+/* Store a multibyte character in three contiguous bytes starting
+ DESTINATION, and increment DESTINATION to the byte after where the
+ character is stored. Therefore, DESTINATION must be an lvalue. */
+
+#define STORE_CHARACTER_AND_INCR(destination, character) \
+ do { \
+ (destination)[0] = (character) & 0377; \
+ (destination)[1] = ((character) >> 8) & 0377; \
+ (destination)[2] = (character) >> 16; \
+ (destination) += 3; \
+ } while (0)
+
+/* Put into DESTINATION a character stored in three contiguous bytes
+ starting at SOURCE. */
+
+#define EXTRACT_CHARACTER(destination, source) \
+ do { \
+ (destination) = ((source)[0] \
+ | ((source)[1] << 8) \
+ | ((source)[2] << 16)); \
+ } while (0)
+
+
+/* Macros for charset. */
+
+/* Size of bitmap of charset P in bytes. P is a start of charset,
+ i.e. *P is (re_opcode_t) charset or (re_opcode_t) charset_not. */
+#define CHARSET_BITMAP_SIZE(p) ((p)[1] & 0x7F)
+
+/* Nonzero if charset P has range table. */
+#define CHARSET_RANGE_TABLE_EXISTS_P(p) ((p)[1] & 0x80)
+
+/* Return the address of range table of charset P. But not the start
+ of table itself, but the before where the number of ranges is
+ stored. `2 +' means to skip re_opcode_t and size of bitmap,
+ and the 2 bytes of flags at the start of the range table. */
+#define CHARSET_RANGE_TABLE(p) (&(p)[4 + CHARSET_BITMAP_SIZE (p)])
+
+/* Extract the bit flags that start a range table. */
+#define CHARSET_RANGE_TABLE_BITS(p) \
+ ((p)[2 + CHARSET_BITMAP_SIZE (p)] \
+ + (p)[3 + CHARSET_BITMAP_SIZE (p)] * 0x100)
+
+/* Test if C is listed in the bitmap of charset P. */
+#define CHARSET_LOOKUP_BITMAP(p, c) \
+ ((c) < CHARSET_BITMAP_SIZE (p) * BYTEWIDTH \
+ && (p)[2 + (c) / BYTEWIDTH] & (1 << ((c) % BYTEWIDTH)))
+
+/* Return the address of end of RANGE_TABLE. COUNT is number of
+ ranges (which is a pair of (start, end)) in the RANGE_TABLE. `* 2'
+ is start of range and end of range. `* 3' is size of each start
+ and end. */
+#define CHARSET_RANGE_TABLE_END(range_table, count) \
+ ((range_table) + (count) * 2 * 3)
+
+/* Test if C is in RANGE_TABLE. A flag NOT is negated if C is in.
