        {
2389        offset = GET(ptr, 1);
2390        if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2391        }
2392    
2393    ptr += 1 + LINK_SIZE;    ptr += 1 + LINK_SIZE;
2394    }    }
2395  }  }
# Line 1434  auto_callout(uschar *code, const uschar Line 2416  auto_callout(uschar *code, const uschar
2416  {  {
2417  *code++ = OP_CALLOUT;  *code++ = OP_CALLOUT;
2418  *code++ = 255;  *code++ = 255;
2419  PUT(code, 0, ptr - cd->start_pattern);  /* Pattern offset */  PUT(code, 0, (int)(ptr - cd->start_pattern));  /* Pattern offset */
2420  PUT(code, LINK_SIZE, 0);                /* Default length */  PUT(code, LINK_SIZE, 0);                       /* Default length */
2421  return code + 2*LINK_SIZE;  return code + 2*LINK_SIZE;
2422  }  }
2423    
# Line 1460  Returns:             nothing Line 2442  Returns:             nothing
2442  static void  static void
2443  complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)  complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2444  {  {
2445  int length = ptr - cd->start_pattern - GET(previous_callout, 2);  int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2446  PUT(previous_callout, 2 + LINK_SIZE, length);  PUT(previous_callout, 2 + LINK_SIZE, length);
2447  }  }
2448    
# Line 1486  Yield:        TRUE when range returned; Line 2468  Yield:        TRUE when range returned;
2468  */  */
2469    
2470  static BOOL  static BOOL
2471  get_othercase_range(int *cptr, int d, int *ocptr, int *odptr)  get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2472      unsigned int *odptr)
2473  {  {
2474  int c, chartype, othercase, next;  unsigned int c, othercase, next;
2475    
2476  for (c = *cptr; c <= d; c++)  for (c = *cptr; c <= d; c++)
2477    {    { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
   if (_pcre_ucp_findchar(c, &chartype, &othercase) == ucp_L && othercase != 0)  
     break;  
   }  
2478    
2479  if (c > d) return FALSE;  if (c > d) return FALSE;
2480    
# Line 1503  next = othercase + 1; Line 2483  next = othercase + 1;
2483    
2484  for (++c; c <= d; c++)  for (++c; c <= d; c++)
2485    {    {
2486    if (_pcre_ucp_findchar(c, &chartype, &othercase) != ucp_L ||    if (UCD_OTHERCASE(c) != next) break;
         othercase != next)  
     break;  
2487    next++;    next++;
2488    }    }
2489    
# Line 1514  for (++c; c <= d; c++) Line 2492  for (++c; c <= d; c++)
2492    
2493  return TRUE;  return TRUE;
2494  }  }
2495  #endif  /* SUPPORT_UCP */  
2496    
2497    
2498  /*************************************************  /*************************************************
2499  *           Compile one branch                   *  *        Check a character and a property        *
2500  *************************************************/  *************************************************/
2501    
2502  /* 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
2503  changed during the branch, the pointer is used to change the external options  is adjacent to a fixed character.
 bits.  
2504    
2505  Arguments:  Arguments:
2506    optionsptr     pointer to the option bits    c            the character
2507    brackets       points to number of extracting brackets used    ptype        the property type
2508    codeptr        points to the pointer to the current code point    pdata        the data for the type
2509    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.  
2510    
2511  Returns:         TRUE on success  Returns:       TRUE if auto-possessifying is OK
                  FALSE, with *errorcodeptr set non-zero on error  
2512  */  */
2513    
2514  static BOOL  static BOOL
2515  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)  
2516  {  {
2517  int repeat_type, op_type;  const ucd_record *prop = GET_UCD(c);
2518  int repeat_min = 0, repeat_max = 0;      /* To please picky compilers */  switch(ptype)
2519  int bravalue = 0;    {
2520  int greedy_default, greedy_non_default;    case PT_LAMP:
2521  int firstbyte, reqbyte;    return (prop->chartype == ucp_Lu ||
2522  int zeroreqbyte, zerofirstbyte;            prop->chartype == ucp_Ll ||
2523  int req_caseopt, reqvary, tempreqvary;            prop->chartype == ucp_Lt) == negated;
2524  int condcount = 0;  
2525  int options = *optionsptr;    case PT_GC:
2526  int after_manual_callout = 0;    return (pdata == _pcre_ucp_gentype[prop->chartype]) == negated;
2527  register int c;  
2528  register uschar *code = *codeptr;    case PT_PC:
2529  uschar *tempcode;    return (pdata == prop->chartype) == negated;
2530  BOOL inescq = FALSE;  
2531  BOOL groupsetfirstbyte = FALSE;    case PT_SC:
2532  const uschar *ptr = *ptrptr;    return (pdata == prop->script) == negated;
2533  const uschar *tempptr;  
2534  uschar *previous = NULL;    /* These are specials */
2535  uschar *previous_callout = NULL;  
2536  uschar classbits[32];    case PT_ALNUM:
2537      return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2538  #ifdef SUPPORT_UTF8            _pcre_ucp_gentype[prop->chartype] == ucp_N) == negated;
2539  BOOL class_utf8;  
2540  BOOL utf8 = (options & PCRE_UTF8) != 0;    case PT_SPACE:    /* Perl space */
2541  uschar *class_utf8data;    return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2542  uschar utf8_char[6];            c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2543  #else            == negated;
2544  BOOL utf8 = FALSE;  
2545  #endif    case PT_PXSPACE:  /* POSIX space */
2546      return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2547              c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2548              c == CHAR_FF || c == CHAR_CR)
2549              == negated;
2550    
2551      case PT_WORD:
2552      return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2553              _pcre_ucp_gentype[prop->chartype] == ucp_N ||
2554              c == CHAR_UNDERSCORE) == negated;
2555      }
2556    return FALSE;
2557    }
2558    #endif  /* SUPPORT_UCP */
2559    
 /* Set up the default and non-default settings for greediness */  
2560    
 greedy_default = ((options & PCRE_UNGREEDY) != 0);  
 greedy_non_default = greedy_default ^ 1;  
2561    
2562  /* Initialize no first byte, no required byte. REQ_UNSET means "no char  /*************************************************
2563  matching encountered yet". It gets changed to REQ_NONE if we hit something that  *     Check if auto-possessifying is possible    *
2564  matches a non-fixed char first char; reqbyte just remains unset if we never  *************************************************/
 find one.  
2565    
2566  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
2567  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
2568  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. */  
2569    
2570  firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;  Arguments:
2571      previous      pointer to the repeated opcode
2572      utf8          TRUE in UTF-8 mode
2573      ptr           next character in pattern
2574      options       options bits
2575      cd            contains pointers to tables etc.
