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

Diff of /code/trunk/pcre_compile.c

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

revision 79 by nigel, Sat Feb 24 21:40:52 2007 UTC revision 762 by ph10, Tue Nov 22 13:36:51 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"  /** DEAD **/
397      "inconsistent NEWLINE options\0"
398      "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
399      "a numbered reference must not be zero\0"
400      "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
401      /* 60 */
402      "(*VERB) not recognized\0"
403      "number is too big\0"
404      "subpattern name expected\0"
405      "digit expected after (?+\0"
406      "] is an invalid data character in JavaScript compatibility mode\0"
407      /* 65 */
408      "different names for subpatterns of the same number are not allowed\0"
409      "(*MARK) must have an argument\0"
410      "this version of PCRE is not compiled with PCRE_UCP support\0"
411      "\\c must be followed by an ASCII character\0"
412      "\\k is not followed by a braced, angle-bracketed, or quoted name\0"
413      /* 70 */
414      "internal error: unknown opcode in find_fixedlength()\0"
415      "\\N is not supported in a class\0"
416      "too many forward references\0"
417      ;
418    
419  /* Table to identify digits and hex digits. This is used when compiling  /* Table to identify digits and hex digits. This is used when compiling
420  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 432  For convenience, we use the same bit def
432    
433  Then we can use ctype_digit and ctype_xdigit in the code. */  Then we can use ctype_digit and ctype_xdigit in the code. */
434    
435  #if !EBCDIC    /* This is the "normal" case, for ASCII systems */  #ifndef EBCDIC
436    
437    /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
438    UTF-8 mode. */
439    
440  static const unsigned char digitab[] =  static const unsigned char digitab[] =
441    {    {
442    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 472  static const unsigned char digitab[] =
472    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
473    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
474    
475  #else          /* This is the "abnormal" case, for EBCDIC systems */  #else
476    
477    /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
478    
479  static const unsigned char digitab[] =  static const unsigned char digitab[] =
480    {    {
481    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 489  static const unsigned char digitab[] =
489    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*    - 71 40 */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*    - 71 40 */
490    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  72- |     */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  72- |     */
491    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  & - 87 50 */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  & - 87 50 */
492    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  88-     */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  88- 95    */
493    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  - -103 60 */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  - -103 60 */
494    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ?     */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ?     */
495    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 523  static const unsigned char ebcdic_charta
523    0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*    - 71 */    0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*    - 71 */
524    0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /*  72- |  */    0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /*  72- |  */
525    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  & - 87 */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  & - 87 */
526    0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /*  88-  */    0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /*  88- 95 */
527    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  - -103 */    0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /*  - -103 */
528    0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ?  */    0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ?  */
529    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 550  static const unsigned char ebcdic_charta
550  /* Definition to allow mutual recursion */  /* Definition to allow mutual recursion */
551    
552  static BOOL  static BOOL
553    compile_regex(int, int, int *, uschar **, const uschar **, int *, BOOL, int,    compile_regex(int, uschar **, const uschar **, int *, BOOL, BOOL, int, int,
554      int *, int *, branch_chain *, compile_data *);      int *, int *, branch_chain *, compile_data *, int *);
555    
556    
557    
558    /*************************************************
559    *            Find an error text                  *
560    *************************************************/
561    
562    /* The error texts are now all in one long string, to save on relocations. As
563    some of the text is of unknown length, we can't use a table of offsets.
564    Instead, just count through the strings. This is not a performance issue
565    because it happens only when there has been a compilation error.
566    
567    Argument:   the error number
568    Returns:    pointer to the error string
569    */
570    
571    static const char *
572    find_error_text(int n)
573    {
574    const char *s = error_texts;
575    for (; n > 0; n--)
576      {
577      while (*s++ != 0) {};
578      if (*s == 0) return "Error text not found (please report)";
579      }
580    return s;
581    }
582    
583    
584    /*************************************************
585    *            Check for counted repeat            *
586    *************************************************/
587    
588    /* This function is called when a '{' is encountered in a place where it might
589    start a quantifier. It looks ahead to see if it really is a quantifier or not.
590    It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
591    where the ddds are digits.
592    
593    Arguments:
594      p         pointer to the first char after '{'
595    
596    Returns:    TRUE or FALSE
597    */
598    
599    static BOOL
600    is_counted_repeat(const uschar *p)
601    {
602    if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
603    while ((digitab[*p] & ctype_digit) != 0) p++;
604    if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
605    
606    if (*p++ != CHAR_COMMA) return FALSE;
607    if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
608    
609    if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
610    while ((digitab[*p] & ctype_digit) != 0) p++;
611    
612    return (*p == CHAR_RIGHT_CURLY_BRACKET);
613    }
614    
615    
616    
# Line 342  static BOOL Line 620  static BOOL
620    
621  /* 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
622  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
623  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
624  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
625  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,
626    ptr is pointing at the \. On exit, it is on the final character of the escape
627    sequence.
628    
629  Arguments:  Arguments:
630    ptrptr         points to the pattern position pointer    ptrptr         points to the pattern position pointer
# Line 355  Arguments: Line 635  Arguments:
635    
636  Returns:         zero or positive => a data character  Returns:         zero or positive => a data character
637                   negative => a special escape sequence                   negative => a special escape sequence
638                   on error, errorptr is set                   on error, errorcodeptr is set
639  */  */
640    
641  static int  static int
642  check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,  check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
643    int options, BOOL isclass)    int options, BOOL isclass)
644  {  {
645  const uschar *ptr = *ptrptr;  BOOL utf8 = (options & PCRE_UTF8) != 0;
646    const uschar *ptr = *ptrptr + 1;
647  int c, i;  int c, i;
648    
649    GETCHARINCTEST(c, ptr);           /* Get character value, increment pointer */
650    ptr--;                            /* Set pointer back to the last byte */
651    
652  /* 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. */
653    
 c = *(++ptr);  
654  if (c == 0) *errorcodeptr = ERR1;  if (c == 0) *errorcodeptr = ERR1;
655    
656  /* 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
657  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.
658  Otherwise further processing may be required. */  Otherwise further processing may be required. */
659    
660  #if !EBCDIC    /* ASCII coding */  #ifndef EBCDIC  /* ASCII/UTF-8 coding */
661  else if (c < '0' || c > 'z') {}                           /* Not alphameric */  else if (c < CHAR_0 || c > CHAR_z) {}                     /* Not alphanumeric */
662  else if ((i = escapes[c - '0']) != 0) c = i;  else if ((i = escapes[c - CHAR_0]) != 0) c = i;
663    
664  #else          /* EBCDIC coding */  #else           /* EBCDIC coding */
665  else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {}   /* Not alphameric */  else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {}   /* Not alphanumeric */
666  else if ((i = escapes[c - 0x48]) != 0)  c = i;  else if ((i = escapes[c - 0x48]) != 0)  c = i;
667  #endif  #endif
668    
# Line 388  else if ((i = escapes[c - 0x48]) != 0) Line 671  else if ((i = escapes[c - 0x48]) != 0)
671  else  else
672    {    {
673    const uschar *oldptr;    const uschar *oldptr;
674      BOOL braced, negated;
675    
676    switch (c)    switch (c)
677      {      {
678      /* 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
679      error. */      error. */
680    
681      case 'l':      case CHAR_l:
682      case 'L':      case CHAR_L:
     case 'N':  
     case 'u':  
     case 'U':  
683      *errorcodeptr = ERR37;      *errorcodeptr = ERR37;
684      break;      break;
685    
686        case CHAR_u:
687        if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
688          {
689          /* In JavaScript, \u must be followed by four hexadecimal numbers.
690          Otherwise it is a lowercase u letter. */
691          if ((digitab[ptr[1]] & ctype_xdigit) != 0 && (digitab[ptr[2]] & ctype_xdigit) != 0
692               && (digitab[ptr[3]] & ctype_xdigit) != 0 && (digitab[ptr[4]] & ctype_xdigit) != 0)
693            {
694            c = 0;
695            for (i = 0; i < 4; ++i)
696              {
697              register int cc = *(++ptr);
698    #ifndef EBCDIC  /* ASCII/UTF-8 coding */
699              if (cc >= CHAR_a) cc -= 32;               /* Convert to upper case */
700              c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
701    #else           /* EBCDIC coding */
702              if (cc >= CHAR_a && cc <= CHAR_z) cc += 64;  /* Convert to upper case */
703              c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
704    #endif
705              }
706            }
707          }
708        else
709          *errorcodeptr = ERR37;
710        break;
711    
712        case CHAR_U:
713        /* In JavaScript, \U is an uppercase U letter. */
714        if ((options & PCRE_JAVASCRIPT_COMPAT) == 0) *errorcodeptr = ERR37;
715        break;
716    
717        /* In a character class, \g is just a literal "g". Outside a character
718        class, \g must be followed by one of a number of specific things:
719    
720        (1) A number, either plain or braced. If positive, it is an absolute
721        backreference. If negative, it is a relative backreference. This is a Perl
722        5.10 feature.
723    
724        (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
725        is part of Perl's movement towards a unified syntax for back references. As
726        this is synonymous with \k{name}, we fudge it up by pretending it really
727        was \k.
728    
729        (3) For Oniguruma compatibility we also support \g followed by a name or a
730        number either in angle brackets or in single quotes. However, these are
731        (possibly recursive) subroutine calls, _not_ backreferences. Just return
732        the -ESC_g code (cf \k). */
733    
734        case CHAR_g:
735        if (isclass) break;
736        if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
737          {
738          c = -ESC_g;
739          break;
740          }
741    
742        /* Handle the Perl-compatible cases */
743    
744        if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
745          {
746          const uschar *p;
747          for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
748            if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break;
749          if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
750            {
751            c = -ESC_k;
752            break;
753            }
754          braced = TRUE;
755          ptr++;
756          }
757        else braced = FALSE;
758    
759        if (ptr[1] == CHAR_MINUS)
760          {
761          negated = TRUE;
762          ptr++;
763          }
764        else negated = FALSE;
765    
766        c = 0;
767        while ((digitab[ptr[1]] & ctype_digit) != 0)
768          c = c * 10 + *(++ptr) - CHAR_0;
769    
770        if (c < 0)   /* Integer overflow */
771          {
772          *errorcodeptr = ERR61;