+ COUNT is number of ranges in RANGE_TABLE. */
+#define CHARSET_LOOKUP_RANGE_TABLE_RAW(not, c, range_table, count) \
+ do \
+ { \
+ int range_start, range_end; \
+ unsigned char *p; \
+ unsigned char *range_table_end \
+ = CHARSET_RANGE_TABLE_END ((range_table), (count)); \
+ \
+ for (p = (range_table); p < range_table_end; p += 2 * 3) \
+ { \
+ EXTRACT_CHARACTER (range_start, p); \
+ EXTRACT_CHARACTER (range_end, p + 3); \
+ \
+ if (range_start <= (c) && (c) <= range_end) \
+ { \
+ (not) = !(not); \
+ break; \
+ } \
+ } \
+ } \
+ while (0)
+
+/* Test if C is in range table of CHARSET. The flag NOT is negated if
+ C is listed in it. */
+#define CHARSET_LOOKUP_RANGE_TABLE(not, c, charset) \
+ do \
+ { \
+ /* Number of ranges in range table. */ \
+ int count; \
+ unsigned char *range_table = CHARSET_RANGE_TABLE (charset); \
+ \
+ EXTRACT_NUMBER_AND_INCR (count, range_table); \
+ CHARSET_LOOKUP_RANGE_TABLE_RAW ((not), (c), range_table, count); \
+ } \
+ while (0)
+\f
- { "Success", /* REG_NOERROR */
- "No match", /* REG_NOMATCH */
- "Invalid regular expression", /* REG_BADPAT */
- "Invalid collation character", /* REG_ECOLLATE */
- "Invalid character class name", /* REG_ECTYPE */
- "Trailing backslash", /* REG_EESCAPE */
- "Invalid back reference", /* REG_ESUBREG */
- "Unmatched [ or [^", /* REG_EBRACK */
- "Unmatched ( or \\(", /* REG_EPAREN */
- "Unmatched \\{", /* REG_EBRACE */
- "Invalid content of \\{\\}", /* REG_BADBR */
- "Invalid range end", /* REG_ERANGE */
- "Memory exhausted", /* REG_ESPACE */
- "Invalid preceding regular expression", /* REG_BADRPT */
- "Premature end of regular expression", /* REG_EEND */
- "Regular expression too big", /* REG_ESIZE */
- "Unmatched ) or \\)", /* REG_ERPAREN */
+ {
+ gettext_noop ("Success"), /* REG_NOERROR */
+ gettext_noop ("No match"), /* REG_NOMATCH */
+ gettext_noop ("Invalid regular expression"), /* REG_BADPAT */
+ gettext_noop ("Invalid collation character"), /* REG_ECOLLATE */
+ gettext_noop ("Invalid character class name"), /* REG_ECTYPE */
+ gettext_noop ("Trailing backslash"), /* REG_EESCAPE */
+ gettext_noop ("Invalid back reference"), /* REG_ESUBREG */
+ gettext_noop ("Unmatched [ or [^"), /* REG_EBRACK */
+ gettext_noop ("Unmatched ( or \\("), /* REG_EPAREN */
+ gettext_noop ("Unmatched \\{"), /* REG_EBRACE */
+ gettext_noop ("Invalid content of \\{\\}"), /* REG_BADBR */
+ gettext_noop ("Invalid range end"), /* REG_ERANGE */
+ gettext_noop ("Memory exhausted"), /* REG_ESPACE */
+ gettext_noop ("Invalid preceding regular expression"), /* REG_BADRPT */
+ gettext_noop ("Premature end of regular expression"), /* REG_EEND */
+ gettext_noop ("Regular expression too big"), /* REG_ESIZE */
+ gettext_noop ("Unmatched ) or \\)"), /* REG_ERPAREN */
- PATFETCH (c);
-
- if (c == '*'
- || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
- ;
-
- else if (syntax & RE_BK_PLUS_QM && c == '\\')
- {
- if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
-
- PATFETCH (c1);
- if (!(c1 == '+' || c1 == '?'))
- {
- PATUNFETCH;
- PATUNFETCH;
- break;
- }
-
- c = c1;
- }
- else
- {
- PATUNFETCH;
- break;
- }
-
- /* If we get here, we found another repeat character. */
- }
-
- /* Star, etc. applied to an empty pattern is equivalent
- to an empty pattern. */
- if (!laststart)
- break;
-
- /* Now we know whether or not zero matches is allowed
- and also whether or not two or more matches is allowed. */
- if (many_times_ok)
- { /* More than one repetition is allowed, so put in at the
- end a backward relative jump from `b' to before the next
- jump we're going to put in below (which jumps from
- laststart to after this jump).