2576    
2577  /* The variable req_caseopt contains either the REQ_CASELESS value or zero,  Returns:        TRUE if possessifying is wanted
2578  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. */  
2579    
2580  req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;  static BOOL
2581    check_auto_possessive(const uschar *previous, BOOL utf8, const uschar *ptr,
2582      int options, compile_data *cd)
2583    {
2584    int c, next;
2585    int op_code = *previous++;
2586    
2587  /* Switch on next character until the end of the branch */  /* Skip whitespace and comments in extended mode */
2588    
2589  for (;; ptr++)  if ((options & PCRE_EXTENDED) != 0)
2590    {    {
2591    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)  
2592      {      {
2593      if (c == '\\' && ptr[1] == 'E')      while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2594        if (*ptr == CHAR_NUMBER_SIGN)
2595        {        {
       inescq = FALSE;  
2596        ptr++;        ptr++;
2597        continue;        while (*ptr != 0)
       }  
     else  
       {  
       if (previous_callout != NULL)  
2598          {          {
2599          complete_callout(previous_callout, ptr, cd);          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2600          previous_callout = NULL;          ptr++;
2601          }  #ifdef SUPPORT_UTF8
2602        if ((options & PCRE_AUTO_CALLOUT) != 0)          if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2603          {  #endif
         previous_callout = code;  
         code = auto_callout(code, ptr, cd);  
2604          }          }
       goto NORMAL_CHAR;  
2605        }        }
2606        else break;
2607      }      }
2608      }
2609    
2610    /* 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
2611    a quantifier. */  value is a character, a negative value is an escape value. */
2612    
2613    is_quantifier = c == '*' || c == '+' || c == '?' ||  if (*ptr == CHAR_BACKSLASH)
2614      (c == '{' && is_counted_repeat(ptr+1));    {
2615      int temperrorcode = 0;
2616      next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2617      if (temperrorcode != 0) return FALSE;
2618      ptr++;    /* Point after the escape sequence */
2619      }
2620    
2621    if (!is_quantifier && previous_callout != NULL &&  else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2622         after_manual_callout-- <= 0)    {
2623      {  #ifdef SUPPORT_UTF8
2624      complete_callout(previous_callout, ptr, cd);    if (utf8) { GETCHARINC(next, ptr); } else
2625      previous_callout = NULL;  #endif
2626      }    next = *ptr++;
2627      }
2628    
2629    /* In extended mode, skip white space and comments */  else return FALSE;
2630    
2631    if ((options & PCRE_EXTENDED) != 0)  /* Skip whitespace and comments in extended mode */
2632    
2633    if ((options & PCRE_EXTENDED) != 0)
2634      {
2635      for (;;)
2636      {      {
2637      if ((cd->ctypes[c] & ctype_space) != 0) continue;      while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2638      if (c == '#')      if (*ptr == CHAR_NUMBER_SIGN)
2639        {        {
2640        /* The space before the ; is to avoid a warning on a silly compiler        ptr++;
2641        on the Macintosh. */        while (*ptr != 0)
2642        while ((c = *(++ptr)) != 0 && c != NEWLINE) ;          {
2643        if (c != 0) continue;   /* Else fall through to handle end of string */          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2644            ptr++;
2645    #ifdef SUPPORT_UTF8
2646            if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2647    #endif
2648            }
2649        }        }
2650        else break;
2651      }      }
2652      }
2653    
2654    /* No auto callout for quantifiers. */  /* If the next thing is itself optional, we have to give up. */
2655    
2656    if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)  if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2657      {    strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2658      previous_callout = code;      return FALSE;
     code = auto_callout(code, ptr, cd);  
     }  
2659    
2660    switch(c)  /* Now compare the next item with the previous opcode. First, handle cases when
2661      {  the next item is a character. */
     /* The branch terminates at end of string, |, or ). */  
2662    
2663      case 0:  if (next >= 0) switch(op_code)
2664      case '|':    {
2665      case ')':    case OP_CHAR:
2666      *firstbyteptr = firstbyte;  #ifdef SUPPORT_UTF8
2667      *reqbyteptr = reqbyte;    GETCHARTEST(c, previous);
2668      *codeptr = code;  #else
2669      *ptrptr = ptr;    c = *previous;
2670      return TRUE;  #endif
2671      return c != next;
2672    
2673      /* Handle single-character metacharacters. In multiline mode, ^ disables    /* For CHARI (caseless character) we must check the other case. If we have
2674      the setting of any following char as a first character. */    Unicode property support, we can use it to test the other case of
2675      high-valued characters. */
2676    
2677      case '^':    case OP_CHARI:
2678      if ((options & PCRE_MULTILINE) != 0)  #ifdef SUPPORT_UTF8
2679        {    GETCHARTEST(c, previous);
2680        if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;  #else
2681        }    c = *previous;
2682      previous = NULL;  #endif
2683      *code++ = OP_CIRC;    if (c == next) return FALSE;
2684      break;  #ifdef SUPPORT_UTF8
2685      if (utf8)
2686        {
2687        unsigned int othercase;
2688        if (next < 128) othercase = cd->fcc[next]; else
2689    #ifdef SUPPORT_UCP
2690        othercase = UCD_OTHERCASE((unsigned int)next);
2691    #else
2692        othercase = NOTACHAR;
2693    #endif
2694        return (unsigned int)c != othercase;
2695        }
2696      else
2697    #endif  /* SUPPORT_UTF8 */
2698      return (c != cd->fcc[next]);  /* Non-UTF-8 mode */
2699    
2700      case '$':    /* For OP_NOT and OP_NOTI, the data is always a single-byte character. These
2701      previous = NULL;    opcodes are not used for multi-byte characters, because they are coded using
2702      *code++ = OP_DOLL;    an XCLASS instead. */
     break;  
2703    
2704      /* There can never be a first char if '.' is first, whatever happens about    case OP_NOT:
2705      repeats. The value of reqbyte doesn't change either. */    return (c = *previous) == next;
2706    
2707      case '.':    case OP_NOTI:
2708      if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;    if ((c = *previous) == next) return TRUE;
2709      zerofirstbyte = firstbyte;  #ifdef SUPPORT_UTF8
2710      zeroreqbyte = reqbyte;    if (utf8)
2711      previous = code;      {
2712      *code++ = OP_ANY;      unsigned int othercase;
2713      break;      if (next < 128) othercase = cd->fcc[next]; else
2714    #ifdef SUPPORT_UCP
2715        othercase = UCD_OTHERCASE(next);
2716    #else
2717        othercase = NOTACHAR;
2718    #endif
2719        return (unsigned int)c == othercase;
2720        }
2721      else
2722    #endif  /* SUPPORT_UTF8 */
2723      return (c == cd->fcc[next]);  /* Non-UTF-8 mode */
2724    
2725      /* 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.
2726      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.  
2727    
2728      If the class contains characters outside the 0-255 range, a different    case OP_DIGIT:
2729      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.  