773          break;
774          }
775    
776        if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
777          {
778          *errorcodeptr = ERR57;
779          break;
780          }
781    
782        if (c == 0)
783          {
784          *errorcodeptr = ERR58;
785          break;
786          }
787    
788        if (negated)
789          {
790          if (c > bracount)
791            {
792            *errorcodeptr = ERR15;
793            break;
794            }
795          c = bracount - (c - 1);
796          }
797    
798        c = -(ESC_REF + c);
799        break;
800    
801      /* The handling of escape sequences consisting of a string of digits      /* The handling of escape sequences consisting of a string of digits
802      starting with one that is not zero is not straightforward. By experiment,      starting with one that is not zero is not straightforward. By experiment,
803      the way Perl works seems to be as follows:      the way Perl works seems to be as follows:
# Line 413  else Line 810  else
810      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
811      character class, \ followed by a digit is always an octal number. */      character class, \ followed by a digit is always an octal number. */
812    
813      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:
814      case '6': case '7': case '8': case '9':      case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
815    
816      if (!isclass)      if (!isclass)
817        {        {
818        oldptr = ptr;        oldptr = ptr;
819        c -= '0';        c -= CHAR_0;
820        while ((digitab[ptr[1]] & ctype_digit) != 0)        while ((digitab[ptr[1]] & ctype_digit) != 0)
821          c = c * 10 + *(++ptr) - '0';          c = c * 10 + *(++ptr) - CHAR_0;
822          if (c < 0)    /* Integer overflow */
823            {
824            *errorcodeptr = ERR61;
825            break;
826            }
827        if (c < 10 || c <= bracount)        if (c < 10 || c <= bracount)
828          {          {
829          c = -(ESC_REF + c);          c = -(ESC_REF + c);
# Line 434  else Line 836  else
836      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.
837      Thus we have to pull back the pointer by one. */      Thus we have to pull back the pointer by one. */
838    
839      if ((c = *ptr) >= '8')      if ((c = *ptr) >= CHAR_8)
840        {        {
841        ptr--;        ptr--;
842        c = 0;        c = 0;
# Line 442  else Line 844  else
844        }        }
845    
846      /* \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
847      larger first octal digit. */      larger first octal digit. The original code used just to take the least
848        significant 8 bits of octal numbers (I think this is what early Perls used
849      case '0':      to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
850      c -= '0';      than 3 octal digits. */
851      while(i++ < 2 && ptr[1] >= '0' && ptr[1] <= '7')  
852          c = c * 8 + *(++ptr) - '0';      case CHAR_0:
853      c &= 255;     /* Take least significant 8 bits */      c -= CHAR_0;
854        while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
855            c = c * 8 + *(++ptr) - CHAR_0;
856        if (!utf8 && c > 255) *errorcodeptr = ERR51;
857      break;      break;
858    
859      /* \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
860      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
861        treated as a data character. */
862    
863        case CHAR_x:
864        if ((options & PCRE_JAVASCRIPT_COMPAT) != 0)
865          {
866          /* In JavaScript, \x must be followed by two hexadecimal numbers.
867          Otherwise it is a lowercase x letter. */
868          if ((digitab[ptr[1]] & ctype_xdigit) != 0 && (digitab[ptr[2]] & ctype_xdigit) != 0)
869            {
870            c = 0;
871            for (i = 0; i < 2; ++i)
872              {
873              register int cc = *(++ptr);
874    #ifndef EBCDIC  /* ASCII/UTF-8 coding */
875              if (cc >= CHAR_a) cc -= 32;               /* Convert to upper case */
876              c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
877    #else           /* EBCDIC coding */
878              if (cc >= CHAR_a && cc <= CHAR_z) cc += 64;  /* Convert to upper case */
879              c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
880    #endif
881              }
882            }
883          break;
884          }
885    
886      case 'x':      if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
 #ifdef SUPPORT_UTF8  
     if (ptr[1] == '{' && (options & PCRE_UTF8) != 0)  
887        {        {
888        const uschar *pt = ptr + 2;        const uschar *pt = ptr + 2;
889        register int count = 0;        int count = 0;
890    
891        c = 0;        c = 0;
892        while ((digitab[*pt] & ctype_xdigit) != 0)        while ((digitab[*pt] & ctype_xdigit) != 0)
893          {          {
894          int cc = *pt++;          register int cc = *pt++;
895            if (c == 0 && cc == CHAR_0) continue;     /* Leading zeroes */
896          count++;          count++;
897  #if !EBCDIC    /* ASCII coding */  
898          if (cc >= 'a') cc -= 32;               /* Convert to upper case */  #ifndef EBCDIC  /* ASCII/UTF-8 coding */
899          c = c * 16 + cc - ((cc < 'A')? '0' : ('A' - 10));          if (cc >= CHAR_a) cc -= 32;               /* Convert to upper case */
900  #else          /* EBCDIC coding */          c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
901          if (cc >= 'a' && cc <= 'z') cc += 64;  /* Convert to upper case */  #else           /* EBCDIC coding */
902          c = c * 16 + cc - ((cc >= '0')? '0' : ('A' - 10));          if (cc >= CHAR_a && cc <= CHAR_z) cc += 64;  /* Convert to upper case */
903            c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
904  #endif  #endif
905          }          }
906        if (*pt == '}')  
907          if (*pt == CHAR_RIGHT_CURLY_BRACKET)
908          {          {
909          if (c < 0 || count > 8) *errorcodeptr = ERR34;          if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
910          ptr = pt;          ptr = pt;
911          break;          break;
912          }          }
913    
914        /* 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
915        recognize this construct; fall through to the normal \x handling. */        recognize this construct; fall through to the normal \x handling. */
916        }        }
 #endif  
917    
918      /* Read just a single hex char */      /* Read just a single-byte hex-defined char */
919    
920      c = 0;      c = 0;
921      while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)      while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
922        {        {
923        int cc;                               /* Some compilers don't like ++ */        int cc;                                  /* Some compilers don't like */
924        cc = *(++ptr);                        /* in initializers */        cc = *(++ptr);                           /* ++ in initializers */
925  #if !EBCDIC    /* ASCII coding */  #ifndef EBCDIC  /* ASCII/UTF-8 coding */
926        if (cc >= 'a') cc -= 32;              /* Convert to upper case */        if (cc >= CHAR_a) cc -= 32;              /* Convert to upper case */
927        c = c * 16 + cc - ((cc < 'A')? '0' : ('A' - 10));        c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
928  #else          /* EBCDIC coding */  #else           /* EBCDIC coding */
929        if (cc <= 'z') cc += 64;              /* Convert to upper case */        if (cc <= CHAR_z) cc += 64;              /* Convert to upper case */
930        c = c * 16 + cc - ((cc >= '0')? '0' : ('A' - 10));        c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
931  #endif  #endif
932        }        }
933      break;      break;
934    
935      /* Other special escapes not starting with a digit are straightforward */      /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
936        An error is given if the byte following \c is not an ASCII character. This
937        coding is ASCII-specific, but then the whole concept of \cx is
938        ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
939    
940      case 'c':      case CHAR_c:
941      c = *(++ptr);      c = *(++ptr);
942      if (c == 0)      if (c == 0)
943        {        {
944        *errorcodeptr = ERR2;        *errorcodeptr = ERR2;
945        return 0;        break;
946        }        }
947    #ifndef EBCDIC    /* ASCII/UTF-8 coding */
948      /* A letter is upper-cased; then the 0x40 bit is flipped. This coding      if (c > 127)  /* Excludes all non-ASCII in either mode */
949      is ASCII-specific, but then the whole concept of \cx is ASCII-specific.        {
950      (However, an EBCDIC equivalent has now been added.) */        *errorcodeptr = ERR68;
951          break;
952  #if !EBCDIC    /* ASCII coding */        }
953      if (c >= 'a' && c <= 'z') c -= 32;      if (c >= CHAR_a && c <= CHAR_z) c -= 32;
954      c ^= 0x40;      c ^= 0x40;
955  #else          /* EBCDIC coding */  #else             /* EBCDIC coding */
956      if (c >= 'a' && c <= 'z') c += 64;      if (c >= CHAR_a && c <= CHAR_z) c += 64;
957      c ^= 0xC0;      c ^= 0xC0;
958  #endif  #endif
959      break;      break;
960    
961      /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any      /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
962      other alphameric following \ is an error if PCRE_EXTRA was set; otherwise,      other alphanumeric following \ is an error if PCRE_EXTRA was set;
963      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
964      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
965      in future, so I haven't "optimized" it. */      be again in future, so I haven't "optimized" it. */
966    
967      default:      default:
968      if ((options & PCRE_EXTRA) != 0) switch(c)      if ((options & PCRE_EXTRA) != 0) switch(c)
# Line 541  else Line 975  else
975      }      }
976    }    }
977    
978    /* Perl supports \N{name} for character names, as well as plain \N for "not
979    newline". PCRE does not support \N{name}. However, it does support
980    quantification such as \N{2,3}. */
981    
982    if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET &&
983         !is_counted_repeat(ptr+2))
984      *errorcodeptr = ERR37;
985    
986    /* If PCRE_UCP is set, we change the values for \d etc. */
987    
988    if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
989      c -= (ESC_DU - ESC_D);
990    
991    /* Set the pointer to the final character before returning. */
992    
993  *ptrptr = ptr;  *ptrptr = ptr;
994  return c;  return c;
995  }  }
# Line 560  escape sequence. Line 1009  escape sequence.
1009  Argument:  Argument:
1010    ptrptr         points to the pattern position pointer    ptrptr         points to the pattern position pointer
1011    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
1012      dptr           points to an int that is set to the detailed property value
1013    errorcodeptr   points to the error code variable    errorcodeptr   points to the error code variable
1014    
1015  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
1016  */  */
1017    
1018  static int  static int
1019  get_ucp(const uschar **ptrptr, BOOL *negptr, int *errorcodeptr)  get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
1020  {  {
1021  int c, i, bot, top;  int c, i, bot, top;
1022  const uschar *ptr = *ptrptr;  const uschar *ptr = *ptrptr;
1023  char name[4];  char name[32];
1024    
1025  c = *(++ptr);  c = *(++ptr);
1026  if (c == 0) goto ERROR_RETURN;  if (c == 0) goto ERROR_RETURN;
1027    
1028  *negptr = FALSE;  *negptr = FALSE;
1029    
1030  /* \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
1031  preceded by ^ for negation. */  negation. */
1032    
1033  if (c == '{')  if (c == CHAR_LEFT_CURLY_BRACKET)
1034    {    {
1035    if (ptr[1] == '^')    if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1036      {      {
1037      *negptr = TRUE;      *negptr = TRUE;
1038      ptr++;      ptr++;
1039      }      }
1040    for (i = 0; i <= 2; i++)    for (i = 0; i < (int)sizeof(name) - 1; i++)
1041      {      {
1042      c = *(++ptr);      c = *(++ptr);
1043      if (c == 0) goto ERROR_RETURN;      if (c == 0) goto ERROR_RETURN;
1044      if (c == '}') break;      if (c == CHAR_RIGHT_CURLY_BRACKET) break;
1045      name[i] = c;      name[i] = c;
1046      }      }
1047    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;  
     }  
1048    name[i] = 0;    name[i] = 0;
1049    }    }
1050    
# Line 619  top = _pcre_utt_size; Line 1065  top = _pcre_utt_size;
1065    
1066  while (bot < top)  while (bot < top)
1067    {    {
1068    i = (bot + top)/2;    i = (bot + top) >> 1;
1069    c = strcmp(name, _pcre_utt[i].name);    c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
1070    if (c == 0) return _pcre_utt[i].value;    if (c == 0)
1071        {
1072        *dptr = _pcre_utt[i].value;
1073        return _pcre_utt[i].type;
1074        }
1075    if (c > 0) bot = i + 1; else top = i;    if (c > 0) bot = i + 1; else top = i;
1076    }    }
1077    
 UNKNOWN_RETURN:  
1078  *errorcodeptr = ERR47;  *errorcodeptr = ERR47;
1079  *ptrptr = ptr;  *ptrptr = ptr;
1080  return -1;  return -1;
# Line 641  return -1; Line 1090  return -1;
1090    
1091    
1092  /*************************************************  /*************************************************
 *            Check for counted repeat            *  
 *************************************************/  
   