-
- But if we are at the `*' in the exact sequence `.*\n',
- insert an unconditional jump backwards to the .,
- instead of the beginning of the loop. This way we only
- push a failure point once, instead of every time
- through the loop. */
- assert (p - 1 > pattern);
-
- /* Allocate the space for the jump. */
- GET_BUFFER_SPACE (3);
-
- /* We know we are not at the first character of the pattern,
- because laststart was nonzero. And we've already
- incremented `p', by the way, to be the character after
- the `*'. Do we have to do something analogous here
- for null bytes, because of RE_DOT_NOT_NULL? */
- if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
+ if (p == pend)
+ break;
+
+ PATFETCH (c);
+
+ if (c == '*'
+ || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
+ ;
+
+ else if (syntax & RE_BK_PLUS_QM && c == '\\')
+ {
+ if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
+
+ PATFETCH (c1);
+ if (!(c1 == '+' || c1 == '?'))
+ {
+ PATUNFETCH;
+ PATUNFETCH;
+ break;
+ }
+
+ c = c1;
+ }
+ else
+ {
+ PATUNFETCH;
+ break;
+ }
+
+ /* If we get here, we found another repeat character. */
+ }
+
+ /* Star, etc. applied to an empty pattern is equivalent
+ to an empty pattern. */
+ if (!laststart)
+ break;
+
+ /* Now we know whether or not zero matches is allowed
+ and also whether or not two or more matches is allowed. */
+ if (many_times_ok)
+ { /* More than one repetition is allowed, so put in at the
+ end a backward relative jump from `b' to before the next
+ jump we're going to put in below (which jumps from
+ laststart to after this jump).
+
+ But if we are at the `*' in the exact sequence `.*\n',
+ insert an unconditional jump backwards to the .,
+ instead of the beginning of the loop. This way we only
+ push a failure point once, instead of every time
+ through the loop. */
+ assert (p - 1 > pattern);
+
+ /* Allocate the space for the jump. */
+ GET_BUFFER_SPACE (3);
+
+ /* We know we are not at the first character of the pattern,
+ because laststart was nonzero. And we've already
+ incremented `p', by the way, to be the character after
+ the `*'. Do we have to do something analogous here
+ for null bytes, because of RE_DOT_NOT_NULL? */
+ if (TRANSLATE ((unsigned char)*(p - 2)) == TRANSLATE ('.')
- /* Look ahead to see if it's a range when the last thing
- was a character class. */
- if (had_char_class && c == '-' && *p != ']')
- FREE_STACK_RETURN (REG_ERANGE);
-
- /* Look ahead to see if it's a range when the last thing
- was a character: if this is a hyphen not at the
- beginning or the end of a list, then it's the range
- operator. */
- if (c == '-'
- && !(p - 2 >= pattern && p[-2] == '[')
- && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
- && *p != ']')
- {
- reg_errcode_t ret
- = compile_range (&p, pend, translate, syntax, b);
- if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
- }
-
- else if (p[0] == '-' && p[1] != ']')
- { /* This handles ranges made up of characters only. */
- reg_errcode_t ret;
-
- /* Move past the `-'. */
- PATFETCH (c1);
-
- ret = compile_range (&p, pend, translate, syntax, b);
- if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
- }
-
- /* See if we're at the beginning of a possible character
- class. */
-
- else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
- { /* Leave room for the null. */
- char str[CHAR_CLASS_MAX_LENGTH + 1];
-
- PATFETCH (c);
- c1 = 0;
-
- /* If pattern is `[[:'. */
- if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
-
- for (;;)
- {
- PATFETCH (c);
- if (c == ':' || c == ']' || p == pend
- || c1 == CHAR_CLASS_MAX_LENGTH)
- break;
- str[c1++] = c;
- }
- str[c1] = '\0';
-
- /* If isn't a word bracketed by `[:' and:`]':
- undo the ending character, the letters, and leave
- the leading `:' and `[' (but set bits for them). */
- if (c == ':' && *p == ']')
- {
- int ch;
- boolean is_alnum = STREQ (str, "alnum");
- boolean is_alpha = STREQ (str, "alpha");
- boolean is_blank = STREQ (str, "blank");
- boolean is_cntrl = STREQ (str, "cntrl");
- boolean is_digit = STREQ (str, "digit");
- boolean is_graph = STREQ (str, "graph");
- boolean is_lower = STREQ (str, "lower");
- boolean is_print = STREQ (str, "print");
- boolean is_punct = STREQ (str, "punct");
- boolean is_space = STREQ (str, "space");
- boolean is_upper = STREQ (str, "upper");
- boolean is_xdigit = STREQ (str, "xdigit");
-
- if (!IS_CHAR_CLASS (str))
- FREE_STACK_RETURN (REG_ECTYPE);
+ /* If C indicates start of multibyte char, get the
+ actual character code in C, and set the pattern
+ pointer P to the next character boundary. */
+ if (bufp->multibyte && BASE_LEADING_CODE_P (c))
+ {
+ PATUNFETCH;
+ c = STRING_CHAR_AND_LENGTH (p, pend - p, len);
+ p += len;
+ }
+ /* What should we do for the character which is
+ greater than 0x7F, but not BASE_LEADING_CODE_P?