     */  
2730    
2731      case '[':    case OP_NOT_DIGIT:
2732      previous = code;    return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2733    
2734      /* PCRE supports POSIX class stuff inside a class. Perl gives an error if    case OP_WHITESPACE:
2735      they are encountered at the top level, so we'll do that too. */    return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2736    
2737      if ((ptr[1] == ':' || ptr[1] == '.' || ptr[1] == '=') &&    case OP_NOT_WHITESPACE:
2738          check_posix_syntax(ptr, &tempptr, cd))    return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
       {  
       *errorcodeptr = (ptr[1] == ':')? ERR13 : ERR31;  
       goto FAILED;  
       }  
2739    
2740      /* If the first character is '^', set the negation flag and skip it. */    case OP_WORDCHAR:
2741      return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2742    
2743      if ((c = *(++ptr)) == '^')    case OP_NOT_WORDCHAR:
2744        {    return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
       negate_class = TRUE;  
       c = *(++ptr);  
       }  
     else  
       {  
       negate_class = FALSE;  
       }  
2745    
2746      /* Keep a count of chars with values < 256 so that we can optimize the case    case OP_HSPACE:
2747      of just a single character (as long as it's < 256). For higher valued UTF-8    case OP_NOT_HSPACE:
2748      characters, we don't yet do any optimization. */    switch(next)
2749        {
2750        case 0x09:
2751        case 0x20:
2752        case 0xa0:
2753        case 0x1680:
2754        case 0x180e:
2755        case 0x2000:
2756        case 0x2001:
2757        case 0x2002:
2758        case 0x2003:
2759        case 0x2004:
2760        case 0x2005:
2761        case 0x2006:
2762        case 0x2007:
2763        case 0x2008:
2764        case 0x2009:
2765        case 0x200A:
2766        case 0x202f:
2767        case 0x205f:
2768        case 0x3000:
2769        return op_code == OP_NOT_HSPACE;
2770        default:
2771        return op_code != OP_NOT_HSPACE;
2772        }
2773    
2774      class_charcount = 0;    case OP_ANYNL:
2775      class_lastchar = -1;    case OP_VSPACE:
2776      case OP_NOT_VSPACE:
2777      switch(next)
2778        {
2779        case 0x0a:
2780        case 0x0b:
2781        case 0x0c:
2782        case 0x0d:
2783        case 0x85:
2784        case 0x2028:
2785        case 0x2029:
2786        return op_code == OP_NOT_VSPACE;
2787        default:
2788        return op_code != OP_NOT_VSPACE;
2789        }
2790    
2791  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UCP
2792      class_utf8 = FALSE;                       /* No chars >= 256 */    case OP_PROP:
2793      class_utf8data = code + LINK_SIZE + 34;   /* For UTF-8 items */    return check_char_prop(next, previous[0], previous[1], FALSE);
2794    
2795      case OP_NOTPROP:
2796      return check_char_prop(next, previous[0], previous[1], TRUE);
2797  #endif  #endif
2798    
2799      /* Initialize the 32-char bit map to all zeros. We have to build the    default:
2800      map in a temporary bit of store, in case the class contains only 1    return FALSE;
2801      character (< 256), because in that case the compiled code doesn't use the    }
     bit map. */  
2802    
     memset(classbits, 0, 32 * sizeof(uschar));  
2803    
2804      /* 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
2805      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
2806      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
2807      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
2808      character. */  replaced by OP_PROP codes when PCRE_UCP is set. */
2809    
2810      do  switch(op_code)
2811        {    {
2812      case OP_CHAR:
2813      case OP_CHARI:
2814  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
2815        if (utf8 && c > 127)    GETCHARTEST(c, previous);
2816          {                           /* Braces are required because the */  #else
2817          GETCHARLEN(c, ptr, ptr);    /* macro generates multiple statements */    c = *previous;
         }  
2818  #endif  #endif
2819      switch(-next)
2820        {
2821        case ESC_d:
2822        return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
2823    
2824        /* Inside \Q...\E everything is literal except \E */      case ESC_D:
2825        return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
2826    
2827        if (inescq)      case ESC_s:
2828          {      return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
         if (c == '\\' && ptr[1] == 'E')  
           {  
           inescq = FALSE;  
           ptr++;  
           continue;  
           }  
         else goto LONE_SINGLE_CHARACTER;  
         }  
2829    
2830        /* Handle POSIX class names. Perl allows a negation extension of the      case ESC_S:
2831        form [:^name:]. A square bracket that doesn't match the syntax is      return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
2832        treated as a literal. We also recognize the POSIX constructions  
2833        [.ch.] and [=ch=] ("collating elements") and fault them, as Perl      case ESC_w:
2834        5.6 and 5.8 do. */      return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
2835    
2836        case ESC_W:
2837        return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
2838    
2839        case ESC_h:
2840        case ESC_H:
2841        switch(c)
2842          {
2843          case 0x09:
2844          case 0x20:
2845          case 0xa0:
2846          case 0x1680:
2847          case 0x180e:
2848          case 0x2000:
2849          case 0x2001:
2850          case 0x2002:
2851          case 0x2003:
2852          case 0x2004:
2853          case 0x2005:
2854          case 0x2006:
2855          case 0x2007:
2856          case 0x2008:
2857          case 0x2009:
2858          case 0x200A:
2859          case 0x202f:
2860          case 0x205f:
2861          case 0x3000:
2862          return -next != ESC_h;
2863          default:
2864          return -next == ESC_h;
2865          }
2866    
2867        if (c == '[' &&      case ESC_v:
2868            (ptr[1] == ':' || ptr[1] == '.' || ptr[1] == '=') &&      case ESC_V:
2869            check_posix_syntax(ptr, &tempptr, cd))      switch(c)
2870          {        {
2871          BOOL local_negate = FALSE;        case 0x0a:
2872          int posix_class, i;        case 0x0b:
2873          register const uschar *cbits = cd->cbits;        case 0x0c:
2874          case 0x0d:
2875          case 0x85:
2876          case 0x2028:
2877          case 0x2029:
2878          return -next != ESC_v;
2879          default:
2880          return -next == ESC_v;
2881          }
2882    
2883          if (ptr[1] != ':')      /* When PCRE_UCP is set, these values get generated for \d etc. Find
2884            {      their substitutions and process them. The result will always be either
2885            *errorcodeptr = ERR31;      -ESC_p or -ESC_P. Then fall through to process those values. */
           goto FAILED;  
           }  
2886    
2887          ptr += 2;  #ifdef SUPPORT_UCP
2888          if (*ptr == '^')      case ESC_du:
2889            {      case ESC_DU:
2890            local_negate = TRUE;      case ESC_wu:
2891            ptr++;      case ESC_WU:
2892            }      case ESC_su:
2893        case ESC_SU:
2894          {
2895          int temperrorcode = 0;
2896          ptr = substitutes[-next - ESC_DU];
2897          next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
2898          if (temperrorcode != 0) return FALSE;
2899          ptr++;    /* For compatibility */
2900          }
2901        /* Fall through */
2902    
2903          posix_class = check_posix_name(ptr, tempptr - ptr);      case ESC_p:
2904          if (posix_class < 0)      case ESC_P:
2905            {        {
2906            *errorcodeptr = ERR30;        int ptype, pdata, errorcodeptr;
2907            goto FAILED;        BOOL negated;
           }  
2908    
2909          /* If matching is caseless, upper and lower are converted to        ptr--;      /* Make ptr point at the p or P */
2910          alpha. This relies on the fact that the class table starts with        ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
2911          alpha, lower, upper as the first 3 entries. */        if (ptype < 0) return FALSE;
2912          ptr++;      /* Point past the final curly ket */
2913    
2914          /* If the property item is optional, we have to give up. (When generated
2915          from \d etc by PCRE_UCP, this test will have been applied much earlier,
2916          to the original \d etc. At this point, ptr will point to a zero byte. */
2917    