 /* This function is called when a '{' is encountered in a place where it might  
 start a quantifier. It looks ahead to see if it really is a quantifier or not.  
 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}  
 where the ddds are digits.  
   
 Arguments:  
   p         pointer to the first char after '{'  
   
 Returns:    TRUE or FALSE  
 */  
   
 static BOOL  
 is_counted_repeat(const uschar *p)  
 {  
 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;  
 while ((digitab[*p] & ctype_digit) != 0) p++;  
 if (*p == '}') return TRUE;  
   
 if (*p++ != ',') return FALSE;  
 if (*p == '}') return TRUE;  
   
 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;  
 while ((digitab[*p] & ctype_digit) != 0) p++;  
   
 return (*p == '}');  
 }  
   
   
   
 /*************************************************  
1093  *         Read repeat counts                     *  *         Read repeat counts                     *
1094  *************************************************/  *************************************************/
1095    
# Line 698  read_repeat_counts(const uschar *p, int Line 1114  read_repeat_counts(const uschar *p, int
1114  int min = 0;  int min = 0;
1115  int max = -1;  int max = -1;
1116    
1117  while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - '0';  /* Read the minimum value and do a paranoid check: a negative value indicates
1118    an integer overflow. */
1119    
1120    while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - CHAR_0;
1121    if (min < 0 || min > 65535)
1122      {
1123      *errorcodeptr = ERR5;
1124      return p;
1125      }
1126    
1127    /* Read the maximum value if there is one, and again do a paranoid on its size.
1128    Also, max must not be less than min. */
1129    
1130  if (*p == '}') max = min; else  if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1131    {    {
1132    if (*(++p) != '}')    if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1133      {      {
1134      max = 0;      max = 0;
1135      while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - '0';      while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0;
1136        if (max < 0 || max > 65535)
1137          {
1138          *errorcodeptr = ERR5;
1139          return p;
1140          }
1141      if (max < min)      if (max < min)
1142        {        {
1143        *errorcodeptr = ERR4;        *errorcodeptr = ERR4;
# Line 714  if (*p == '}') max = min; else Line 1146  if (*p == '}') max = min; else
1146      }      }
1147    }    }
1148    
1149  /* Do paranoid checks, then fill in the required variables, and pass back the  /* Fill in the required variables, and pass back the pointer to the terminating
1150  pointer to the terminating '}'. */  '}'. */
1151    
1152  if (min > 65535 || max > 65535)  *minp = min;
1153    *errorcodeptr = ERR5;  *maxp = max;
 else  
   {  
   *minp = min;  
   *maxp = max;  
   }  
1154  return p;  return p;
1155  }  }
1156    
1157    
1158    
1159  /*************************************************  /*************************************************
1160  *      Find first significant op code            *  *  Subroutine for finding forward reference      *
1161  *************************************************/  *************************************************/
1162    
1163  /* This is called by several functions that scan a compiled expression looking  /* This recursive function is called only from find_parens() below. The
1164  for a fixed first character, or an anchoring op code etc. It skips over things  top-level call starts at the beginning of the pattern. All other calls must
1165  that do not influence this. For some calls, a change of option is important.  start at a parenthesis. It scans along a pattern's text looking for capturing
1166  For some calls, it makes sense to skip negative forward and all backward  subpatterns, and counting them. If it finds a named pattern that matches the
1167  assertions, and also the \b assertion; for others it does not.  name it is given, it returns its number. Alternatively, if the name is NULL, it
1168    returns when it reaches a given numbered subpattern. Recursion is used to keep
1169    track of subpatterns that reset the capturing group numbers - the (?| feature.
1170    
1171    This function was originally called only from the second pass, in which we know
1172    that if (?< or (?' or (?P< is encountered, the name will be correctly
1173    terminated because that is checked in the first pass. There is now one call to
1174    this function in the first pass, to check for a recursive back reference by
1175    name (so that we can make the whole group atomic). In this case, we need check
1176    only up to the current position in the pattern, and that is still OK because
1177    and previous occurrences will have been checked. To make this work, the test
1178    for "end of pattern" is a check against cd->end_pattern in the main loop,
1179    instead of looking for a binary zero. This means that the special first-pass
1180    call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1181    processing items within the loop are OK, because afterwards the main loop will
1182    terminate.)
1183    
1184  Arguments:  Arguments:
1185    code         pointer to the start of the group    ptrptr       address of the current character pointer (updated)
1186    options      pointer to external options    cd           compile background data
1187    optbit       the option bit whose changing is significant, or    name         name to seek, or NULL if seeking a numbered subpattern
1188                   zero if none are    lorn         name length, or subpattern number if name is NULL
1189    skipassert   TRUE if certain assertions are to be skipped    xmode        TRUE if we are in /x mode
1190      utf8         TRUE if we are in UTF-8 mode
1191      count        pointer to the current capturing subpattern number (updated)
1192    
1193  Returns:       pointer to the first significant opcode  Returns:       the number of the named subpattern, or -1 if not found
1194  */  */
1195    
1196  static const uschar*  static int
1197  first_significant_code(const uschar *code, int *options, int optbit,  find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
1198    BOOL skipassert)    BOOL xmode, BOOL utf8, int *count)
1199  {  {
1200  for (;;)  uschar *ptr = *ptrptr;
1201    {  int start_count = *count;
1202    switch ((int)*code)  int hwm_count = start_count;
1203      {  BOOL dup_parens = FALSE;
     case OP_OPT:  
     if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))  
       *options = (int)code[1];  
     code += 2;  
     break;  
1204    
1205      case OP_ASSERT_NOT:  /* If the first character is a parenthesis, check on the type of group we are
1206      case OP_ASSERTBACK:  dealing with. The very first call may not start with a parenthesis. */
     case OP_ASSERTBACK_NOT:  
     if (!skipassert) return code;  
     do code += GET(code, 1); while (*code == OP_ALT);  
     code += _pcre_OP_lengths[*code];  
     break;  
1207    
1208      case OP_WORD_BOUNDARY:  if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1209      case OP_NOT_WORD_BOUNDARY:    {
1210      if (!skipassert) return code;    /* Handle specials such as (*SKIP) or (*UTF8) etc. */
     /* Fall through */  
1211    
1212      case OP_CALLOUT:    if (ptr[1] == CHAR_ASTERISK) ptr += 2;
     case OP_CREF:  
     case OP_BRANUMBER:  
     code += _pcre_OP_lengths[*code];  
     break;  
1213    
1214      default:    /* Handle a normal, unnamed capturing parenthesis. */
1215      return code;  
1216      else if (ptr[1] != CHAR_QUESTION_MARK)
1217        {
1218        *count += 1;
1219        if (name == NULL && *count == lorn) return *count;
1220        ptr++;
1221      }      }
   }  
 /* Control never reaches here */  
 }  
1222    
1223      /* All cases now have (? at the start. Remember when we are in a group
1224      where the parenthesis numbers are duplicated. */
1225    
1226      else if (ptr[2] == CHAR_VERTICAL_LINE)
1227        {
1228        ptr += 3;
1229        dup_parens = TRUE;
1230        }
1231    
1232      /* Handle comments; all characters are allowed until a ket is reached. */
1233    
1234  /*************************************************    else if (ptr[2] == CHAR_NUMBER_SIGN)
1235  *        Find the fixed length of a pattern      *      {
1236  *************************************************/      for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1237        goto FAIL_EXIT;
1238        }
1239    
1240  /* Scan a pattern and compute the fixed length of subject that will match it,    /* Handle a condition. If it is an assertion, just carry on so that it
1241  if the length is fixed. This is needed for dealing with backward assertions.    is processed as normal. If not, skip to the closing parenthesis of the
1242  In UTF8 mode, the result is in characters rather than bytes.    condition (there can't be any nested parens). */
1243    
1244  Arguments:    else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1245    code     points to the start of the pattern (the bracket)      {
1246    options  the compiling options      ptr += 2;
1247        if (ptr[1] != CHAR_QUESTION_MARK)
1248          {
1249          while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1250          if (*ptr != 0) ptr++;
1251          }
1252        }
1253    
1254  Returns:   the fixed length, or -1 if there is no fixed length,    /* Start with (? but not a condition. */
1255               or -2 if \C was encountered  
1256  */    else
1257        {
1258        ptr += 2;
1259        if (*ptr == CHAR_P) ptr++;                      /* Allow optional P */
1260    
1261        /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1262    
1263        if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1264            ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1265          {
1266          int term;
1267          const uschar *thisname;
1268          *count += 1;
1269          if (name == NULL && *count == lorn) return *count;
1270          term = *ptr++;
1271          if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1272          thisname = ptr;
1273          while (*ptr != term) ptr++;
1274          if (name != NULL && lorn == ptr - thisname &&
1275              strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1276            return *count;
1277          term++;
1278          }
1279        }
1280      }
1281    
1282    /* Past any initial parenthesis handling, scan for parentheses or vertical
1283    bars. Stop if we get to cd->end_pattern. Note that this is important for the
1284    first-pass call when this value is temporarily adjusted to stop at the current
1285    position. So DO NOT change this to a test for binary zero. */
1286    
1287    for (; ptr < cd->end_pattern; ptr++)
1288      {
1289      /* Skip over backslashed characters and also entire \Q...\E */
1290    
1291      if (*ptr == CHAR_BACKSLASH)
1292        {
1293        if (*(++ptr) == 0) goto FAIL_EXIT;
1294        if (*ptr == CHAR_Q) for (;;)
1295          {
1296          while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1297          if (*ptr == 0) goto FAIL_EXIT;
1298          if (*(++ptr) == CHAR_E) break;
1299          }
1300        continue;
1301        }
1302    
1303      /* Skip over character classes; this logic must be similar to the way they
1304      are handled for real. If the first character is '^', skip it. Also, if the
1305      first few characters (either before or after ^) are \Q\E or \E we skip them
1306      too. This makes for compatibility with Perl. Note the use of STR macros to
1307      encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1308    
1309      if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1310        {
1311        BOOL negate_class = FALSE;
1312        for (;;)
1313          {
1314          if (ptr[1] == CHAR_BACKSLASH)
1315            {
1316            if (ptr[2] == CHAR_E)
1317              ptr+= 2;
1318            else if (strncmp((const char *)ptr+2,
1319                     STR_Q STR_BACKSLASH STR_E, 3) == 0)
1320              ptr += 4;
1321            else
1322              break;
1323            }
1324          else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1325            {
1326            negate_class = TRUE;
1327            ptr++;
1328            }
1329          else break;
1330          }
1331    
1332        /* If the next character is ']', it is a data character that must be
1333        skipped, except in JavaScript compatibility mode. */
1334    
1335        if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1336            (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1337          ptr++;
1338    
1339        while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1340          {
1341          if (*ptr == 0) return -1;
1342          if (*ptr == CHAR_BACKSLASH)
1343            {
1344            if (*(++ptr) == 0) goto FAIL_EXIT;
1345            if (*ptr == CHAR_Q) for (;;)
1346              {
1347              while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1348              if (*ptr == 0) goto FAIL_EXIT;
1349              if (*(++ptr) == CHAR_E) break;
1350              }
1351            continue;
1352            }
1353          }
1354        continue;
1355        }
1356    
1357      /* Skip comments in /x mode */
1358    
1359      if (xmode && *ptr == CHAR_NUMBER_SIGN)
1360        {
1361        ptr++;
1362        while (*ptr != 0)
1363          {
1364          if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1365          ptr++;
1366    #ifdef SUPPORT_UTF8
1367          if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
1368    #endif
1369          }
1370        if (*ptr == 0) goto FAIL_EXIT;
1371        continue;
1372        }
1373    
1374      /* Check for the special metacharacters */
1375    
1376      if (*ptr == CHAR_LEFT_PARENTHESIS)
1377        {
1378        int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, count);
1379        if (rc > 0) return rc;
1380        if (*ptr == 0) goto FAIL_EXIT;
1381        }
1382    
1383      else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1384        {
1385        if (dup_parens && *count < hwm_count) *count = hwm_count;
1386        goto FAIL_EXIT;
1387        }
1388    
1389      else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1390        {
1391        if (*count > hwm_count) hwm_count = *count;
1392        *count = start_count;
1393        }
1394      }
1395    
1396    FAIL_EXIT:
1397    *ptrptr = ptr;
1398    return -1;
1399    }
1400    
1401    
1402    
1403    
1404    /*************************************************
1405    *       Find forward referenced subpattern       *
1406    *************************************************/
1407    
1408    /* This function scans along a pattern's text looking for capturing
1409    subpatterns, and counting them. If it finds a named pattern that matches the
1410    name it is given, it returns its number. Alternatively, if the name is NULL, it
1411    returns when it reaches a given numbered subpattern. This is used for forward
1412    references to subpatterns. We used to be able to start this scan from the
1413    current compiling point, using the current count value from cd->bracount, and
1414    do it all in a single loop, but the addition of the possibility of duplicate
1415    subpattern numbers means that we have to scan from the very start, in order to
1416    take account of such duplicates, and to use a recursive function to keep track
1417    of the different types of group.
1418    
1419    Arguments:
1420      cd           compile background data
1421      name         name to seek, or NULL if seeking a numbered subpattern
1422      lorn         name length, or subpattern number if name is NULL
1423      xmode        TRUE if we are in /x mode
1424      utf8         TRUE if we are in UTF-8 mode
1425    
1426    Returns:       the number of the found subpattern, or -1 if not found
1427    */
1428    
1429    static int
1430    find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode,
1431      BOOL utf8)
1432    {
1433    uschar *ptr = (uschar *)cd->start_pattern;
1434    int count = 0;
1435    int rc;
1436    
1437    /* If the pattern does not start with an opening parenthesis, the first call
1438    to find_parens_sub() will scan right to the end (if necessary). However, if it
1439    does start with a parenthesis, find_parens_sub() will return when it hits the
1440    matching closing parens. That is why we have to have a loop. */
1441    
1442    for (;;)
1443      {
1444      rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, &count);
1445      if (rc > 0 || *ptr++ == 0) break;
1446      }
1447    
1448    return rc;
1449    }
1450    
1451    
1452    
1453    
1454    /*************************************************
1455    *      Find first significant op code            *
1456    *************************************************/
1457    
1458    /* This is called by several functions that scan a compiled expression looking
1459    for a fixed first character, or an anchoring op code etc. It skips over things
1460    that do not influence this. For some calls, it makes sense to skip negative
1461    forward and all backward assertions, and also the \b assertion; for others it
1462    does not.
1463    
1464    Arguments:
1465      code         pointer to the start of the group
1466      skipassert   TRUE if certain assertions are to be skipped
1467    
1468    Returns:       pointer to the first significant opcode
1469    */
1470    
1471    static const uschar*
1472    first_significant_code(const uschar *code, BOOL skipassert)
1473    {
1474    for (;;)
1475      {
1476      switch ((int)*code)
1477        {
1478        case OP_ASSERT_NOT:
1479        case OP_ASSERTBACK:
1480        case OP_ASSERTBACK_NOT:
1481        if (!skipassert) return code;
1482        do code += GET(code, 1); while (*code == OP_ALT);
1483        code += _pcre_OP_lengths[*code];
1484        break;
1485    
1486        case OP_WORD_BOUNDARY:
1487        case OP_NOT_WORD_BOUNDARY:
1488        if (!skipassert) return code;
1489        /* Fall through */
1490    
1491        case OP_CALLOUT:
1492        case OP_CREF:
1493        case OP_NCREF:
1494        case OP_RREF:
1495        case OP_NRREF:
1496        case OP_DEF:
1497        code += _pcre_OP_lengths[*code];
1498        break;
1499    
1500        default:
1501        return code;
1502        }
1503      }
1504    /* Control never reaches here */
1505    }
1506    
1507    
1508    
1509    
1510    /*************************************************
1511    *        Find the fixed length of a branch       *
1512    *************************************************/
1513    
1514    /* Scan a branch and compute the fixed length of subject that will match it,
1515    if the length is fixed. This is needed for dealing with backward assertions.
1516    In UTF8 mode, the result is in characters rather than bytes. The branch is
1517    temporarily terminated with OP_END when this function is called.
1518    
1519    This function is called when a backward assertion is encountered, so that if it
1520    fails, the error message can point to the correct place in the pattern.
1521    However, we cannot do this when the assertion contains subroutine calls,
1522    because they can be forward references. We solve this by remembering this case
1523    and doing the check at the end; a flag specifies which mode we are running in.
1524    
1525    Arguments:
1526      code     points to the start of the pattern (the bracket)
1527      utf8     TRUE in UTF-8 mode
1528      atend    TRUE if called when the pattern is complete
1529      cd       the "compile data" structure
1530    
1531    Returns:   the fixed length,
1532                 or -1 if there is no fixed length,
1533                 or -2 if \C was encountered (in UTF-8 mode only)
1534                 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1535                 or -4 if an unknown opcode was encountered (internal error)
1536    */
1537    
1538  static int  static int
1539  find_fixedlength(uschar *code, int options)  find_fixedlength(uschar *code, BOOL utf8, BOOL atend, compile_data *cd)
1540  {  {
1541  int length = -1;  int length = -1;
1542    
# Line 822  branch, check the length against that of Line 1549  branch, check the length against that of
1549  for (;;)  for (;;)
1550    {    {
1551    int d;    int d;
1552      uschar *ce, *cs;
1553    register int op = *cc;    register int op = *cc;
   if (op >= OP_BRA) op = OP_BRA;  
   