+ XXX */
- || (is_space && ISSPACE (ch))
- || (is_upper && ISUPPER (ch))
- || (is_xdigit && ISXDIGIT (ch)))
- SET_LIST_BIT (ch);
- }
- had_char_class = true;
- }
- else
- {
- c1++;
- while (c1--)
- PATUNFETCH;
- SET_LIST_BIT ('[');
- SET_LIST_BIT (':');
- had_char_class = false;
- }
- }
- else
- {
- had_char_class = false;
- SET_LIST_BIT (c);
- }
- }
-
- /* Discard any (non)matching list bytes that are all 0 at the
- end of the map. Decrease the map-length byte too. */
- while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
- b[-1]--;
- b += b[-1];
- }
- break;
+ || (is_space && ISSPACE (ch))
+ || (is_upper && ISUPPER (ch))
+ || (is_xdigit && ISXDIGIT (ch)))
+ SET_LIST_BIT (translated);
+ if ( (is_ascii && IS_REAL_ASCII (ch))
+ || (is_nonascii && !IS_REAL_ASCII (ch))
+ || (is_unibyte && ISUNIBYTE (ch))
+ || (is_multibyte && !ISUNIBYTE (ch)))
+ SET_LIST_BIT (translated);
+
+ if ( (is_word && ISWORD (ch)))
+ SET_LIST_BIT (translated);
+ }
+
+ /* Repeat the loop. */
+ continue;
+ }
+ else
+ {
+ c1++;
+ while (c1--)
+ PATUNFETCH;
+ SET_LIST_BIT ('[');
+
+ /* Because the `:' may starts the range, we
+ can't simply set bit and repeat the loop.
+ Instead, just set it to C and handle below. */
+ c = ':';
+ }
+ }
+
+ if (p < pend && p[0] == '-' && p[1] != ']')
+ {
+
+ /* Discard the `-'. */
+ PATFETCH (c1);
+
+ /* Fetch the character which ends the range. */
+ PATFETCH (c1);
+ if (bufp->multibyte && BASE_LEADING_CODE_P (c1))
+ {
+ PATUNFETCH;
+ c1 = STRING_CHAR_AND_LENGTH (p, pend - p, len);
+ p += len;
+ }
+
+ if (SINGLE_BYTE_CHAR_P (c)
+ && ! SINGLE_BYTE_CHAR_P (c1))
+ {
+ /* Handle a range such as \177-\377 in multibyte mode.