2918          if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2919            strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2920              return FALSE;
2921    
2922          if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)        /* Do the property check. */
           posix_class = 0;  
2923    
2924          /* 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);
2925          tables, or their negations. The [:blank:] class sets up the same        }
2926          chars as the [:space:] class (all white space). We remove the vertical  #endif
         white space chars afterwards. */  
2927    
2928          posix_class *= 3;      default:
2929          for (i = 0; i < 3; i++)      return FALSE;
2930            {      }
           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;  
             }  
           }  
2931    
2932          ptr = tempptr + 1;    /* In principle, support for Unicode properties should be integrated here as
2933          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
2934          continue;    /* End of POSIX syntax handling */    values before switching on the op-code. However, I wonder how many patterns
2935          }    combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
2936      these op-codes are never generated.) */
2937    
2938        /* Backslash may introduce a single character, or it may introduce one    case OP_DIGIT:
2939        of the specials, which just set a flag. Escaped items are checked for    return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
2940        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. */  
2941    
2942        if (c == '\\')    case OP_NOT_DIGIT:
2943          {    return next == -ESC_d;
         c = check_escape(&ptr, errorcodeptr, *brackets, options, TRUE);  
2944    
2945          if (-c == ESC_b) c = '\b';       /* \b is backslash in a class */    case OP_WHITESPACE:
2946          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;  
           }  
2947    
2948          if (c < 0)    case OP_NOT_WHITESPACE:
2949            {    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;  
2950    
2951              case ESC_D:    case OP_HSPACE:
2952              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];    return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
2953              continue;           next == -ESC_w || next == -ESC_v || next == -ESC_R;
2954    
2955              case ESC_w:    case OP_NOT_HSPACE:
2956              for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];    return next == -ESC_h;
             continue;  
2957    
2958              case ESC_W:    /* Can't have \S in here because VT matches \S (Perl anomaly) */
2959              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];    case OP_ANYNL:
2960              continue;    case OP_VSPACE:
2961      return next == -ESC_V || next == -ESC_d || next == -ESC_w;
2962    
2963              case ESC_s:    case OP_NOT_VSPACE:
2964              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;  
2965    
2966              case ESC_S:    case OP_WORDCHAR:
2967              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];    return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
2968              classbits[1] |= 0x08;    /* Perl 5.004 onwards omits VT from \s */           next == -ESC_v || next == -ESC_R;
             continue;  
2969    
2970  #ifdef SUPPORT_UCP    case OP_NOT_WORDCHAR:
2971              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  
2972    
2973              /* Unrecognized escapes are faulted if PCRE is running in its    default:
2974              strict mode. By default, for compatibility with Perl, they are    return FALSE;
2975              treated as literals. */    }
2976    
2977              default:  /* Control does not reach here */
2978              if ((options & PCRE_EXTRA) != 0)  }
               {  
               *errorcodeptr = ERR7;  
               goto FAILED;  
               }  
             c = *ptr;              /* The final character */  
             class_charcount -= 2;  /* Undo the default count from above */  
             }  
           }  
2979    
         /* Fall through if we have a single character (c >= 0). This may be  
         > 256 in UTF-8 mode. */  
2980    
         }   /* End of backslash handling */  
2981    
2982        /* A single character may be followed by '-' to form a range. However,  /*************************************************
2983        Perl does not permit ']' to be the end of the range. A '-' character  *           Compile one branch                   *
2984        here is treated as a literal. */  *************************************************/
2985    
2986        if (ptr[1] == '-' && ptr[2] != ']')  /* Scan the pattern, compiling it into the a vector. If the options are
2987          {  changed during the branch, the pointer is used to change the external options
2988          int d;  bits. This function is used during the pre-compile phase when we are trying
2989          ptr += 2;  to find out the amount of memory needed, as well as during the real compile
2990    phase. The value of lengthptr distinguishes the two phases.
2991    
2992    Arguments:
2993      optionsptr     pointer to the option bits
2994      codeptr        points to the pointer to the current code point
2995      ptrptr         points to the current pattern pointer
2996      errorcodeptr   points to error code variable
2997      firstbyteptr   set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
2998      reqbyteptr     set to the last literal character required, else < 0
2999      bcptr          points to current branch chain
3000      cd             contains pointers to tables etc.
3001      lengthptr      NULL during the real compile phase
3002                     points to length accumulator during pre-compile phase
3003    
3004    Returns:         TRUE on success
3005                     FALSE, with *errorcodeptr set non-zero on error
3006    */
3007    
3008    static BOOL
3009    compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
3010      int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
3011      compile_data *cd, int *lengthptr)
3012    {
3013    int repeat_type, op_type;
3014    int repeat_min = 0, repeat_max = 0;      /* To please picky compilers */
3015    int bravalue = 0;
3016    int greedy_default, greedy_non_default;
3017    int firstbyte, reqbyte;
3018    int zeroreqbyte, zerofirstbyte;
3019    int req_caseopt, reqvary, tempreqvary;
3020    int options = *optionsptr;
3021    int after_manual_callout = 0;
3022    int length_prevgroup = 0;
3023    register int c;
3024    register uschar *code = *codeptr;
3025    uschar *last_code = code;
3026    uschar *orig_code = code;
3027    uschar *tempcode;
3028    BOOL inescq = FALSE;
3029    BOOL groupsetfirstbyte = FALSE;
3030    const uschar *ptr = *ptrptr;
3031    const uschar *tempptr;
3032    const uschar *nestptr = NULL;
3033    uschar *previous = NULL;
3034    uschar *previous_callout = NULL;
3035    uschar *save_hwm = NULL;
3036    uschar classbits[32];
3037    
3038  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
3039          if (utf8)  BOOL class_utf8;
3040            {                           /* Braces are required because the */  BOOL utf8 = (options & PCRE_UTF8) != 0;
3041            GETCHARLEN(d, ptr, ptr);    /* macro generates multiple statements */  uschar *class_utf8data;
3042            }  uschar *class_utf8data_base;
3043          else  uschar utf8_char[6];
3044    #else
3045    BOOL utf8 = FALSE;
3046    uschar *utf8_char = NULL;
3047  #endif  #endif
         d = *ptr;  /* Not UTF-8 mode */  
3048    
3049          /* The second part of a range can be a single-character escape, but  #ifdef PCRE_DEBUG
3050          not any of the other escapes. Perl 5.6 treats a hyphen as a literal  if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3051          in such circumstances. */  #endif
3052    
3053          if (d == '\\')  /* Set up the default and non-default settings for greediness */
           {  
           const uschar *oldptr = ptr;  
           d = check_escape(&ptr, errorcodeptr, *brackets, options, TRUE);  
3054    
3055            /* \b is backslash; \X is literal X; any other special means the '-'  greedy_default = ((options & PCRE_UNGREEDY) != 0);
3056            was literal */  greedy_non_default = greedy_default ^ 1;
3057    
3058            if (d < 0)  /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3059              {  matching encountered yet". It gets changed to REQ_NONE if we hit something that
3060              if (d == -ESC_b) d = '\b';  matches a non-fixed char first char; reqbyte just remains unset if we never
3061              else if (d == -ESC_X) d = 'X'; else  find one.