1554    switch (op)    switch (op)
1555      {      {
1556        /* We only need to continue for OP_CBRA (normal capturing bracket) and
1557        OP_BRA (normal non-capturing bracket) because the other variants of these
1558        opcodes are all concerned with unlimited repeated groups, which of course
1559        are not of fixed length. */
1560    
1561        case OP_CBRA:
1562      case OP_BRA:      case OP_BRA:
1563      case OP_ONCE:      case OP_ONCE:
1564        case OP_ONCE_NC:
1565      case OP_COND:      case OP_COND:
1566      d = find_fixedlength(cc, options);      d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), utf8, atend, cd);
1567      if (d < 0) return d;      if (d < 0) return d;
1568      branchlength += d;      branchlength += d;
1569      do cc += GET(cc, 1); while (*cc == OP_ALT);      do cc += GET(cc, 1); while (*cc == OP_ALT);
1570      cc += 1 + LINK_SIZE;      cc += 1 + LINK_SIZE;
1571      break;      break;
1572    
1573      /* 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 call.
1574      call. If it's ALT it is an alternation in a nested call. If it is      If it's ALT it is an alternation in a nested call. An ACCEPT is effectively
1575      END it's the end of the outer call. All can be handled by the same code. */      an ALT. If it is END it's the end of the outer call. All can be handled by
1576        the same code. Note that we must not include the OP_KETRxxx opcodes here,
1577        because they all imply an unlimited repeat. */
1578    
1579      case OP_ALT:      case OP_ALT:
1580      case OP_KET:      case OP_KET:
     case OP_KETRMAX:  
     case OP_KETRMIN:  
1581      case OP_END:      case OP_END:
1582        case OP_ACCEPT:
1583        case OP_ASSERT_ACCEPT:
1584      if (length < 0) length = branchlength;      if (length < 0) length = branchlength;
1585        else if (length != branchlength) return -1;        else if (length != branchlength) return -1;
1586      if (*cc != OP_ALT) return length;      if (*cc != OP_ALT) return length;
# Line 853  for (;;) Line 1588  for (;;)
1588      branchlength = 0;      branchlength = 0;
1589      break;      break;
1590    
1591        /* A true recursion implies not fixed length, but a subroutine call may
1592        be OK. If the subroutine is a forward reference, we can't deal with
1593        it until the end of the pattern, so return -3. */
1594    
1595        case OP_RECURSE:
1596        if (!atend) return -3;
1597        cs = ce = (uschar *)cd->start_code + GET(cc, 1);  /* Start subpattern */
1598        do ce += GET(ce, 1); while (*ce == OP_ALT);       /* End subpattern */
1599        if (cc > cs && cc < ce) return -1;                /* Recursion */
1600        d = find_fixedlength(cs + 2, utf8, atend, cd);
1601        if (d < 0) return d;
1602        branchlength += d;
1603        cc += 1 + LINK_SIZE;
1604        break;
1605    
1606      /* Skip over assertive subpatterns */      /* Skip over assertive subpatterns */
1607    
1608      case OP_ASSERT:      case OP_ASSERT:
# Line 864  for (;;) Line 1614  for (;;)
1614    
1615      /* Skip over things that don't match chars */      /* Skip over things that don't match chars */
1616    
1617      case OP_REVERSE:      case OP_MARK:
1618      case OP_BRANUMBER:      case OP_PRUNE_ARG:
1619      case OP_CREF:      case OP_SKIP_ARG:
1620      case OP_OPT:      case OP_THEN_ARG:
1621        cc += cc[1] + _pcre_OP_lengths[*cc];
1622        break;
1623    
1624      case OP_CALLOUT:      case OP_CALLOUT:
     case OP_SOD:  
     case OP_SOM:  
     case OP_EOD:  
     case OP_EODN:  
1625      case OP_CIRC:      case OP_CIRC:
1626        case OP_CIRCM:
1627        case OP_CLOSE:
1628        case OP_COMMIT:
1629        case OP_CREF:
1630        case OP_DEF:
1631      case OP_DOLL:      case OP_DOLL:
1632        case OP_DOLLM:
1633        case OP_EOD:
1634        case OP_EODN:
1635        case OP_FAIL:
1636        case OP_NCREF:
1637        case OP_NRREF:
1638      case OP_NOT_WORD_BOUNDARY:      case OP_NOT_WORD_BOUNDARY:
1639        case OP_PRUNE:
1640        case OP_REVERSE:
1641        case OP_RREF:
1642        case OP_SET_SOM:
1643        case OP_SKIP:
1644        case OP_SOD:
1645        case OP_SOM:
1646        case OP_THEN:
1647      case OP_WORD_BOUNDARY:      case OP_WORD_BOUNDARY:
1648      cc += _pcre_OP_lengths[*cc];      cc += _pcre_OP_lengths[*cc];
1649      break;      break;
# Line 883  for (;;) Line 1651  for (;;)
1651      /* Handle literal characters */      /* Handle literal characters */
1652    
1653      case OP_CHAR:      case OP_CHAR:
1654      case OP_CHARNC:      case OP_CHARI:
1655        case OP_NOT:
1656        case OP_NOTI:
1657      branchlength++;      branchlength++;
1658      cc += 2;      cc += 2;
1659  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
1660      if ((options & PCRE_UTF8) != 0)      if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
       {  
       while ((*cc & 0xc0) == 0x80) cc++;  
       }  
1661  #endif  #endif
1662      break;      break;
1663    
# Line 898  for (;;) Line 1665  for (;;)
1665      need to skip over a multibyte character in UTF8 mode.  */      need to skip over a multibyte character in UTF8 mode.  */
1666    
1667      case OP_EXACT:      case OP_EXACT:
1668        case OP_EXACTI:
1669        case OP_NOTEXACT:
1670        case OP_NOTEXACTI:
1671      branchlength += GET2(cc,1);      branchlength += GET2(cc,1);
1672      cc += 4;      cc += 4;
1673  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
1674      if ((options & PCRE_UTF8) != 0)      if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
       {  
       while((*cc & 0x80) == 0x80) cc++;  
       }  
1675  #endif  #endif
1676      break;      break;
1677    
1678      case OP_TYPEEXACT:      case OP_TYPEEXACT:
1679      branchlength += GET2(cc,1);      branchlength += GET2(cc,1);
1680        if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1681      cc += 4;      cc += 4;
1682      break;      break;
1683    
# Line 917  for (;;) Line 1685  for (;;)
1685    
1686      case OP_PROP:      case OP_PROP:
1687      case OP_NOTPROP:      case OP_NOTPROP:
1688      cc++;      cc += 2;
1689      /* Fall through */      /* Fall through */
1690    
1691        case OP_HSPACE:
1692        case OP_VSPACE:
1693        case OP_NOT_HSPACE:
1694        case OP_NOT_VSPACE:
1695      case OP_NOT_DIGIT:      case OP_NOT_DIGIT:
1696      case OP_DIGIT:      case OP_DIGIT:
1697      case OP_NOT_WHITESPACE:      case OP_NOT_WHITESPACE:
# Line 927  for (;;) Line 1699  for (;;)
1699      case OP_NOT_WORDCHAR:      case OP_NOT_WORDCHAR:
1700      case OP_WORDCHAR:      case OP_WORDCHAR:
1701      case OP_ANY:      case OP_ANY:
1702        case OP_ALLANY:
1703      branchlength++;      branchlength++;
1704      cc++;      cc++;
1705      break;      break;
1706    
1707      /* The single-byte matcher isn't allowed */      /* The single-byte matcher isn't allowed. This only happens in UTF-8 mode;
1708        otherwise \C is coded as OP_ALLANY. */
1709    
1710      case OP_ANYBYTE:      case OP_ANYBYTE:
1711      return -2;      return -2;
# Line 950  for (;;) Line 1724  for (;;)
1724    
1725      switch (*cc)      switch (*cc)
1726        {        {
1727          case OP_CRPLUS:
1728          case OP_CRMINPLUS:
1729        case OP_CRSTAR:        case OP_CRSTAR:
1730        case OP_CRMINSTAR:        case OP_CRMINSTAR:
1731        case OP_CRQUERY:        case OP_CRQUERY:
# Line 970  for (;;) Line 1746  for (;;)
1746    
1747      /* Anything else is variable length */      /* Anything else is variable length */
1748    
1749      default:      case OP_ANYNL:
1750        case OP_BRAMINZERO:
1751        case OP_BRAPOS:
1752        case OP_BRAPOSZERO:
1753        case OP_BRAZERO:
1754        case OP_CBRAPOS:
1755        case OP_EXTUNI:
1756        case OP_KETRMAX:
1757        case OP_KETRMIN:
1758        case OP_KETRPOS:
1759        case OP_MINPLUS:
1760        case OP_MINPLUSI:
1761        case OP_MINQUERY:
1762        case OP_MINQUERYI:
1763        case OP_MINSTAR:
1764        case OP_MINSTARI:
1765        case OP_MINUPTO:
1766        case OP_MINUPTOI:
1767        case OP_NOTMINPLUS:
1768        case OP_NOTMINPLUSI:
1769        case OP_NOTMINQUERY:
1770        case OP_NOTMINQUERYI:
1771        case OP_NOTMINSTAR:
1772        case OP_NOTMINSTARI:
1773        case OP_NOTMINUPTO:
1774        case OP_NOTMINUPTOI:
1775        case OP_NOTPLUS:
1776        case OP_NOTPLUSI:
1777        case OP_NOTPOSPLUS:
1778        case OP_NOTPOSPLUSI:
1779        case OP_NOTPOSQUERY:
1780        case OP_NOTPOSQUERYI:
1781        case OP_NOTPOSSTAR:
1782        case OP_NOTPOSSTARI:
1783        case OP_NOTPOSUPTO:
1784        case OP_NOTPOSUPTOI:
1785        case OP_NOTQUERY:
1786        case OP_NOTQUERYI:
1787        case OP_NOTSTAR:
1788        case OP_NOTSTARI:
1789        case OP_NOTUPTO:
1790        case OP_NOTUPTOI:
1791        case OP_PLUS:
1792        case OP_PLUSI:
1793        case OP_POSPLUS:
1794        case OP_POSPLUSI:
1795        case OP_POSQUERY:
1796        case OP_POSQUERYI:
1797        case OP_POSSTAR:
1798        case OP_POSSTARI:
1799        case OP_POSUPTO:
1800        case OP_POSUPTOI:
1801        case OP_QUERY:
1802        case OP_QUERYI:
1803        case OP_REF:
1804        case OP_REFI:
1805        case OP_SBRA:
1806        case OP_SBRAPOS:
1807        case OP_SCBRA:
1808        case OP_SCBRAPOS:
1809        case OP_SCOND:
1810        case OP_SKIPZERO:
1811        case OP_STAR:
1812        case OP_STARI:
1813        case OP_TYPEMINPLUS:
1814        case OP_TYPEMINQUERY:
1815        case OP_TYPEMINSTAR:
1816        case OP_TYPEMINUPTO:
1817        case OP_TYPEPLUS:
1818        case OP_TYPEPOSPLUS:
1819        case OP_TYPEPOSQUERY:
1820        case OP_TYPEPOSSTAR:
1821        case OP_TYPEPOSUPTO:
1822        case OP_TYPEQUERY:
1823        case OP_TYPESTAR:
1824        case OP_TYPEUPTO:
1825        case OP_UPTO:
1826        case OP_UPTOI:
1827      return -1;      return -1;
1828    
1829        /* Catch unrecognized opcodes so that when new ones are added they
1830        are not forgotten, as has happened in the past. */
1831    
1832        default:
1833        return -4;
1834      }      }
1835    }    }
1836  /* Control never gets here */  /* Control never gets here */
# Line 981  for (;;) Line 1840  for (;;)
1840    
1841    
1842  /*************************************************  /*************************************************
1843  *    Scan compiled regex for numbered bracket    *  *    Scan compiled regex for specific bracket    *
1844  *************************************************/  *************************************************/
1845    
1846  /* This little function scans through a compiled pattern until it finds a  /* This little function scans through a compiled pattern until it finds a
1847  capturing bracket with the given number.  capturing bracket with the given number, or, if the number is negative, an
1848    instance of OP_REVERSE for a lookbehind. The function is global in the C sense
1849    so that it can be called from pcre_study() when finding the minimum matching
1850    length.
1851    
1852  Arguments:  Arguments:
1853    code        points to start of expression    code        points to start of expression
1854    utf8        TRUE in UTF-8 mode    utf8        TRUE in UTF-8 mode
1855    number      the required bracket number    number      the required bracket number or negative to find a lookbehind
1856    
1857  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
1858  */  */
1859    
1860  static const uschar *  const uschar *
1861  find_bracket(const uschar *code, BOOL utf8, int number)  _pcre_find_bracket(const uschar *code, BOOL utf8, int number)
1862  {  {
 #ifndef SUPPORT_UTF8  
 utf8 = utf8;               /* Stop pedantic compilers complaining */  
 #endif  
   
1863  for (;;)  for (;;)
1864    {    {
1865    register int c = *code;    register int c = *code;
1866    
1867    if (c == OP_END) return NULL;    if (c == OP_END) return NULL;
1868    else if (c > OP_BRA)  
1869      /* XCLASS is used for classes that cannot be represented just by a bit
1870      map. This includes negated single high-valued characters. The length in
1871      the table is zero; the actual length is stored in the compiled code. */
1872    
1873      if (c == OP_XCLASS) code += GET(code, 1);
1874    
1875      /* Handle recursion */
1876    
1877      else if (c == OP_REVERSE)
1878        {
1879        if (number < 0) return (uschar *)code;
1880        code += _pcre_OP_lengths[c];
1881        }
1882    
1883      /* Handle capturing bracket */
1884    
1885      else if (c == OP_CBRA || c == OP_SCBRA ||
1886               c == OP_CBRAPOS || c == OP_SCBRAPOS)
1887      {      {
1888      int n = c - OP_BRA;      int n = GET2(code, 1+LINK_SIZE);
     if (n > EXTRACT_BASIC_MAX) n = GET2(code, 2+LINK_SIZE);  
1889      if (n == number) return (uschar *)code;      if (n == number) return (uschar *)code;
1890      code += _pcre_OP_lengths[OP_BRA];      code += _pcre_OP_lengths[c];
1891      }      }
1892    
1893      /* Otherwise, we can get the item's length from the table, except that for
1894      repeated character types, we have to test for \p and \P, which have an extra
1895      two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1896      must add in its length. */
1897    
1898    else    else
1899      {      {
1900      code += _pcre_OP_lengths[c];      switch(c)
1901          {
1902          case OP_TYPESTAR:
1903          case OP_TYPEMINSTAR:
1904          case OP_TYPEPLUS:
1905          case OP_TYPEMINPLUS:
1906          case OP_TYPEQUERY:
1907          case OP_TYPEMINQUERY:
1908          case OP_TYPEPOSSTAR:
1909          case OP_TYPEPOSPLUS:
1910          case OP_TYPEPOSQUERY:
1911          if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1912          break;
1913    
1914  #ifdef SUPPORT_UTF8        case OP_TYPEUPTO:
1915          case OP_TYPEMINUPTO:
1916          case OP_TYPEEXACT:
1917          case OP_TYPEPOSUPTO:
1918          if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1919          break;
1920    
1921      /* In UTF-8 mode, opcodes that are followed by a character may be followed        case OP_MARK:
1922      by a multi-byte character. The length in the table is a minimum, so we have        case OP_PRUNE_ARG:
1923      to scan along to skip the extra bytes. All opcodes are less than 128, so we        case OP_SKIP_ARG:
1924      can use relatively efficient code. */        code += code[1];
1925          break;
1926    
1927          case OP_THEN_ARG:
1928          code += code[1];
1929          break;
1930          }
1931    
1932        /* Add in the fixed length from the table */
1933    
1934        code += _pcre_OP_lengths[c];
1935    
1936      /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1937      a multi-byte character. The length in the table is a minimum, so we have to
1938      arrange to skip the extra bytes. */
1939    
1940    #ifdef SUPPORT_UTF8
1941      if (utf8) switch(c)      if (utf8) switch(c)
1942        {        {
1943        case OP_CHAR:        case OP_CHAR:
1944        case OP_CHARNC:        case OP_CHARI:
1945        case OP_EXACT:        case OP_EXACT:
1946          case OP_EXACTI:
1947        case OP_UPTO:        case OP_UPTO:
1948          case OP_UPTOI:
1949        case OP_MINUPTO:        case OP_MINUPTO:
1950          case OP_MINUPTOI:
1951          case OP_POSUPTO:
1952          case OP_POSUPTOI:
1953        case OP_STAR:        case OP_STAR:
1954          case OP_STARI:
1955        case OP_MINSTAR:        case OP_MINSTAR:
1956          case OP_MINSTARI:
1957          case OP_POSSTAR:
1958          case OP_POSSTARI:
1959        case OP_PLUS:        case OP_PLUS:
1960          case OP_PLUSI:
1961        case OP_MINPLUS:        case OP_MINPLUS:
1962          case OP_MINPLUSI:
1963          case OP_POSPLUS:
1964          case OP_POSPLUSI:
1965        case OP_QUERY:        case OP_QUERY:
1966          case OP_QUERYI:
1967        case OP_MINQUERY:        case OP_MINQUERY:
1968        while ((*code & 0xc0) == 0x80) code++;        case OP_MINQUERYI:
1969        break;        case OP_POSQUERY:
1970          case OP_POSQUERYI:
1971        /* 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;  
1972        break;        break;
1973        }        }
1974    #else
1975        (void)(utf8);  /* Keep compiler happy by referencing function argument */
1976  #endif  #endif
1977      }      }
1978    }    }
# Line 1072  Returns: pointer to the opcode for Line 1997  Returns: pointer to the opcode for
1997  static const uschar *  static const uschar *
1998  find_recurse(const uschar *code, BOOL utf8)  find_recurse(const uschar *code, BOOL utf8)
1999  {  {
 #ifndef SUPPORT_UTF8  
 utf8 = utf8;               /* Stop pedantic compilers complaining */  
 #endif  
   