+ Split that into two ranges,,
+ the low one ending at 0237, and the high one
+ starting at ...040. */
+ int c1_base = (c1 & ~0177) | 040;
+ SET_RANGE_TABLE_WORK_AREA (range_table_work, c, c1);
+ c1 = 0237;
+ }
+ else if (!SAME_CHARSET_P (c, c1))
+ FREE_STACK_RETURN (REG_ERANGE);
+ }
+ else
+ /* Range from C to C. */
+ c1 = c;
+
+ /* Set the range ... */
+ if (SINGLE_BYTE_CHAR_P (c))
+ /* ... into bitmap. */
+ {
+ unsigned this_char;
+ int range_start = c, range_end = c1;
+
+ /* If the start is after the end, the range is empty. */
+ if (range_start > range_end)
+ {
+ if (syntax & RE_NO_EMPTY_RANGES)
+ FREE_STACK_RETURN (REG_ERANGE);
+ /* Else, repeat the loop. */
+ }
+ else
+ {
+ for (this_char = range_start; this_char <= range_end;
+ this_char++)
+ SET_LIST_BIT (TRANSLATE (this_char));
+ }
+ }
+ else
+ /* ... into range table. */
+ SET_RANGE_TABLE_WORK_AREA (range_table_work, c, c1);
+ }
+
+ /* Discard any (non)matching list bytes that are all 0 at the
+ end of the map. Decrease the map-length byte too. */
+ while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
+ b[-1]--;
+ b += b[-1];
+
+ /* Build real range table from work area. */
+ if (RANGE_TABLE_WORK_USED (range_table_work)
+ || RANGE_TABLE_WORK_BITS (range_table_work))
+ {
+ int i;
+ int used = RANGE_TABLE_WORK_USED (range_table_work);
+
+ /* Allocate space for COUNT + RANGE_TABLE. Needs two
+ bytes for flags, two for COUNT, and three bytes for
+ each character. */
+ GET_BUFFER_SPACE (4 + used * 3);
+
+ /* Indicate the existence of range table. */
+ laststart[1] |= 0x80;
+
+ /* Store the character class flag bits into the range table.
+ If not in emacs, these flag bits are always 0. */
+ *b++ = RANGE_TABLE_WORK_BITS (range_table_work) & 0xff;
+ *b++ = RANGE_TABLE_WORK_BITS (range_table_work) >> 8;
+
+ STORE_NUMBER_AND_INCR (b, used / 2);
+ for (i = 0; i < used; i++)
+ STORE_CHARACTER_AND_INCR
+ (b, RANGE_TABLE_WORK_ELT (range_table_work, i));
+ }
+ }
+ break;
- /* We're at the end of the group, so now we know how many
- groups were inside this one. */
- if (this_group_regnum <= MAX_REGNUM)
- {
- unsigned char *inner_group_loc
- = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset;
-
- *inner_group_loc = regnum - this_group_regnum;
- BUF_PUSH_3 (stop_memory, this_group_regnum,
- regnum - this_group_regnum);
- }
- }
- break;
-
-
- case '|': /* `\|'. */
- if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
- goto normal_backslash;
- handle_alt:
- if (syntax & RE_LIMITED_OPS)
- goto normal_char;
-
- /* Insert before the previous alternative a jump which
- jumps to this alternative if the former fails. */
- GET_BUFFER_SPACE (3);
- INSERT_JUMP (on_failure_jump, begalt, b + 6);
- pending_exact = 0;
- b += 3;
-
- /* The alternative before this one has a jump after it
- which gets executed if it gets matched. Adjust that
- jump so it will jump to this alternative's analogous
- jump (put in below, which in turn will jump to the next
- (if any) alternative's such jump, etc.). The last such
- jump jumps to the correct final destination. A picture:
- _____ _____
- | | | |
- | v | v
- a | b | c
-
- If we are at `b', then fixup_alt_jump right now points to a
- three-byte space after `a'. We'll put in the jump, set
- fixup_alt_jump to right after `b', and leave behind three
- bytes which we'll fill in when we get to after `c'. */
-
- if (fixup_alt_jump)
- STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