               {  
               ptr = oldptr - 2;  
               goto LONE_SINGLE_CHARACTER;  /* A few lines below */  
               }  
             }  
           }  
3062    
3063          /* 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
3064          the pre-pass. Optimize one-character ranges */  to take the zero repeat into account. This is implemented by setting them to
3065    zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
3066    item types that can be repeated set these backoff variables appropriately. */
3067    
3068          if (d == c) goto LONE_SINGLE_CHARACTER;  /* A few lines below */  firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
3069    
3070          /* 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,
3071          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
3072          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
3073          available. */  case status of the value. This is used only for ASCII characters. */
3074    
3075  #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;  
3076    
3077            /* 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. */  
3078    
3079  #ifdef SUPPORT_UCP  for (;; ptr++)
3080            if ((options & PCRE_CASELESS) != 0)    {
3081              {    BOOL negate_class;
3082              int occ, ocd;    BOOL should_flip_negation;
3083              int cc = c;    BOOL possessive_quantifier;
3084              int origd = d;    BOOL is_quantifier;
3085              while (get_othercase_range(&cc, origd, &occ, &ocd))    BOOL is_recurse;
3086                {    BOOL reset_bracount;
3087                if (occ >= c && ocd <= d) continue;  /* Skip embedded ranges */    int class_charcount;
3088      int class_lastchar;
3089      int newoptions;
3090      int recno;
3091      int refsign;
3092      int skipbytes;
3093      int subreqbyte;
3094      int subfirstbyte;
3095      int terminator;
3096      int mclength;
3097      uschar mcbuffer[8];
3098    
3099                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;  
                 }  
3100    
3101                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 */  
3102    
3103            /* 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
3104            overlapping ranges. */    string. Nesting only happens one level deep. */
3105    
3106            *class_utf8data++ = XCL_RANGE;    if (c == 0 && nestptr != NULL)
3107            class_utf8data += _pcre_ord2utf8(c, class_utf8data);      {
3108            class_utf8data += _pcre_ord2utf8(d, class_utf8data);      ptr = nestptr;
3109        nestptr = NULL;
3110        c = *ptr;
3111        }
3112    
3113            /* 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
3114            caseless matching for UTF-8 characters > 127; we can use the bit map    previous cycle of this loop. */
           for the smaller ones. */  
3115    
3116  #ifdef SUPPORT_UCP    if (lengthptr != NULL)
3117            continue;    /* With next character in the class */      {
3118  #else  #ifdef PCRE_DEBUG
3119            if ((options & PCRE_CASELESS) == 0 || c > 127) continue;      if (code > cd->hwm) cd->hwm = code;                 /* High water info */
3120    #endif
3121        if (code > cd->start_workspace + WORK_SIZE_CHECK)   /* Check for overrun */
3122          {
3123          *errorcodeptr = ERR52;
3124          goto FAILED;
3125          }
3126    
3127            /* Adjust upper limit and fall through to set up the map */      /* There is at least one situation where code goes backwards: this is the
3128        case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3129        the class is simply eliminated. However, it is created first, so we have to
3130        allow memory for it. Therefore, don't ever reduce the length at this point.
3131        */
3132    
3133            d = 127;      if (code < last_code) code = last_code;
3134    
3135  #endif  /* SUPPORT_UCP */      /* Paranoid check for integer overflow */
           }  
 #endif  /* SUPPORT_UTF8 */  
3136    
3137          /* We use the bit map for all cases when not in UTF-8 mode; else      if (OFLOW_MAX - *lengthptr < code - last_code)
3138          ranges that lie entirely within 0-127 when there is UCP support; else        {
3139          for partial ranges without UCP support. */        *errorcodeptr = ERR20;
3140          goto FAILED;
3141          }
3142    
3143          for (; c <= d; c++)      *lengthptr += (int)(code - last_code);
3144            {      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;  
           }  
3145    
3146          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
3147        it back to there, in order to avoid filling up the work space. Otherwise,
3148        if "previous" is NULL, reset the current code pointer to the start. */
3149    
3150        if (previous != NULL)
3151          {
3152          if (previous > orig_code)
3153            {
3154            memmove(orig_code, previous, code - previous);
3155            code -= previous - orig_code;
3156            previous = orig_code;
3157          }          }
3158          }
3159        else code = orig_code;
3160    
3161        /* 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
3162        non-escape char, or after \ that introduces a single character or for an      next time round. */
       apparent range that isn't. */  
3163    
3164        LONE_SINGLE_CHARACTER:      last_code = code;
3165        }
3166    
3167        /* Handle a character that cannot go in the bit map */    /* In the real compile phase, just check the workspace used by the forward
3168      reference list. */
3169    
3170  #ifdef SUPPORT_UTF8    else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
3171        if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))      {
3172          {      *errorcodeptr = ERR52;
3173          class_utf8 = TRUE;      goto FAILED;
3174          *class_utf8data++ = XCL_SINGLE;      }
         class_utf8data += _pcre_ord2utf8(c, class_utf8data);  
3175    
3176  #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 */  
3177    
3178      if (inescq && c != 0)
3179        {
3180        if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3181          {
3182          inescq = FALSE;
3183          ptr++;
3184          continue;
3185          }
3186        else
3187          {
3188          if (previous_callout != NULL)
3189            {
3190            if (lengthptr == NULL)  /* Don't attempt in pre-compile phase */
3191              complete_callout(previous_callout, ptr, cd);
3192            previous_callout = NULL;
3193          }          }
3194        else        if ((options & PCRE_AUTO_CALLOUT) != 0)
 #endif  /* SUPPORT_UTF8 */  
   
       /* Handle a single-byte character */  
3195          {          {
3196          classbits[c/8] |= (1 << (c&7));          previous_callout = code;
3197          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;  
3198          }          }
3199          goto NORMAL_CHAR;
3200        }        }
3201        }
3202    
3203      /* 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
3204      loop. This "while" is the end of the "do" above. */    a quantifier. */
3205    
3206      while ((c = *(++ptr)) != ']' || inescq);    is_quantifier =