2000  for (;;)  for (;;)
2001    {    {
2002    register int c = *code;    register int c = *code;
2003    if (c == OP_END) return NULL;    if (c == OP_END) return NULL;
2004    else if (c == OP_RECURSE) return code;    if (c == OP_RECURSE) return code;
2005    else if (c > OP_BRA)  
2006      {    /* XCLASS is used for classes that cannot be represented just by a bit
2007      code += _pcre_OP_lengths[OP_BRA];    map. This includes negated single high-valued characters. The length in
2008      }    the table is zero; the actual length is stored in the compiled code. */
2009    
2010      if (c == OP_XCLASS) code += GET(code, 1);
2011    
2012      /* Otherwise, we can get the item's length from the table, except that for
2013      repeated character types, we have to test for \p and \P, which have an extra
2014      two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
2015      must add in its length. */
2016    
2017    else    else
2018      {      {
2019      code += _pcre_OP_lengths[c];      switch(c)
2020          {
2021          case OP_TYPESTAR:
2022          case OP_TYPEMINSTAR:
2023          case OP_TYPEPLUS:
2024          case OP_TYPEMINPLUS:
2025          case OP_TYPEQUERY:
2026          case OP_TYPEMINQUERY:
2027          case OP_TYPEPOSSTAR:
2028          case OP_TYPEPOSPLUS:
2029          case OP_TYPEPOSQUERY:
2030          if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2031          break;
2032    
2033  #ifdef SUPPORT_UTF8        case OP_TYPEPOSUPTO:
2034          case OP_TYPEUPTO:
2035          case OP_TYPEMINUPTO:
2036          case OP_TYPEEXACT:
2037          if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
2038          break;
2039    
2040      /* In UTF-8 mode, opcodes that are followed by a character may be followed        case OP_MARK:
2041      by a multi-byte character. The length in the table is a minimum, so we have        case OP_PRUNE_ARG:
2042      to scan along to skip the extra bytes. All opcodes are less than 128, so we        case OP_SKIP_ARG:
2043      can use relatively efficient code. */        code += code[1];
2044          break;
2045    
2046      if (utf8) switch(c)        case OP_THEN_ARG:
2047        {        code += code[1];
2048        case OP_CHAR:        break;
2049        case OP_CHARNC:        }
2050    
2051        /* Add in the fixed length from the table */
2052    
2053        code += _pcre_OP_lengths[c];
2054    
2055        /* In UTF-8 mode, opcodes that are followed by a character may be followed
2056        by a multi-byte character. The length in the table is a minimum, so we have
2057        to arrange to skip the extra bytes. */
2058    
2059    #ifdef SUPPORT_UTF8
2060        if (utf8) switch(c)
2061          {
2062          case OP_CHAR:
2063          case OP_CHARI:
2064        case OP_EXACT:        case OP_EXACT:
2065          case OP_EXACTI:
2066        case OP_UPTO:        case OP_UPTO:
2067          case OP_UPTOI:
2068        case OP_MINUPTO:        case OP_MINUPTO:
2069          case OP_MINUPTOI:
2070          case OP_POSUPTO:
2071          case OP_POSUPTOI:
2072        case OP_STAR:        case OP_STAR:
2073          case OP_STARI:
2074        case OP_MINSTAR:        case OP_MINSTAR:
2075          case OP_MINSTARI:
2076          case OP_POSSTAR:
2077          case OP_POSSTARI:
2078        case OP_PLUS:        case OP_PLUS:
2079          case OP_PLUSI:
2080        case OP_MINPLUS:        case OP_MINPLUS:
2081          case OP_MINPLUSI:
2082          case OP_POSPLUS:
2083          case OP_POSPLUSI:
2084        case OP_QUERY:        case OP_QUERY:
2085          case OP_QUERYI:
2086        case OP_MINQUERY:        case OP_MINQUERY:
2087        while ((*code & 0xc0) == 0x80) code++;        case OP_MINQUERYI:
2088        break;        case OP_POSQUERY:
2089          case OP_POSQUERYI:
2090        /* 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;  
2091        break;        break;
2092        }        }
2093    #else
2094        (void)(utf8);  /* Keep compiler happy by referencing function argument */
2095  #endif  #endif
2096      }      }
2097    }    }
# Line 1132  for (;;) Line 2104  for (;;)
2104  *************************************************/  *************************************************/
2105    
2106  /* 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
2107  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()
2108  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
2109  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
2110  whose current branch will already have been scanned.  backward and negative forward assertions when its final argument is TRUE. If we
2111    hit an unclosed bracket, we return "empty" - this means we've struck an inner
2112    bracket whose current branch will already have been scanned.
2113    
2114  Arguments:  Arguments:
2115    code        points to start of search    code        points to start of search
2116    endcode     points to where to stop    endcode     points to where to stop
2117    utf8        TRUE if in UTF8 mode    utf8        TRUE if in UTF8 mode
2118      cd          contains pointers to tables etc.
2119    
2120  Returns:      TRUE if what is matched could be empty  Returns:      TRUE if what is matched could be empty
2121  */  */
2122    
2123  static BOOL  static BOOL
2124  could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8)  could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8,
2125      compile_data *cd)
2126  {  {
2127  register int c;  register int c;
2128  for (code = first_significant_code(code + 1 + LINK_SIZE, NULL, 0, TRUE);  for (code = first_significant_code(code + _pcre_OP_lengths[*code], TRUE);
2129       code < endcode;       code < endcode;
2130       code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))       code = first_significant_code(code + _pcre_OP_lengths[c], TRUE))
2131    {    {
2132    const uschar *ccode;    const uschar *ccode;
2133    
2134    c = *code;    c = *code;
2135    
2136    if (c >= OP_BRA)    /* Skip over forward assertions; the other assertions are skipped by
2137      first_significant_code() with a TRUE final argument. */
2138    
2139      if (c == OP_ASSERT)
2140        {
2141        do code += GET(code, 1); while (*code == OP_ALT);
2142        c = *code;
2143        continue;
2144        }
2145    
2146      /* For a recursion/subroutine call, if its end has been reached, which
2147      implies a backward reference subroutine call, we can scan it. If it's a
2148      forward reference subroutine call, we can't. To detect forward reference
2149      we have to scan up the list that is kept in the workspace. This function is
2150      called only when doing the real compile, not during the pre-compile that
2151      measures the size of the compiled pattern. */
2152    
2153      if (c == OP_RECURSE)
2154      {      {
2155        const uschar *scode;
2156      BOOL empty_branch;      BOOL empty_branch;
     if (GET(code, 1) == 0) return TRUE;    /* Hit unclosed bracket */  
2157    
2158      /* Scan a closed bracket */      /* Test for forward reference */
2159    
2160        for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
2161          if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
2162    
2163        /* Not a forward reference, test for completed backward reference */
2164    
2165      empty_branch = FALSE;      empty_branch = FALSE;
2166        scode = cd->start_code + GET(code, 1);
2167        if (GET(scode, 1) == 0) return TRUE;    /* Unclosed */
2168    
2169        /* Completed backwards reference */
2170    
2171      do      do
2172        {        {
2173        if (!empty_branch && could_be_empty_branch(code, endcode, utf8))        if (could_be_empty_branch(scode, endcode, utf8, cd))
2174            {
2175          empty_branch = TRUE;          empty_branch = TRUE;
2176            break;
2177            }
2178          scode += GET(scode, 1);
2179          }
2180        while (*scode == OP_ALT);
2181    
2182        if (!empty_branch) return FALSE;  /* All branches are non-empty */
2183        continue;
2184        }
2185    
2186      /* Groups with zero repeats can of course be empty; skip them. */
2187    
2188      if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2189          c == OP_BRAPOSZERO)
2190        {
2191        code += _pcre_OP_lengths[c];
2192        do code += GET(code, 1); while (*code == OP_ALT);
2193        c = *code;
2194        continue;
2195        }
2196    
2197      /* A nested group that is already marked as "could be empty" can just be
2198      skipped. */
2199    
2200      if (c == OP_SBRA  || c == OP_SBRAPOS ||
2201          c == OP_SCBRA || c == OP_SCBRAPOS)
2202        {
2203        do code += GET(code, 1); while (*code == OP_ALT);
2204        c = *code;
2205        continue;
2206        }
2207    
2208      /* For other groups, scan the branches. */
2209    
2210      if (c == OP_BRA  || c == OP_BRAPOS ||
2211          c == OP_CBRA || c == OP_CBRAPOS ||
2212          c == OP_ONCE || c == OP_ONCE_NC ||
2213          c == OP_COND)
2214        {
2215        BOOL empty_branch;
2216        if (GET(code, 1) == 0) return TRUE;    /* Hit unclosed bracket */
2217    
2218        /* If a conditional group has only one branch, there is a second, implied,
2219        empty branch, so just skip over the conditional, because it could be empty.
2220        Otherwise, scan the individual branches of the group. */
2221    
2222        if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2223        code += GET(code, 1);        code += GET(code, 1);
2224        else
2225          {
2226          empty_branch = FALSE;
2227          do
2228            {
2229            if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
2230              empty_branch = TRUE;
2231            code += GET(code, 1);
2232            }
2233          while (*code == OP_ALT);
2234          if (!empty_branch) return FALSE;   /* All branches are non-empty */
2235        }        }
2236      while (*code == OP_ALT);  
     if (!empty_branch) return FALSE;   /* All branches are non-empty */  
     code += 1 + LINK_SIZE;  
2237      c = *code;      c = *code;
2238        continue;
2239      }      }
2240    
2241    else switch (c)    /* Handle the other opcodes */
2242    
2243      switch (c)
2244      {      {
2245      /* Check for quantifiers after a class */      /* Check for quantifiers after a class. XCLASS is used for classes that
2246        cannot be represented just by a bit map. This includes negated single
2247        high-valued characters. The length in _pcre_OP_lengths[] is zero; the
2248        actual length is stored in the compiled code, so we must update "code"
2249        here. */
2250    
2251  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
2252      case OP_XCLASS:      case OP_XCLASS:
2253      ccode = code + GET(code, 1);      ccode = code += GET(code, 1);
2254      goto CHECK_CLASS_REPEAT;      goto CHECK_CLASS_REPEAT;
2255  #endif  #endif
2256    
# Line 1227  for (code = first_significant_code(code Line 2294  for (code = first_significant_code(code
2294      case OP_NOT_WORDCHAR:      case OP_NOT_WORDCHAR:
2295      case OP_WORDCHAR:      case OP_WORDCHAR:
2296      case OP_ANY:      case OP_ANY:
2297        case OP_ALLANY:
2298      case OP_ANYBYTE:      case OP_ANYBYTE:
2299      case OP_CHAR:      case OP_CHAR:
2300      case OP_CHARNC:      case OP_CHARI:
2301      case OP_NOT:      case OP_NOT:
2302        case OP_NOTI:
2303      case OP_PLUS:      case OP_PLUS:
2304      case OP_MINPLUS:      case OP_MINPLUS:
2305        case OP_POSPLUS:
2306      case OP_EXACT:      case OP_EXACT:
2307      case OP_NOTPLUS:      case OP_NOTPLUS:
2308      case OP_NOTMINPLUS:      case OP_NOTMINPLUS:
2309        case OP_NOTPOSPLUS:
2310      case OP_NOTEXACT:      case OP_NOTEXACT:
2311      case OP_TYPEPLUS:      case OP_TYPEPLUS:
2312      case OP_TYPEMINPLUS:      case OP_TYPEMINPLUS:
2313        case OP_TYPEPOSPLUS:
2314      case OP_TYPEEXACT:      case OP_TYPEEXACT:
2315      return FALSE;      return FALSE;
2316    
2317        /* These are going to continue, as they may be empty, but we have to
2318        fudge the length for the \p and \P cases. */
2319    
2320        case OP_TYPESTAR:
2321        case OP_TYPEMINSTAR:
2322        case OP_TYPEPOSSTAR:
2323        case OP_TYPEQUERY:
2324        case OP_TYPEMINQUERY:
2325        case OP_TYPEPOSQUERY:
2326        if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2327        break;
2328    
2329        /* Same for these */
2330    
2331        case OP_TYPEUPTO:
2332        case OP_TYPEMINUPTO:
2333        case OP_TYPEPOSUPTO:
2334        if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
2335        break;
2336    
2337      /* End of branch */      /* End of branch */
2338    
2339      case OP_KET:      case OP_KET:
2340      case OP_KETRMAX:      case OP_KETRMAX:
2341      case OP_KETRMIN:      case OP_KETRMIN:
2342        case OP_KETRPOS:
2343      case OP_ALT:      case OP_ALT:
2344      return TRUE;      return TRUE;
2345    
2346      /* In UTF-8 mode, STAR, MINSTAR, QUERY, MINQUERY, UPTO, and MINUPTO  may be      /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2347      followed by a multibyte character */      MINUPTO, and POSUPTO may be followed by a multibyte character */
2348    
2349  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
2350      case OP_STAR:      case OP_STAR:
2351        case OP_STARI:
2352      case OP_MINSTAR:      case OP_MINSTAR:
2353        case OP_MINSTARI:
2354        case OP_POSSTAR:
2355        case OP_POSSTARI:
2356      case OP_QUERY:      case OP_QUERY:
2357        case OP_QUERYI:
2358      case OP_MINQUERY:      case OP_MINQUERY:
2359        case OP_MINQUERYI:
2360        case OP_POSQUERY:
2361        case OP_POSQUERYI:
2362        if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
2363        break;
2364    
2365      case OP_UPTO:      case OP_UPTO:
2366        case OP_UPTOI:
2367      case OP_MINUPTO:      case OP_MINUPTO:
2368      if (utf8) while ((code[2] & 0xc0) == 0x80) code++;      case OP_MINUPTOI:
2369        case OP_POSUPTO:
2370        case OP_POSUPTOI:
2371        if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
2372      break;      break;
2373  #endif  #endif
2374    
2375        /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2376        string. */
2377    
2378        case OP_MARK:
2379        case OP_PRUNE_ARG:
2380        case OP_SKIP_ARG:
2381        code += code[1];
2382        break;
2383    
2384        case OP_THEN_ARG:
2385        code += code[1];
2386        break;
2387    
2388        /* None of the remaining opcodes are required to match a character. */
2389    
2390        default:
2391        break;
2392      }      }
2393    }    }
2394    
# Line 1279  return TRUE; Line 2405  return TRUE;
2405  the current branch of the current pattern to see if it could match the empty  the current branch of the current pattern to see if it could match the empty
2406  string. If it could, we must look outwards for branches at other levels,  string. If it could, we must look outwards for branches at other levels,
2407  stopping when we pass beyond the bracket which is the subject of the recursion.  stopping when we pass beyond the bracket which is the subject of the recursion.
2408    This function is called only during the real compile, not during the
2409    pre-compile.
2410    
2411  Arguments:  Arguments:
2412    code        points to start of the recursion    code        points to start of the recursion
2413    endcode     points to where to stop (current RECURSE item)    endcode     points to where to stop (current RECURSE item)
2414    bcptr       points to the chain of current (unclosed) branch starts    bcptr       points to the chain of current (unclosed) branch starts
2415    utf8        TRUE if in UTF-8 mode    utf8        TRUE if in UTF-8 mode
2416      cd          pointers to tables etc
2417    
2418  Returns:      TRUE if what is matched could be empty  Returns:      TRUE if what is matched could be empty
2419  */  */
2420    
2421  static BOOL  static BOOL
2422  could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,  could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
2423    BOOL utf8)    BOOL utf8, compile_data *cd)
2424  {  {
2425  while (bcptr != NULL && bcptr->current >= code)  while (bcptr != NULL && bcptr->current_branch >= code)
2426    {    {
2427    if (!could_be_empty_branch(bcptr->current, endcode, utf8)) return FALSE;    if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
2428        return FALSE;
2429    bcptr = bcptr->outer;    bcptr = bcptr->outer;
2430    }    }
2431  return TRUE;  return TRUE;
# Line 1308  return TRUE; Line 2438  return TRUE;
2438  *************************************************/  *************************************************/
2439    
2440  /* This function is called when the sequence "[:" or "[." or "[=" is  /* This function is called when the sequence "[:" or "[." or "[=" is
2441  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
2442  optional ^ and then a sequence of letters, terminated by a matching ":]" or  sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2443  ".]" or "=]".  reach an unescaped ']' without the special preceding character, return FALSE.
2444    
2445    Originally, this function only recognized a sequence of letters between the
2446    terminators, but it seems that Perl recognizes any sequence of characters,
2447    though of course unknown POSIX names are subsequently rejected. Perl gives an
2448    "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2449    didn't consider this to be a POSIX class. Likewise for [:1234:].
2450    
2451    The problem in trying to be exactly like Perl is in the handling of escapes. We
2452    have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2453    class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2454    below handles the special case of \], but does not try to do any other escape
2455    processing. This makes it different from Perl for cases such as [:l\ower:]
2456    where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2457    "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2458    I think.
2459    
2460    A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
2461    It seems that the appearance of a nested POSIX class supersedes an apparent
2462    external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
2463    a digit.
2464    
2465    In Perl, unescaped square brackets may also appear as part of class names. For
2466    example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
2467    [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
2468    seem right at all. PCRE does not allow closing square brackets in POSIX class
2469    names.
2470    
2471  Argument:  Arguments:
2472    ptr      pointer to the initial [    ptr      pointer to the initial [
2473    endptr   where to return the end pointer    endptr   where to return the end pointer
   cd       pointer to compile data  
2474    
2475  Returns:   TRUE or FALSE  Returns:   TRUE or FALSE
2476  */  */
2477    
2478  static BOOL  static BOOL
2479  check_posix_syntax(const uschar *ptr, const uschar **endptr, compile_data *cd)  check_posix_syntax(const uschar *ptr, const uschar **endptr)
2480  {  {
2481  int terminator;          /* Don't combine these lines; the Solaris cc */  int terminator;          /* Don't combine these lines; the Solaris cc */
2482  terminator = *(++ptr);   /* compiler warns about "non-constant" initializer. */  terminator = *(++ptr);   /* compiler warns about "non-constant" initializer. */
2483  if (*(++ptr) == '^') ptr++;  for (++ptr; *ptr != 0; ptr++)
 while ((cd->ctypes[*ptr] & ctype_letter) != 0) ptr++;  
 if (*ptr == terminator && ptr[1] == ']')  
2484    {    {
2485    *endptr = ptr;    if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2486    return TRUE;      ptr++;
2487      else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2488      else
2489        {
2490        if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2491          {
2492          *endptr = ptr;
2493          return TRUE;
2494          }
2495        if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
2496             (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2497              ptr[1] == CHAR_EQUALS_SIGN) &&
2498            check_posix_syntax(ptr, endptr))
2499          return FALSE;
2500        }
2501    }    }
2502  return FALSE;  return FALSE;
2503  }  }
# Line 1355  Returns: a value representing the na Line 2522  Returns: a value representing the na
2522  static int  static int
2523  check_posix_name(const uschar *ptr, int len)  check_posix_name(const uschar *ptr, int len)
2524  {  {
2525    const char *pn = posix_names;
2526  register int yield = 0;  register int yield = 0;
2527  while (posix_name_lengths[yield] != 0)  while (posix_name_lengths[yield] != 0)
2528    {    {
2529    if (len == posix_name_lengths[yield] &&    if (len == posix_name_lengths[yield] &&
2530      strncmp((const char *)ptr, posix_names[yield], len) == 0) return yield;      strncmp((const char *)ptr, pn, len) == 0) return yield;
2531      pn += posix_name_lengths[yield] + 1;
2532    yield++;    yield++;
2533    }    }
2534  return -1;  return -1;
# Line 1374  return -1; Line 2543  return -1;
2543  that is referenced. This means that groups can be replicated for fixed  that is referenced. This means that groups can be replicated for fixed
2544  repetition simply by copying (because the recursion is allowed to refer to  repetition simply by copying (because the recursion is allowed to refer to
2545  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
2546  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
2547  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
2548  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
2549  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
2550  partially compiled regex must be temporarily terminated with OP_END.  is called, the partially compiled regex must be temporarily terminated with
2551    OP_END.
2552    
2553    This function has been extended with the possibility of forward references for
2554    recursions and subroutine calls. It must also check the list of such references
2555    for the group we are dealing with. If it finds that one of the recursions in
2556    the current group is on this list, it adjusts the offset in the list, not the
2557    value in the reference (which is a group number).
2558    
2559  Arguments:  Arguments:
2560    group      points to the start of the group    group      points to the start of the group
2561    adjust     the amount by which the group is to be moved    adjust     the amount by which the group is to be moved
2562    utf8       TRUE in UTF-8 mode    utf8       TRUE in UTF-8 mode
2563    cd         contains pointers to tables etc.    cd         contains pointers to tables etc.
2564      save_hwm   the hwm forward reference pointer at the start of the group
2565    
2566  Returns:     nothing  Returns:     nothing
2567  */  */
2568    
2569  static void  static void
2570  adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd)  adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
2571      uschar *save_hwm)
2572  {  {
2573  uschar *ptr = group;  uschar *ptr = group;
2574    
2575  while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)  while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
2576    {    {
2577    int offset = GET(ptr, 1);    int offset;
2578    if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);    uschar *hc;
2579    
2580      /* See if this recursion is on the forward reference list. If so, adjust the
2581      reference. */
2582    
2583      for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2584        {
2585        offset = GET(hc, 0);
2586        if (cd->start_code + offset == ptr + 1)
2587          {
2588          PUT(hc, 0, offset + adjust);
2589          break;
2590          }
2591        }
2592    
2593      /* Otherwise, adjust the recursion offset if it's after the start of this
2594      group. */
2595    
2596      if (hc >= cd->hwm)
2597        {
2598        offset = GET(ptr, 1);
2599        if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2600        }
2601    
2602    ptr += 1 + LINK_SIZE;    ptr += 1 + LINK_SIZE;
2603    }    }
2604  }  }
# Line 1423  auto_callout(uschar *code, const uschar Line 2625  auto_callout(uschar *code, const uschar
2625  {  {
2626  *code++ = OP_CALLOUT;  *code++ = OP_CALLOUT;
2627  *code++ = 255;  *code++ = 255;
2628  PUT(code, 0, ptr - cd->start_pattern);  /* Pattern offset */  PUT(code, 0, (int)(ptr - cd->start_pattern));  /* Pattern offset */
2629  PUT(code, LINK_SIZE, 0);                /* Default length */  PUT(code, LINK_SIZE, 0);                       /* Default length */
2630  return code + 2*LINK_SIZE;  return code + 2*LINK_SIZE;
2631  }  }
2632    
# Line 1449  Returns: nothing Line 2651  Returns: nothing
2651  static void  static void
2652  complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)  complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2653  {  {
2654  int length = ptr - cd->start_pattern - GET(previous_callout, 2);  int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2655  PUT(previous_callout, 2 + LINK_SIZE, length);  PUT(previous_callout, 2 + LINK_SIZE, length);
2656  }  }
2657    
# Line 1475  Yield: TRUE when range returned; Line 2677  Yield: TRUE when range returned;
2677  */  */
2678    
2679  static BOOL  static BOOL
2680  get_othercase_range(int *cptr, int d, int *ocptr, int *odptr)  get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2681      unsigned int *odptr)
2682  {  {
2683  int c, chartype, othercase, next;  unsigned int c, othercase, next;
2684    
2685  for (c = *cptr; c <= d; c++)  for (c = *cptr; c <= d; c++)
2686    {    { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
   if (_pcre_ucp_findchar(c, &chartype, &othercase) == ucp_L && othercase != 0)  
     break;  
   }  
2687    
2688  if (c > d) return FALSE;  if (c > d) return FALSE;
2689    
# Line 1492  next = othercase + 1; Line 2692  next = othercase + 1;
2692    
2693  for (++c; c <= d; c++)  for (++c; c <= d; c++)
2694    {    {
2695    if (_pcre_ucp_findchar(c, &chartype, &othercase) != ucp_L ||    if (UCD_OTHERCASE(c) != next) break;
         othercase != next)  
     break;  
2696    next++;    next++;
2697    }    }
2698    
# Line 1503  for (++c; c <= d; c++) Line 2701  for (++c; c <= d; c++)
2701    
2702  return TRUE;  return TRUE;
2703  }  }
2704  #endif  /* SUPPORT_UCP */  
2705    
2706    
2707  /*************************************************  /*************************************************
2708  *           Compile one branch                   *  *        Check a character and a property        *
2709  *************************************************/  *************************************************/
2710    
2711  /* 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
2712  changed during the branch, the pointer is used to change the external options  is adjacent to a fixed character.
 bits.  
2713    
2714  Arguments:  Arguments:
2715    optionsptr     pointer to the option bits    c            the character
2716    brackets       points to number of extracting brackets used    ptype        the property type
2717    codeptr        points to the pointer to the current code point    pdata        the data for the type
2718    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.  
2719    
2720  Returns:         TRUE on success  Returns:       TRUE if auto-possessifying is OK
                  FALSE, with *errorcodeptr set non-zero on error  
2721  */  */
2722    
2723  static BOOL  static BOOL
2724  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)  
2725  {  {
2726  int repeat_type, op_type;  const ucd_record *prop = GET_UCD(c);
2727  int repeat_min = 0, repeat_max = 0;      /* To please picky compilers */  switch(ptype)
2728  int bravalue = 0;    {
2729  int greedy_default, greedy_non_default;    case PT_LAMP:
2730  int firstbyte, reqbyte;    return (prop->chartype == ucp_Lu ||
2731  int zeroreqbyte, zerofirstbyte;            prop->chartype == ucp_Ll ||
2732  int req_caseopt, reqvary, tempreqvary;            prop->chartype == ucp_Lt) == negated;
2733  int condcount = 0;  
2734  int options = *optionsptr;    case PT_GC:
2735  int after_manual_callout = 0;    return (pdata == _pcre_ucp_gentype[prop->chartype]) == negated;
2736  register int c;  
2737  register uschar *code = *codeptr;    case PT_PC:
2738  uschar *tempcode;    return (pdata == prop->chartype) == negated;
2739  BOOL inescq = FALSE;  
2740  BOOL groupsetfirstbyte = FALSE;    case PT_SC:
2741  const uschar *ptr = *ptrptr;    return (pdata == prop->script) == negated;
2742  const uschar *tempptr;  
2743  uschar *previous = NULL;    /* These are specials */