-
- /* Mark and leave space for a jump after this alternative,
- to be filled in later either by next alternative or
- when know we're at the end of a series of alternatives. */
- fixup_alt_jump = b;
- GET_BUFFER_SPACE (3);
- b += 3;
-
- laststart = 0;
- begalt = b;
- break;
-
-
- case '{':
- /* If \{ is a literal. */
- if (!(syntax & RE_INTERVALS)
- /* If we're at `\{' and it's not the open-interval
- operator. */
- || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
- || (p - 2 == pattern && p == pend))
- goto normal_backslash;
-
- handle_interval:
- {
- /* If got here, then the syntax allows intervals. */
-
- /* At least (most) this many matches must be made. */
- int lower_bound = -1, upper_bound = -1;
-
- beg_interval = p - 1;
-
- if (p == pend)
- {
- if (syntax & RE_NO_BK_BRACES)
- goto unfetch_interval;
- else
- FREE_STACK_RETURN (REG_EBRACE);
- }
-
- GET_UNSIGNED_NUMBER (lower_bound);
-
- if (c == ',')
- {
- GET_UNSIGNED_NUMBER (upper_bound);
- if (upper_bound < 0) upper_bound = RE_DUP_MAX;
- }
- else
- /* Interval such as `{1}' => match exactly once. */
- upper_bound = lower_bound;
-
- if (lower_bound < 0 || upper_bound > RE_DUP_MAX
- || lower_bound > upper_bound)
- {
- if (syntax & RE_NO_BK_BRACES)
- goto unfetch_interval;
- else
- FREE_STACK_RETURN (REG_BADBR);
- }
-
- if (!(syntax & RE_NO_BK_BRACES))
- {
- if (c != '\\') FREE_STACK_RETURN (REG_EBRACE);
-
- PATFETCH (c);
- }
-
- if (c != '}')
- {
- if (syntax & RE_NO_BK_BRACES)
- goto unfetch_interval;
- else
- FREE_STACK_RETURN (REG_BADBR);
- }
-
- /* We just parsed a valid interval. */
-
- /* If it's invalid to have no preceding re. */
- if (!laststart)
- {
- if (syntax & RE_CONTEXT_INVALID_OPS)
- FREE_STACK_RETURN (REG_BADRPT);
- else if (syntax & RE_CONTEXT_INDEP_OPS)
- laststart = b;
- else
- goto unfetch_interval;
- }
-
- /* If the upper bound is zero, don't want to succeed at
- all; jump from `laststart' to `b + 3', which will be
- the end of the buffer after we insert the jump. */
- if (upper_bound == 0)
- {
- GET_BUFFER_SPACE (3);
- INSERT_JUMP (jump, laststart, b + 3);
- b += 3;
- }
-
- /* Otherwise, we have a nontrivial interval. When
- we're all done, the pattern will look like:
- set_number_at <jump count> <upper bound>
- set_number_at <succeed_n count> <lower bound>
- succeed_n <after jump addr> <succeed_n count>
- <body of loop>
- jump_n <succeed_n addr> <jump count>
- (The upper bound and `jump_n' are omitted if
- `upper_bound' is 1, though.) */
- else
- { /* If the upper bound is > 1, we need to insert
- more at the end of the loop. */
- unsigned nbytes = 10 + (upper_bound > 1) * 10;
-
- GET_BUFFER_SPACE (nbytes);
-
- /* Initialize lower bound of the `succeed_n', even
- though it will be set during matching by its
- attendant `set_number_at' (inserted next),
- because `re_compile_fastmap' needs to know.
- Jump to the `jump_n' we might insert below. */
- INSERT_JUMP2 (succeed_n, laststart,
- b + 5 + (upper_bound > 1) * 5,
- lower_bound);
- b += 5;
-
- /* Code to initialize the lower bound. Insert
- before the `succeed_n'. The `5' is the last two
- bytes of this `set_number_at', plus 3 bytes of
- the following `succeed_n'. */
- insert_op2 (set_number_at, laststart, 5, lower_bound, b);
- b += 5;
-
- if (upper_bound > 1)
- { /* More than one repetition is allowed, so
- append a backward jump to the `succeed_n'
- that starts this interval.