3207        c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3208        (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3209    
3210      /* If class_charcount is 1, we saw precisely one character whose value is    if (!is_quantifier && previous_callout != NULL &&
3211      less than 256. In non-UTF-8 mode we can always optimize. In UTF-8 mode, we         after_manual_callout-- <= 0)
3212      can optimize the negative case only if there were no characters >= 128      {
3213      because OP_NOT and the related opcodes like OP_NOTSTAR operate on      if (lengthptr == NULL)      /* Don't attempt in pre-compile phase */
3214      single-bytes only. This is an historical hangover. Maybe one day we can        complete_callout(previous_callout, ptr, cd);
3215      tidy these opcodes to handle multi-byte characters.      previous_callout = NULL;
3216        }
3217    
3218      The optimization throws away the bit map. We turn the item into a    /* In extended mode, skip white space and comments */
     1-character OP_CHAR[NC] if it's positive, or OP_NOT if it's negative. Note  
     that OP_NOT does not support multibyte characters. In the positive case, it  
     can cause firstbyte to be set. Otherwise, there can be no first char if  
     this item is first, whatever repeat count may follow. In the case of  
     reqbyte, save the previous value for reinstating. */  
3219    
3220      if ((options & PCRE_EXTENDED) != 0)
3221        {
3222        if ((cd->ctypes[c] & ctype_space) != 0) continue;
3223        if (c == CHAR_NUMBER_SIGN)
3224          {
3225          ptr++;
3226          while (*ptr != 0)
3227            {
3228            if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3229            ptr++;
3230  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
3231      if (class_charcount == 1 &&          if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
           (!utf8 ||  
           (!class_utf8 && (!negate_class || class_lastchar < 128))))  
   
 #else  
     if (class_charcount == 1)  
3232  #endif  #endif
3233        {          }
3234        zeroreqbyte = reqbyte;        if (*ptr != 0) continue;
3235    
3236        /* The OP_NOT opcode works on one-byte characters only. */        /* Else fall through to handle end of string */
3237          c = 0;
3238          }
3239        }
3240    
3241        if (negate_class)    /* No auto callout for quantifiers. */
         {  
         if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;  
         zerofirstbyte = firstbyte;  
         *code++ = OP_NOT;  
         *code++ = class_lastchar;  
         break;  
         }  
3242    
3243        /* For a single, positive character, get the value into mcbuffer, and    if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3244        then we can handle this with the normal one-character code. */      {
3245        previous_callout = code;
3246        code = auto_callout(code, ptr, cd);
3247        }
3248    
3249  #ifdef SUPPORT_UTF8    switch(c)
3250        if (utf8 && class_lastchar > 127)      {
3251          mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);      /* ===================================================================*/
3252        else      case 0:                        /* The branch terminates at string end */
3253  #endif      case CHAR_VERTICAL_LINE:       /* or | or ) */
3254        case CHAR_RIGHT_PARENTHESIS:
3255        *firstbyteptr = firstbyte;
3256        *reqbyteptr = reqbyte;
3257        *codeptr = code;
3258        *ptrptr = ptr;
3259        if (lengthptr != NULL)
3260          {
3261          if (OFLOW_MAX - *lengthptr < code - last_code)
3262          {          {
3263          mcbuffer[0] = class_lastchar;          *errorcodeptr = ERR20;
3264          mclength = 1;          goto FAILED;
3265          }          }
3266        goto ONE_CHAR;        *lengthptr += (int)(code - last_code);   /* To include callout length */
3267        }       /* End of 1-char optimization */        DPRINTF((">> end branch\n"));
3268          }
3269      /* The general case - not the one-char optimization. If this is the first      return TRUE;
     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. */  
3270    
     if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;  
     zerofirstbyte = firstbyte;  
     zeroreqbyte = reqbyte;  
3271    
3272      /* If there are characters with values > 255, we have to compile an      /* ===================================================================*/
3273      extended class, with its own opcode. If there are no characters < 256,      /* Handle single-character metacharacters. In multiline mode, ^ disables
3274      we can omit the bitmap. */      the setting of any following char as a first character. */
3275    
3276  #ifdef SUPPORT_UTF8      case CHAR_CIRCUMFLEX_ACCENT:
3277      if (class_utf8)      previous = NULL;
3278        if ((options & PCRE_MULTILINE) != 0)
3279        {        {
3280        *class_utf8data++ = XCL_END;    /* Marks the end of extra data */        if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3281        *code++ = OP_XCLASS;        *code++ = OP_CIRCM;
3282        code += LINK_SIZE;        }
3283        *code = negate_class? XCL_NOT : 0;      else *code++ = OP_CIRC;
3284        break;
3285    
3286        /* If the map is required, install it, and move on to the end of      case CHAR_DOLLAR_SIGN:
3287        the extra data */      previous = NULL;
3288        *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3289        break;
3290    
3291        if (class_charcount > 0)      /* There can never be a first char if '.' is first, whatever happens about
3292          {      repeats. The value of reqbyte doesn't change either. */
         *code++ |= XCL_MAP;  
         memcpy(code, classbits, 32);  
         code = class_utf8data;  
         }  
3293    
3294        /* If the map is not required, slide down the extra data. */      case CHAR_DOT:
3295        if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3296        zerofirstbyte = firstbyte;
3297        zeroreqbyte = reqbyte;
3298        previous = code;
3299        *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3300        break;
3301    
       else  
         {  
         int len = class_utf8data - (code + 33);  
         memmove(code + 1, code + 33, len);  
         code += len + 1;  
         }  
3302    
3303        /* Now fill in the complete length of the item */      /* ===================================================================*/
3304        /* Character classes. If the included characters are all < 256, we build a
3305        32-byte bitmap of the permitted characters, except in the special case
3306        where there is only one such character. For negated classes, we build the
3307        map as usual, then invert it at the end. However, we use a different opcode
3308        so that data characters > 255 can be handled correctly.
3309    
3310        PUT(previous, 1, code - previous);      If the class contains characters outside the 0-255 range, a different
3311        break;   /* End of class handling */      opcode is compiled. It may optionally have a bit map for characters < 256,
3312        }      but those above are are explicitly listed afterwards. A flag byte tells
3313  #endif      whether the bitmap is present, and whether this is a negated class or not.