2744  uschar *previous_callout = NULL;  
2745  uschar classbits[32];    case PT_ALNUM:
2746      return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2747  #ifdef SUPPORT_UTF8            _pcre_ucp_gentype[prop->chartype] == ucp_N) == negated;
2748  BOOL class_utf8;  
2749  BOOL utf8 = (options & PCRE_UTF8) != 0;    case PT_SPACE:    /* Perl space */
2750  uschar *class_utf8data;    return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2751  uschar utf8_char[6];            c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2752  #else            == negated;
2753  BOOL utf8 = FALSE;  
2754  #endif    case PT_PXSPACE:  /* POSIX space */
2755      return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2756              c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2757              c == CHAR_FF || c == CHAR_CR)
2758              == negated;
2759    
2760      case PT_WORD:
2761      return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2762              _pcre_ucp_gentype[prop->chartype] == ucp_N ||
2763              c == CHAR_UNDERSCORE) == negated;
2764      }
2765    return FALSE;
2766    }
2767    #endif  /* SUPPORT_UCP */
2768    
 /* Set up the default and non-default settings for greediness */  
2769    
 greedy_default = ((options & PCRE_UNGREEDY) != 0);  
 greedy_non_default = greedy_default ^ 1;  
2770    
2771  /* Initialize no first byte, no required byte. REQ_UNSET means "no char  /*************************************************
2772  matching encountered yet". It gets changed to REQ_NONE if we hit something that  *     Check if auto-possessifying is possible    *
2773  matches a non-fixed char first char; reqbyte just remains unset if we never  *************************************************/
 find one.  
2774    
2775  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
2776  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
2777  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. */  
2778    
2779  firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;  Arguments:
2780      previous      pointer to the repeated opcode
2781      utf8          TRUE in UTF-8 mode
2782      ptr           next character in pattern
2783      options       options bits
2784      cd            contains pointers to tables etc.
2785    
2786  /* The variable req_caseopt contains either the REQ_CASELESS value or zero,  Returns:        TRUE if possessifying is wanted
2787  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. */  
2788    
2789  req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;  static BOOL
2790    check_auto_possessive(const uschar *previous, BOOL utf8, const uschar *ptr,
2791      int options, compile_data *cd)
2792    {
2793    int c, next;
2794    int op_code = *previous++;
2795    
2796  /* Switch on next character until the end of the branch */  /* Skip whitespace and comments in extended mode */
2797    
2798  for (;; ptr++)  if ((options & PCRE_EXTENDED) != 0)
2799    {    {
2800    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)  
2801      {      {
2802      if (c == '\\' && ptr[1] == 'E')      while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2803        if (*ptr == CHAR_NUMBER_SIGN)
2804        {        {
       inescq = FALSE;  
2805        ptr++;        ptr++;
2806        continue;        while (*ptr != 0)
       }  
     else  
       {  
       if (previous_callout != NULL)  
         {  
         complete_callout(previous_callout, ptr, cd);  
         previous_callout = NULL;  
         }  
       if ((options & PCRE_AUTO_CALLOUT) != 0)  
2807          {          {
2808          previous_callout = code;          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2809          code = auto_callout(code, ptr, cd);          ptr++;
2810    #ifdef SUPPORT_UTF8
2811            if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2812    #endif
2813          }          }
       goto NORMAL_CHAR;  
2814        }        }
2815        else break;
2816      }      }
2817      }
2818    
2819    /* 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
2820    a quantifier. */  value is a character, a negative value is an escape value. */
2821    
2822    if (*ptr == CHAR_BACKSLASH)
2823      {
2824      int temperrorcode = 0;
2825      next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2826      if (temperrorcode != 0) return FALSE;
2827      ptr++;    /* Point after the escape sequence */
2828      }
2829    
2830    is_quantifier = c == '*' || c == '+' || c == '?' ||  else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2831      (c == '{' && is_counted_repeat(ptr+1));    {
2832    #ifdef SUPPORT_UTF8
2833      if (utf8) { GETCHARINC(next, ptr); } else
2834    #endif
2835      next = *ptr++;
2836      }
2837    
2838    if (!is_quantifier && previous_callout != NULL &&  else return FALSE;
        after_manual_callout-- <= 0)  
     {  
     complete_callout(previous_callout, ptr, cd);  
     previous_callout = NULL;  
     }  
2839    
2840    /* In extended mode, skip white space and comments */  /* Skip whitespace and comments in extended mode */
2841    
2842    if ((options & PCRE_EXTENDED) != 0)  if ((options & PCRE_EXTENDED) != 0)
2843      {
2844      for (;;)
2845      {      {
2846      if ((cd->ctypes[c] & ctype_space) != 0) continue;      while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2847      if (c == '#')      if (*ptr == CHAR_NUMBER_SIGN)
2848        {        {
2849        /* The space before the ; is to avoid a warning on a silly compiler        ptr++;
2850        on the Macintosh. */        while (*ptr != 0)
2851        while ((c = *(++ptr)) != 0 && c != NEWLINE) ;          {
2852        if (c != 0) continue;   /* Else fall through to handle end of string */          if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2853            ptr++;
2854    #ifdef SUPPORT_UTF8
2855            if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2856    #endif
2857            }
2858        }        }
2859        else break;
2860      }      }
2861      }
2862    
2863    /* No auto callout for quantifiers. */  /* If the next thing is itself optional, we have to give up. */
2864    
2865    if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)  if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2866      {    strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2867      previous_callout = code;      return FALSE;
     code = auto_callout(code, ptr, cd);  
     }  
2868    
2869    switch(c)  /* Now compare the next item with the previous opcode. First, handle cases when
2870      {  the next item is a character. */
     /* The branch terminates at end of string, |, or ). */  
2871    
2872      case 0:  if (next >= 0) switch(op_code)
2873      case '|':    {
2874      case ')':    case OP_CHAR:
2875      *firstbyteptr = firstbyte;  #ifdef SUPPORT_UTF8
2876      *reqbyteptr = reqbyte;    GETCHARTEST(c, previous);
2877      *codeptr = code;  #else
2878      *ptrptr = ptr;    c = *previous;
2879      return TRUE;  #endif
2880      return c != next;
2881    
2882      /* Handle single-character metacharacters. In multiline mode, ^ disables    /* For CHARI (caseless character) we must check the other case. If we have
2883      the setting of any following char as a first character. */    Unicode property support, we can use it to test the other case of
2884      high-valued characters. */
2885    
2886      case '^':    case OP_CHARI:
2887      if ((options & PCRE_MULTILINE) != 0)  #ifdef SUPPORT_UTF8
2888        {    GETCHARTEST(c, previous);
2889        if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;  #else
2890        }    c = *previous;
2891      previous = NULL;  #endif
2892      *code++ = OP_CIRC;    if (c == next) return FALSE;
2893      break;  #ifdef SUPPORT_UTF8
2894      if (utf8)
2895        {
2896        unsigned int othercase;
2897        if (next < 128) othercase = cd->fcc[next]; else
2898    #ifdef SUPPORT_UCP
2899        othercase = UCD_OTHERCASE((unsigned int)next);
2900    #else
2901        othercase = NOTACHAR;
2902    #endif
2903        return (unsigned int)c != othercase;
2904        }
2905      else
2906    #endif  /* SUPPORT_UTF8 */
2907      return (c != cd->fcc[next]);  /* Non-UTF-8 mode */
2908    
2909      case '$':    /* For OP_NOT and OP_NOTI, the data is always a single-byte character. These
2910      previous = NULL;    opcodes are not used for multi-byte characters, because they are coded using
2911      *code++ = OP_DOLL;    an XCLASS instead. */
     break;  
2912    
2913      /* There can never be a first char if '.' is first, whatever happens about    case OP_NOT:
2914      repeats. The value of reqbyte doesn't change either. */    return (c = *previous) == next;
2915    
2916      case '.':    case OP_NOTI:
2917      if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;    if ((c = *previous) == next) return TRUE;
2918      zerofirstbyte = firstbyte;  #ifdef SUPPORT_UTF8
2919      zeroreqbyte = reqbyte;    if (utf8)
2920      previous = code;      {
2921      *code++ = OP_ANY;      unsigned int othercase;
2922      break;      if (next < 128) othercase = cd->fcc[next]; else
2923    #ifdef SUPPORT_UCP
2924        othercase = UCD_OTHERCASE(next);
2925    #else
2926        othercase = NOTACHAR;
2927    #endif
2928        return (unsigned int)c == othercase;
2929        }
2930      else
2931    #endif  /* SUPPORT_UTF8 */
2932      return (c == cd->fcc[next]);  /* Non-UTF-8 mode */
2933    
2934      /* 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.
2935      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.  
2936    
2937      If the class contains characters outside the 0-255 range, a different    case OP_DIGIT:
2938      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.  
     */  
2939    
2940      case '[':    case OP_NOT_DIGIT:
2941      previous = code;    return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2942    
2943      /* PCRE supports POSIX class stuff inside a class. Perl gives an error if    case OP_WHITESPACE:
2944      they are encountered at the top level, so we'll do that too. */    return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2945    
2946      if ((ptr[1] == ':' || ptr[1] == '.' || ptr[1] == '=') &&    case OP_NOT_WHITESPACE:
2947          check_posix_syntax(ptr, &tempptr, cd))    return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
       {  
       *errorcodeptr = (ptr[1] == ':')? ERR13 : ERR31;  
       goto FAILED;  
       }  
2948    
2949      /* If the first character is '^', set the negation flag and skip it. */    case OP_WORDCHAR:
2950      return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2951    
2952      if ((c = *(++ptr)) == '^')    case OP_NOT_WORDCHAR:
2953        {    return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
       negate_class = TRUE;  
       c = *(++ptr);  
       }  
     else  
       {  
       negate_class = FALSE;  
       }  
2954    
2955      /* Keep a count of chars with values < 256 so that we can optimize the case    case OP_HSPACE:
2956      of just a single character (as long as it's < 256). For higher valued UTF-8    case OP_NOT_HSPACE:
2957      characters, we don't yet do any optimization. */    switch(next)
2958        {
2959        case 0x09:
2960        case 0x20:
2961        case 0xa0:
2962        case 0x1680:
2963        case 0x180e:
2964        case 0x2000:
2965        case 0x2001:
2966        case 0x2002:
2967        case 0x2003:
2968        case 0x2004:
2969        case 0x2005:
2970        case 0x2006:
2971        case 0x2007:
2972        case 0x2008:
2973        case 0x2009:
2974        case 0x200A:
2975        case 0x202f:
2976        case 0x205f:
2977        case 0x3000:
2978        return op_code == OP_NOT_HSPACE;
2979        default:
2980        return op_code != OP_NOT_HSPACE;
2981        }
2982    
2983      class_charcount = 0;    case OP_ANYNL:
2984      class_lastchar = -1;    case OP_VSPACE:
2985      case OP_NOT_VSPACE:
2986      switch(next)
2987        {
2988        case 0x0a:
2989        case 0x0b:
2990        case 0x0c:
2991        case 0x0d:
2992        case 0x85:
2993        case 0x2028:
2994        case 0x2029:
2995        return op_code == OP_NOT_VSPACE;
2996        default:
2997        return op_code != OP_NOT_VSPACE;
2998        }
2999    
3000  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UCP
3001      class_utf8 = FALSE;                       /* No chars >= 256 */    case OP_PROP:
3002      class_utf8data = code + LINK_SIZE + 34;   /* For UTF-8 items */    return check_char_prop(next, previous[0], previous[1], FALSE);
3003    
3004      case OP_NOTPROP:
3005      return check_char_prop(next, previous[0], previous[1], TRUE);
3006  #endif  #endif
3007    
3008      /* Initialize the 32-char bit map to all zeros. We have to build the    default:
3009      map in a temporary bit of store, in case the class contains only 1    return FALSE;
3010      character (< 256), because in that case the compiled code doesn't use the    }
     bit map. */  
3011    
     memset(classbits, 0, 32 * sizeof(uschar));  
3012    
3013      /* 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
3014      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
3015      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
3016      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
3017      character. */  replaced by OP_PROP codes when PCRE_UCP is set. */
3018    
3019      do  switch(op_code)
3020        {    {
3021      case OP_CHAR:
3022      case OP_CHARI:
3023  #ifdef SUPPORT_UTF8  #ifdef SUPPORT_UTF8
3024        if (utf8 && c > 127)    GETCHARTEST(c, previous);
3025          {                           /* Braces are required because the */  #else
3026          GETCHARLEN(c, ptr, ptr);    /* macro generates multiple statements */    c = *previous;
         }  
3027  #endif  #endif
3028      switch(-next)
3029        {
3030        case ESC_d:
3031        return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
3032    
3033        /* Inside \Q...\E everything is literal except \E */      case ESC_D:
3034        return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
       if (inescq)  
         {  
         if (c == '\\' && ptr[1] == 'E')  
           {  
           inescq = FALSE;  
           ptr++;  
           continue;  
           }  
         else goto LONE_SINGLE_CHARACTER;  
         }  
   