3314    
3315      /* If there are no characters > 255, negate the 32-byte map if necessary,      In JavaScript compatibility mode, an isolated ']' causes an error. In
3316      and copy it into the code vector. If this is the first thing in the branch,      default (Perl) mode, it is treated as a data character. */
     there can be no first char setting, whatever the repeat count. Any reqbyte  
     setting must remain unchanged after any kind of repeat. */  
3317    
3318      if (negate_class)      case CHAR_RIGHT_SQUARE_BRACKET:
3319        {      if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
       *code++ = OP_NCLASS;  
       for (c = 0; c < 32; c++) code[c] = ~classbits[c];  
       }  
     else  
3320        {        {
3321        *code++ = OP_CLASS;        *errorcodeptr = ERR64;
3322        memcpy(code, classbits, 32);        goto FAILED;
3323        }        }
3324      code += 32;      goto NORMAL_CHAR;
     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;  
3325    
3326      case '+':      case CHAR_LEFT_SQUARE_BRACKET:
3327      repeat_min = 1;      previous = code;
     repeat_max = -1;  
     goto REPEAT;  
3328    
3329      case '?':      /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3330      repeat_min = 0;      they are encountered at the top level, so we'll do that too. */
     repeat_max = 1;  
3331    
3332      REPEAT:      if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3333      if (previous == NULL)           ptr[1] == CHAR_EQUALS_SIGN) &&
3334            check_posix_syntax(ptr, &tempptr))
3335        {        {
3336        *errorcodeptr = ERR9;        *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3337        goto FAILED;        goto FAILED;
3338        }        }
3339    
3340      if (repeat_min == 0)      /* If the first character is '^', set the negation flag and skip it. Also,
3341        if the first few characters (either before or after ^) are \Q\E or \E we
3342        skip them too. This makes for compatibility with Perl. */
3343    
3344        negate_class = FALSE;
3345        for (;;)
3346        {        {
3347        firstbyte = zerofirstbyte;    /* Adjust for zero repeat */        c = *(++ptr);
3348        reqbyte = zeroreqbyte;        /* Ditto */        if (c == CHAR_BACKSLASH)
3349            {
3350            if (ptr[1] == CHAR_E)
3351              ptr++;
3352            else if (strncmp((const char *)ptr+1,
3353                              STR_Q STR_BACKSLASH STR_E, 3) == 0)
3354              ptr += 3;
3355            else
3356              break;
3357            }
3358          else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3359            negate_class = TRUE;
3360          else break;
3361        }        }
3362    
3363      /* Remember whether this is a variable length repeat */      /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3364        an initial ']' is taken as a data character -- the code below handles
3365        that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3366        [^] must match any character, so generate OP_ALLANY. */
3367    
3368      reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;      if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3369            (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3370          {
3371          *code++ = negate_class? OP_ALLANY : OP_FAIL;
3372          if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3373          zerofirstbyte = firstbyte;
3374          break;
3375          }
3376    
3377      op_type = 0;                    /* Default single-char op codes */      /* If a class contains a negative special such as \S, we need to flip the
3378      possessive_quantifier = FALSE;  /* Default not possessive quantifier */      negation flag at the end, so that support for characters > 255 works
3379        correctly (they are all included in the class). */
3380    
3381      /* Save start of previous item, in case we have to move it up to make space      should_flip_negation = FALSE;
     for an inserted OP_ONCE for the additional '+' extension. */  
3382    
3383      tempcode = previous;      /* Keep a count of chars with values < 256 so that we can optimize the case
3384        of just a single character (as long as it's < 256). However, For higher
3385        valued UTF-8 characters, we don't yet do any optimization. */
3386    
3387      /* If the next character is '+', we have a possessive quantifier. This      class_charcount = 0;
3388      implies greediness, whatever the setting of the PCRE_UNGREEDY option.      class_lastchar = -1;
     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. */  
3389    
3390      if (ptr[1] == '+')      /* Initialize the 32-char bit map to all zeros. We build the map in a
3391        {      temporary bit of memory, in case the class contains only 1 character (less
3392        repeat_type = 0;                  /* Force greedy */      than 256), because in that case the compiled code doesn't use the bit map.
3393        possessive_quantifier = TRUE;      */
       ptr++;  
       }  
     else if (ptr[1] == '?')  
       {  
       repeat_type = greedy_non_default;  
       ptr++;  
       }  
     else repeat_type = greedy_default;  
3394    
3395      /* If previous was a recursion, we need to wrap it inside brackets so that      memset(classbits, 0, 32 * sizeof(uschar));
     it can be replicated if necessary. */  
3396    
3397      if (*previous == OP_RECURSE)  #ifdef SUPPORT_UTF8
3398        {      class_utf8 = FALSE;                       /* No chars >= 256 */
3399        memmove(previous + 1 + LINK_SIZE, previous, 1 + LINK_SIZE);      class_utf8data = code + LINK_SIZE + 2;    /* For UTF-8 items */
3400        code += 1 + LINK_SIZE;      class_utf8data_base = class_utf8data;     /* For resetting in pass 1 */
3401        *previous = OP_BRA;  #endif
       PUT(previous, 1, code - previous);  
       *code = OP_KET;  
       PUT(code, 1, code - previous);  
       code += 1 + LINK_SIZE;  
       }  
3402    
3403      /* If previous was a character match, abolish the item and generate a      /* Process characters until ] is reached. By writing this as a "do" it
3404      repeat item instead. If a char item has a minumum of more than one, ensure      means that an initial ] is taken as a data character. At the start of the
3405      that it is set in reqbyte - it might not be if a sequence such as x{3} is      loop, c contains the first byte of the character. */
     the first thing in a branch because the x will have gone into firstbyte  
     instead.  */  
3406    
3407      if (*previous == OP_CHAR || *previous == OP_CHARNC)      if (c != 0) do
3408        {        {
3409        /* Deal with UTF-8 characters that take up more than one byte. It's        const uschar *oldptr;
       easier to write this out separately than try to macrify it. Use c to  
       hold the length of the character in bytes, plus 0x80 to flag that it's a  
       length rather than a small character. */  
3410    
3411  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
3412        if (utf8 && (code[-1] & 0x80) != 0)        if (utf8 && c > 127)
3413          {          {                           /* Braces are required because the */
3414          uschar *lastchar = code - 1;          GETCHARLEN(c, ptr, ptr);    /* macro generates multiple statements */
         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 */  
3415          }          }
       else  
 #endif  
3416    
3417        /* Handle the case of a single byte - either with no UTF8 support, or        /* In the pre-compile phase, accumulate the length of any UTF-8 extra
3418        with UTF-8 disabled, or for a UTF-8 character < 128. */        data and reset the pointer. This is so that very large classes that
3419          contain a zillion UTF-8 characters no longer overwrite the work space
3420          (which is on the stack). */
3421    
3422          if (lengthptr != NULL)
3423          {          {
3424          c = code[-1];          *lengthptr += class_utf8data - class_utf8data_base;
3425          if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;          class_utf8data = class_utf8data_base;
3426          }          }
3427    
3428        goto OUTPUT_SINGLE_REPEAT;   /* Code shared with single character types */  #endif
       }  
3429    
3430      /* If previous was a single negated character ([^a] or similar), we use        /* Inside \Q...\E everything is literal except \E */
     one of the special opcodes, replacing it. The code is shared with single-  
     character repeats by setting opt_type to add a suitable offset into  
     repeat_type. OP_NOT is currently used only for single-byte chars. */  
3431    
3432      else if (*previous == OP_NOT)        if (inescq)
3433        {          {
3434        op_type = OP_NOTSTAR - OP_STAR;  /* Use "not" opcodes */          if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)  /* If we are at \E */
3435        c = previous[1];            {