       /* Handle POSIX class names. Perl allows a negation extension of the  
       form [:^name:]. A square bracket that doesn't match the syntax is  
       treated as a literal. We also recognize the POSIX constructions  
       [.ch.] and [=ch=] ("collating elements") and fault them, as Perl  
       5.6 and 5.8 do. */  
   
       if (c == '[' &&  
           (ptr[1] == ':' || ptr[1] == '.' || ptr[1] == '=') &&  
           check_posix_syntax(ptr, &tempptr, cd))  
         {  
         BOOL local_negate = FALSE;  
         int posix_class, i;  
         register const uschar *cbits = cd->cbits;  
   
         if (ptr[1] != ':')  
           {  
           *errorcodeptr = ERR31;  
           goto FAILED;  
           }  
   
         ptr += 2;  
         if (*ptr == '^')  
           {  
           local_negate = TRUE;  
           ptr++;  
           }  
3035    
3036          posix_class = check_posix_name(ptr, tempptr - ptr);      case ESC_s:
3037          if (posix_class < 0)      return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
           {  
           *errorcodeptr = ERR30;  
           goto FAILED;  
           }  
3038    
3039          /* If matching is caseless, upper and lower are converted to      case ESC_S:
3040          alpha. This relies on the fact that the class table starts with      return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
3041          alpha, lower, upper as the first 3 entries. */  
3042        case ESC_w:
3043        return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
3044    
3045        case ESC_W:
3046        return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
3047    
3048        case ESC_h:
3049        case ESC_H:
3050        switch(c)
3051          {
3052          case 0x09:
3053          case 0x20:
3054          case 0xa0:
3055          case 0x1680:
3056          case 0x180e:
3057          case 0x2000:
3058          case 0x2001:
3059          case 0x2002:
3060          case 0x2003:
3061          case 0x2004:
3062          case 0x2005:
3063          case 0x2006:
3064          case 0x2007:
3065          case 0x2008:
3066          case 0x2009:
3067          case 0x200A:
3068          case 0x202f:
3069          case 0x205f:
3070          case 0x3000:
3071          return -next != ESC_h;
3072          default:
3073          return -next == ESC_h;
3074          }
3075    
3076          if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)      case ESC_v:
3077            posix_class = 0;      case ESC_V:
3078        switch(c)
3079          {
3080          case 0x0a:
3081          case 0x0b:
3082          case 0x0c:
3083          case 0x0d:
3084          case 0x85:
3085          case 0x2028:
3086          case 0x2029:
3087          return -next != ESC_v;
3088          default:
3089          return -next == ESC_v;
3090          }
3091    
3092          /* Or into the map we are building up to 3 of the static class      /* When PCRE_UCP is set, these values get generated for \d etc. Find
3093          tables, or their negations. The [:blank:] class sets up the same      their substitutions and process them. The result will always be either
3094          chars as the [:space:] class (all white space). We remove the vertical      -ESC_p or -ESC_P. Then fall through to process those values. */
         white space chars afterwards. */  
3095    
3096          posix_class *= 3;  #ifdef SUPPORT_UCP
3097          for (i = 0; i < 3; i++)      case ESC_du:
3098            {      case ESC_DU:
3099            BOOL blankclass = strncmp((char *)ptr, "blank", 5) == 0;      case ESC_wu:
3100            int taboffset = posix_class_maps[posix_class + i];      case ESC_WU:
3101            if (taboffset < 0) break;      case ESC_su:
3102            if (local_negate)      case ESC_SU:
3103              {        {
3104              if (i == 0)        int temperrorcode = 0;
3105                for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+taboffset];        ptr = substitutes[-next - ESC_DU];
3106              else        next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
3107                for (c = 0; c < 32; c++) classbits[c] &= ~cbits[c+taboffset];        if (temperrorcode != 0) return FALSE;
3108              if (blankclass) classbits[1] |= 0x3c;        ptr++;    /* For compatibility */
3109              }        }
3110            else      /* Fall through */
             {  
             for (c = 0; c < 32; c++) classbits[c] |= cbits[c+taboffset];  
             if (blankclass) classbits[1] &= ~0x3c;  
             }  
           }  
3111    
3112          ptr = tempptr + 1;      case ESC_p:
3113          class_charcount = 10;  /* Set > 1; assumes more than 1 per class */      case ESC_P:
3114          continue;    /* End of POSIX syntax handling */        {
3115          }        int ptype, pdata, errorcodeptr;
3116          BOOL negated;
3117    
3118        /* Backslash may introduce a single character, or it may introduce one        ptr--;      /* Make ptr point at the p or P */
3119        of the specials, which just set a flag. Escaped items are checked for        ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
3120        validity in the pre-compiling pass. The sequence \b is a special case.        if (ptype < 0) return FALSE;
3121        Inside a class (and only there) it is treated as backspace. Elsewhere        ptr++;      /* Point past the final curly ket */
3122        it marks a word boundary. Other escapes have preset maps ready to  
3123        or into the one we are building. We assume they have more than one        /* If the property item is optional, we have to give up. (When generated
3124        character in them, so set class_charcount bigger than one. */        from \d etc by PCRE_UCP, this test will have been applied much earlier,
3125          to the original \d etc. At this point, ptr will point to a zero byte. */
3126    
3127          if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
3128            strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
3129              return FALSE;
3130    
3131        if (c == '\\')        /* Do the property check. */
         {  
         c = check_escape(&ptr, errorcodeptr, *brackets, options, TRUE);  
3132    
3133          if (-c == ESC_b) c = '\b';       /* \b is backslash in a class */        return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
3134          else if (-c == ESC_X) c = 'X';   /* \X is literal X in a class */        }
3135          else if (-c == ESC_Q)            /* Handle start of quoted string */  #endif
           {  
           if (ptr[1] == '\\' && ptr[2] == 'E')  
             {  
             ptr += 2; /* avoid empty string */  
             }  
           else inescq = TRUE;  
           continue;  
           }  
3136    
3137          if (c < 0)      default:
3138            {      return FALSE;
3139            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;  
3140    
3141              case ESC_D:    /* In principle, support for Unicode properties should be integrated here as
3142              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];    well. It means re-organizing the above code so as to get hold of the property
3143              continue;    values before switching on the op-code. However, I wonder how many patterns
3144      combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
3145      these op-codes are never generated.) */
3146    
3147              case ESC_w:    case OP_DIGIT:
3148              for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];    return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
3149              continue;           next == -ESC_h || next == -ESC_v || next == -ESC_R;
3150    
3151              case ESC_W:    case OP_NOT_DIGIT:
3152              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];    return next == -ESC_d;
             continue;  
3153    
3154              case ESC_s:    case OP_WHITESPACE:
3155              for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];    return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
             classbits[1] &= ~0x08;   /* Perl 5.004 onwards omits VT from \s */  
             continue;  
3156    
3157              case ESC_S:    case OP_NOT_WHITESPACE:
3158              for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];    return next == -ESC_s || next == -ESC_h || next == -ESC_v;
             classbits[1] |= 0x08;    /* Perl 5.004 onwards omits VT from \s */  
             continue;  
3159    
3160  #ifdef SUPPORT_UCP    case OP_HSPACE:
3161              case ESC_p:    return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
3162              case ESC_P:           next == -ESC_w || next == -ESC_v || next == -ESC_R;
               {  
               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  
3163    
3164              /* Unrecognized escapes are faulted if PCRE is running in its    case OP_NOT_HSPACE:
3165              strict mode. By default, for compatibility with Perl, they are    return next == -ESC_h;
             treated as literals. */  
3166    
3167              default:    /* Can't have \S in here because VT matches \S (Perl anomaly) */
3168              if ((options & PCRE_EXTRA) != 0)    case OP_ANYNL:
3169                {    case OP_VSPACE:
3170                *errorcodeptr = ERR7;    return next == -ESC_V || next == -ESC_d || next == -ESC_w;
               goto FAILED;  
               }  
             c = *ptr;              /* The final character */  
             class_charcount -= 2;  /* Undo the default count from above */  
             }  
           }  
3171    
3172          /* Fall through if we have a single character (c >= 0). This may be    case OP_NOT_VSPACE:
3173          > 256 in UTF-8 mode. */    return next == -ESC_v || next == -ESC_R;
3174    
3175          }   /* End of backslash handling */    case OP_WORDCHAR:
3176      return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
3177             next == -ESC_v || next == -ESC_R;
3178    
3179        /* A single character may be followed by '-' to form a range. However,    case OP_NOT_WORDCHAR:
3180        Perl does not permit ']' to be the end of the range. A '-' character    return next == -ESC_w || next == -ESC_d;
       here is treated as a literal. */  
3181    
3182        if (ptr[1] == '-' && ptr[2] != ']')    default:
3183          {    return FALSE;
3184          int d;    }
         ptr += 2;  
3185    
3186  #ifdef SUPPORT_UTF8  /* Control does not reach here */
3187          if (utf8)  }
           {                           /* Braces are required because the */  
           GETCHARLEN(d, ptr, ptr);    /* macro generates multiple statements */  
           }  
         else  
 #endif  
         d = *ptr;  /* Not UTF-8 mode */  
3188    
         /* The second part of a range can be a single-character escape, but  
         not any of the other escapes. Perl 5.6 treats a hyphen as a literal  
         in such circumstances. */  
3189    
         if (d == '\\')  
           {  
           const uschar *oldptr = ptr;  
           d = check_escape(&ptr, errorcodeptr, *brackets, options, TRUE);  
3190    
3191            /* \b is backslash; \X is literal X; any other special means the '-'  /*************************************************
3192            was literal */  *           Compile one branch                   *
3193    *************************************************/
3194    
3195            if (d < 0)  /* Scan the pattern, compiling it into the a vector. If the options are
3196              {  changed during the branch, the pointer is used to change the external options
3197              if (d == -ESC_b) d = '\b';  bits. This function is used during the pre-compile phase when we are trying
3198              else if (d == -ESC_X) d = 'X'; else  to find out the amount of memory needed, as well as during the real compile
3199                {  phase. The value of lengthptr distinguishes the two phases.
               ptr = oldptr - 2;  
               goto LONE_SINGLE_CHARACTER;  /* A few lines below */  
               }  
             }  
           }  
3200    
3201          /* The check that the two values are in the correct order happens in  Arguments:
3202          the pre-pass. Optimize one-character ranges */    optionsptr     pointer to the option bits
3203      codeptr        points to the pointer to the current code point
3204      ptrptr         points to the current pattern pointer
3205      errorcodeptr   points to error code variable
3206      firstbyteptr   set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3207      reqbyteptr     set to the last literal character required, else < 0
3208      bcptr          points to current branch chain
3209      cond_depth     conditional nesting depth
3210      cd             contains pointers to tables etc.
3211      lengthptr      NULL during the real compile phase
3212                     points to length accumulator during pre-compile phase
3213    
3214          if (d == c) goto LONE_SINGLE_CHARACTER;  /* A few lines below */  Returns:         TRUE on success
3215                     FALSE, with *errorcodeptr set non-zero on error
3216    */
3217    
3218          /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless  static BOOL
3219          matching, we have to use an XCLASS with extra data items. Caseless  compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
3220          matching for characters > 127 is available only if UCP support is    int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
3221          available. */    int cond_depth, compile_data *cd, int *lengthptr)
3222    {
3223    int repeat_type, op_type;
3224    int repeat_min = 0, repeat_max = 0;      /* To please picky compilers */
3225    int bravalue = 0;
3226    int greedy_default, greedy_non_default;
3227    int firstbyte, reqbyte;
3228    int zeroreqbyte, zerofirstbyte;
3229    int req_caseopt, reqvary, tempreqvary;
3230    int options = *optionsptr;               /* May change dynamically */
3231    int after_manual_callout = 0;
3232    int length_prevgroup = 0;
3233    register int c;
3234    register uschar *code = *codeptr;
3235    uschar *last_code = code;
3236    uschar *orig_code = code;
3237    uschar *tempcode;
3238    BOOL inescq = FALSE;
3239    BOOL groupsetfirstbyte = FALSE;
3240    const uschar *ptr = *ptrptr;
3241    const uschar *tempptr;
3242    const uschar *nestptr = NULL;
3243    uschar *previous = NULL;
3244    uschar *previous_callout = NULL;
3245    uschar *save_hwm = NULL;
3246    uschar classbits[32];
3247    
3248    /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3249    must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3250    dynamically as we process the pattern. */
3251    
3252    #ifdef SUPPORT_UTF8
3253    BOOL class_utf8;
3254    BOOL utf8 = (options & PCRE_UTF8) != 0;
3255    uschar *class_utf8data;
3256    uschar *class_utf8data_base;
3257    uschar utf8_char[6];
3258    #else
3259    BOOL utf8 = FALSE;
3260    #endif
3261    
3262    #ifdef PCRE_DEBUG
3263    if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3264    #endif
3265    
3266    /* Set up the default and non-default settings for greediness */
3267    
3268    greedy_default = ((options & PCRE_UNGREEDY) != 0);
3269    greedy_non_default = greedy_default ^ 1;
3270    
3271    /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3272    matching encountered yet". It gets changed to REQ_NONE if we hit something that
3273    matches a non-fixed char first char; reqbyte just remains unset if we never
3274    find one.
3275    
3276    When we hit a repeat whose minimum is zero, we may have to adjust these values
3277    to take the zero repeat into account. This is implemented by setting them to
3278    zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
3279    item types that can be repeated set these backoff variables appropriately. */
3280    
3281    firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
3282    
3283    /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
3284    according to the current setting of the caseless flag. REQ_CASELESS is a bit
3285    value > 255. It is added into the firstbyte or reqbyte variables to record the
3286    case status of the value. This is used only for ASCII characters. */
3287    
3288    req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
3289    
3290    /* Switch on next character until the end of the branch */
3291    
3292    for (;; ptr++)
3293      {
3294      BOOL negate_class;
3295      BOOL should_flip_negation;
3296      BOOL possessive_quantifier;
3297      BOOL is_quantifier;
3298      BOOL is_recurse;
3299      BOOL reset_bracount;
3300      int class_charcount;
3301      int class_lastchar;
3302      int newoptions;
3303      int recno;
3304      int refsign;
3305      int skipbytes;
3306      int subreqbyte;
3307      int subfirstbyte;
3308      int terminator;
3309      int mclength;
3310      int tempbracount;
3311      uschar mcbuffer[8];
3312    
3313      /* Get next byte in the pattern */
3314    
3315      c = *ptr;
3316    
3317      /* If we are at the end of a nested substitution, revert to the outer level
3318      string. Nesting only happens one level deep. */
3319    
3320      if (c == 0 && nestptr != NULL)
3321        {
3322        ptr = nestptr;
3323        nestptr = NULL;
3324        c = *ptr;
3325        }
3326    
3327      /* If we are in the pre-compile phase, accumulate the length used for the
3328      previous cycle of this loop. */
3329    
3330      if (lengthptr != NULL)
3331        {
3332    #ifdef PCRE_DEBUG
3333        if (code > cd->hwm) cd->hwm = code;                 /* High water info */
3334    #endif
3335        if (code > cd->start_workspace + WORK_SIZE_CHECK)   /* Check for overrun */
3336          {
3337          *errorcodeptr = ERR52;
3338          goto FAILED;
3339          }
3340    
3341        /* There is at least one situation where code goes backwards: this is the
3342        case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3343        the class is simply eliminated. However, it is created first, so we have to
3344        allow memory for it. Therefore, don't ever reduce the length at this point.
3345        */
3346    
3347        if (code < last_code) code = last_code;
3348    
3349        /* Paranoid check for integer overflow */
3350    
3351        if (OFLOW_MAX - *lengthptr < code - last_code)
3352          {
3353          *errorcodeptr = ERR20;
3354          goto FAILED;
3355          }
3356    
3357        *lengthptr += (int)(code - last_code);
3358        DPRINTF(("length=%d added %d c=%c\n", *lengthptr, (int)(code - last_code),
3359          c));
3360    
3361        /* If "previous" is set and it is not at the start of the work space, move
3362        it back to there, in order to avoid filling up the work space. Otherwise,
3363        if "previous" is NULL, reset the current code pointer to the start. */
3364    
3365        if (previous != NULL)
3366          {
3367          if (previous > orig_code)
3368            {
3369            memmove(orig_code, previous, code - previous);
3370            code -= previous - orig_code;
3371            previous = orig_code;
3372            }
3373          }
3374        else code = orig_code;
3375    
3376        /* Remember where this code item starts so we can pick up the length
3377        next time round. */
3378    
3379        last_code = code;
3380        }
3381    
3382      /* In the real compile phase, just check the workspace used by the forward
3383      reference list. */
3384    
3385      else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
3386        {
3387        *errorcodeptr = ERR52;
3388        goto FAILED;
3389        }
3390    
3391      /* If in \Q...\E, check for the end; if not, we have a literal */
3392    
3393      if (inescq && c != 0)
3394        {
3395        if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3396          {
3397          inescq = FALSE;
3398          ptr++;
3399          continue;
3400          }
3401        else
3402          {
3403          if (previous_callout != NULL)
3404            {
3405            if (lengthptr == NULL)  /* Don't attempt in pre-compile phase */
3406              complete_callout(previous_callout, ptr, cd);
3407            previous_callout = NULL;
3408            }
3409          if ((options & PCRE_AUTO_CALLOUT) != 0)
3410            {
3411            previous_callout = code;
3412            code = auto_callout(code, ptr, cd);
3413            }
3414          goto NORMAL_CHAR;
3415          }
3416        }
3417    
3418      /* Fill in length of a previous callout, except when the next thing is
3419      a quantifier. */
3420    
3421      is_quantifier =
3422        c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3423        (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3424    
3425      if (!is_quantifier && previous_callout != NULL &&
3426           after_manual_callout-- <= 0)
3427        {
3428        if (lengthptr == NULL)      /* Don't attempt in pre-compile phase */
3429          complete_callout(previous_callout, ptr, cd);
3430        previous_callout = NULL;
3431        }
3432    
3433      /* In extended mode, skip white space and comments. */
3434    
3435      if ((options & PCRE_EXTENDED) != 0)
3436        {
3437        if ((cd->ctypes[c] & ctype_space) != 0) continue;
3438        if (c == CHAR_NUMBER_SIGN)
3439          {
3440          ptr++;
3441          while (*ptr != 0)
3442            {
3443            if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3444            ptr++;
3445    #ifdef SUPPORT_UTF8
3446            if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
3447    #endif
3448            }
3449          if (*ptr != 0) continue;
3450    
3451          /* Else fall through to handle end of string */
3452          c = 0;
3453          }
3454        }
3455    
3456      /* No auto callout for quantifiers. */
3457    
3458      if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3459        {
3460        previous_callout = code;
3461        code = auto_callout(code, ptr, cd);
3462        }
3463    
3464      switch(c)
3465        {
3466        /* ===================================================================*/
3467        case 0:                        /* The branch terminates at string end */
3468        case CHAR_VERTICAL_LINE:       /* or | or ) */
3469        case CHAR_RIGHT_PARENTHESIS:
3470        *firstbyteptr = firstbyte;
3471        *reqbyteptr = reqbyte;
3472        *codeptr = code;
3473        *ptrptr = ptr;
3474        if (lengthptr != NULL)
3475          {
3476          if (OFLOW_MAX - *lengthptr < code - last_code)
3477            {
3478            *errorcodeptr = ERR20;
3479            goto FAILED;
3480            }
3481          *lengthptr += (int)(code - last_code);   /* To include callout length */
3482          DPRINTF((">> end branch\n"));
3483          }
3484        return TRUE;
3485    
3486    
3487        /* ===================================================================*/
3488        /* Handle single-character metacharacters. In multiline mode, ^ disables
3489        the setting of any following char as a first character. */
3490    
3491        case CHAR_CIRCUMFLEX_ACCENT:
3492        previous = NULL;
3493        if ((options & PCRE_MULTILINE) != 0)
3494          {
3495          if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3496          *code++ = OP_CIRCM;
3497          }
3498        else *code++ = OP_CIRC;
3499        break;
3500    
3501        case CHAR_DOLLAR_SIGN:
3502        previous = NULL;
3503        *code++ = ((option