3436        goto OUTPUT_SINGLE_REPEAT;            inescq = FALSE;                   /* Reset literal state */
3437        }            ptr++;                            /* Skip the 'E' */
3438              continue;                         /* Carry on with next */
3439              }
3440            goto CHECK_RANGE;                   /* Could be range if \E follows */
3441            }
3442    
3443      /* If previous was a character type match (\d or similar), abolish it and        /* Handle POSIX class names. Perl allows a negation extension of the
3444      create a suitable repeat item. The code is shared with single-character        form [:^name:]. A square bracket that doesn't match the syntax is
3445      repeats by setting op_type to add a suitable offset into repeat_type. Note        treated as a literal. We also recognize the POSIX constructions
3446      the the Unicode property types will be present only when SUPPORT_UCP is        [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3447      defined, but we don't wrap the little bits of code here because it just        5.6 and 5.8 do. */
     makes it horribly messy. */  
3448    
3449      else if (*previous < OP_EODN)        if (c == CHAR_LEFT_SQUARE_BRACKET &&
3450        {            (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3451        uschar *oldcode;             ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3452        int prop_type;          {
3453        op_type = OP_TYPESTAR - OP_STAR;  /* Use type opcodes */          BOOL local_negate = FALSE;
3454        c = *previous;          int posix_class, taboffset, tabopt;
3455            register const uschar *cbits = cd->cbits;
3456            uschar pbits[32];
3457    
3458        OUTPUT_SINGLE_REPEAT:          if (ptr[1] != CHAR_COLON)
3459        prop_type = (*previous == OP_PROP || *previous == OP_NOTPROP)?            {
3460          previous[1] : -1;            *errorcodeptr = ERR31;
3461              goto FAILED;
3462              }
3463    
3464        oldcode = code;          ptr += 2;
3465        code = previous;                  /* Usually overwrite previous item */          if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3466              {
3467              local_negate = TRUE;
3468              should_flip_negation = TRUE;  /* Note negative special */
3469              ptr++;
3470              }
3471    
3472        /* If the maximum is zero then the minimum must also be zero; Perl allows          posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3473        this case, so we do too - by simply omitting the item altogether. */          if (posix_class < 0)
3474              {
3475              *errorcodeptr = ERR30;
3476              goto FAILED;
3477              }
3478    
3479        if (repeat_max == 0) goto END_REPEAT;          /* If matching is caseless, upper and lower are converted to
3480            alpha. This relies on the fact that the class table starts with
3481            alpha, lower, upper as the first 3 entries. */
3482    
3483        /* All real repeats make it impossible to handle partial matching (maybe          if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3484        one day we will be able to remove this restriction). */            posix_class = 0;
3485    
3486        if (repeat_max != 1) cd->nopartial = TRUE;          /* When PCRE_UCP is set, some of the POSIX classes are converted to
3487            different escape sequences that use Unicode properties. */
3488    
3489        /* Combine the op_type with the repeat_type */  #ifdef SUPPORT_UCP
3490            if ((options & PCRE_UCP) != 0)
3491              {
3492              int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3493              if (posix_substitutes[pc] != NULL)
3494                {
3495                nestptr = tempptr + 1;
3496                ptr = posix_substitutes[pc] - 1;
3497                continue;
3498                }
3499              }
3500    #endif
3501            /* In the non-UCP case, we build the bit map for the POSIX class in a
3502            chunk of local store because we may be adding and subtracting from it,
3503            and we don't want to subtract bits that may be in the main map already.
3504            At the end we or the result into the bit map that is being built. */
3505    
3506        repeat_type += op_type;          posix_class *= 3;
3507    
3508        /* A minimum of zero is handled either as the special case * or ?, or as          /* Copy in the first table (always present) */
       an UPTO, with the maximum given. */  
3509    
3510        if (repeat_min == 0)          memcpy(pbits, cbits + posix_class_maps[posix_class],
3511          {            32 * sizeof(uschar));
3512          if (repeat_max == -1) *code++ = OP_STAR + repeat_type;  
3513            else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;          /* If there is a second table, add or remove it as required. */
3514          else  
3515            taboffset = posix_class_maps[posix_class + 1];
3516            tabopt = posix_class_maps[posix_class + 2];
3517    
3518            if (taboffset >= 0)
3519            {            {
3520            *code++ = OP_UPTO + repeat_type;            if (tabopt >= 0)
3521            PUT2INC(code, 0, repeat_max);              for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3522              else
3523                for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3524            }            }
         }  
3525    
3526        /* A repeat minimum of 1 is optimized into some special cases. If the          /* Not see if we need to remove any special characters. An option
3527        maximum is unlimited, we use OP_PLUS. Otherwise, the original item it          value of 1 removes vertical space and 2 removes underscore. */
       left in place and, if the maximum is greater than 1, we use OP_UPTO with  
       one less than the maximum. */  
3528    
3529        else if (repeat_min == 1)          if (tabopt < 0) tabopt = -tabopt;
3530          {          if (tabopt == 1) pbits[1] &= ~0x3c;
3531          if (repeat_max == -1)            else if (tabopt == 2) pbits[11] &= 0x7f;
3532            *code++ = OP_PLUS + repeat_type;  
3533            /* Add the POSIX table or its complement into the main table that is
3534            being built and we are done. */
3535    
3536            if (local_negate)
3537              for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3538          else          else
3539            {            for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3540            code = oldcode;                 /* leave previous item in place */  
3541            if (repeat_max == 1) goto END_REPEAT;          ptr = tempptr + 1;
3542            *code++ = OP_UPTO + repeat_type;          class_charcount = 10;  /* Set > 1; assumes more than 1 per class */
3543            PUT2INC(code, 0, repeat_max - 1);          continue;    /* End of POSIX syntax handling */
           }  
3544          }          }
3545    
3546        /* The case {n,n} is just an EXACT, while the general case {n,m} is        /* Backslash may introduce a single character, or it may introduce one
3547        handled as an EXACT followed by an UPTO. */        of the specials, which just set a flag. The sequence \b is a special
3548          case. Inside a class (and only there) it is treated as backspace. We
3549          assume that other escapes have more than one character in them, so set
3550          class_charcount bigger than one. Unrecognized escapes fall through and
3551          are either treated as literal characters (by default), or are faulted if
3552          PCRE_EXTRA is set. */
3553    
3554        else        if (c == CHAR_BACKSLASH)
3555          {          {
3556          *code++ = OP_EXACT + op_type;  /* NB EXACT doesn't have repeat_type */          c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3557          PUT2INC(code, 0, repeat_min);          if (*errorcodeptr != 0) goto FAILED;
3558    
3559          /* If the maximum is unlimited, insert an OP_STAR. Before doing so,          if (-c == ESC_b) c = CHAR_BS;    /* \b is backspace in a class */
3560          we have to insert the character for the previous code. For a repeated          else if (-c == ESC_Q)            /* Handle start of quoted string */
         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. */  
   
         if (repeat_max < 0)  
3561            {            {
3562  #ifdef SUPPORT_UTF8            if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
           if (utf8 && c >= 128)  
             {  
             memcpy(code, utf8_char, c & 7);  
             code += c & 7;  
             }  
           else  
 #endif  
3563              {              {
3564              *code++ = c;              ptr += 2; /* avoid empty string */
             if (prop_type >= 0) *code++ = prop_type;  
3565              }              }
3566            *code++ = OP_STAR + repeat_type;            else inescq = TRUE;
3567              continue;
3568</