-
- When we've reached this during matching,
- we'll have matched the interval once, so
- jump back only `upper_bound - 1' times. */
- STORE_JUMP2 (jump_n, b, laststart + 5,
- upper_bound - 1);
- b += 5;
-
- /* The location we want to set is the second
- parameter of the `jump_n'; that is `b-2' as
- an absolute address. `laststart' will be
- the `set_number_at' we're about to insert;
- `laststart+3' the number to set, the source
- for the relative address. But we are
- inserting into the middle of the pattern --
- so everything is getting moved up by 5.
- Conclusion: (b - 2) - (laststart + 3) + 5,
- i.e., b - laststart.
-
- We insert this at the beginning of the loop
- so that if we fail during matching, we'll
- reinitialize the bounds. */
- insert_op2 (set_number_at, laststart, b - laststart,
- upper_bound - 1, b);
- b += 5;
- }
- }
- pending_exact = 0;
- beg_interval = NULL;
- }
- break;
-
- unfetch_interval:
- /* If an invalid interval, match the characters as literals. */
- assert (beg_interval);
- p = beg_interval;
- beg_interval = NULL;
-
- /* normal_char and normal_backslash need `c'. */
- PATFETCH (c);
-
- if (!(syntax & RE_NO_BK_BRACES))
- {
- if (p > pattern && p[-1] == '\\')
- goto normal_backslash;
- }
- goto normal_char;
+ /* We're at the end of the group, so now we know how many
+ groups were inside this one. */
+ if (this_group_regnum <= MAX_REGNUM)
+ {
+ unsigned char *inner_group_loc
+ = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset;
+
+ *inner_group_loc = regnum - this_group_regnum;
+ BUF_PUSH_3 (stop_memory, this_group_regnum,
+ regnum - this_group_regnum);
+ }
+ }
+ break;
+
+
+ case '|': /* `\|'. */
+ if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
+ goto normal_backslash;
+ handle_alt:
+ if (syntax & RE_LIMITED_OPS)
+ goto normal_char;
+
+ /* Insert before the previous alternative a jump which
+ jumps to this alternative if the former fails. */
+ GET_BUFFER_SPACE (3);
+ INSERT_JUMP (on_failure_jump, begalt, b + 6);
+ pending_exact = 0;
+ b += 3;
+
+ /* The alternative before this one has a jump after it
+ which gets executed if it gets matched. Adjust that
+ jump so it will jump to this alternative's analogous
+ jump (put in below, which in turn will jump to the next
+ (if any) alternative's such jump, etc.). The last such
+ jump jumps to the correct final destination. A picture:
+ _____ _____
+ | | | |
+ | v | v
+ a | b | c
+
+ If we are at `b', then fixup_alt_jump right now points to a
+ three-byte space after `a'. We'll put in the jump, set
+ fixup_alt_jump to right after `b', and leave behind three
+ bytes which we'll fill in when we get to after `c'. */
+
+ if (fixup_alt_jump)
+ STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
+
+ /* Mark and leave space for a jump after this alternative,
+ to be filled in later either by next alternative or
+ when know we're at the end of a series of alternatives. */
+ fixup_alt_jump = b;
+ GET_BUFFER_SPACE (3);
+ b += 3;
+
+ laststart = 0;
+ begalt = b;
+ break;
+
+
+ case '{':
+ /* If \{ is a literal. */
+ if (!(syntax & RE_INTERVALS)
+ /* If we're at `\{' and it's not the open-interval
+ operator. */
+ || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
+ || (p - 2 == pattern && p == pend))
+ goto normal_backslash;
+
+ handle_interval:
+ {
+ /* If got here, then the syntax allows intervals. */
+
+ /* At least (most) this many matches must be made. */
+ int lower_bound = -1, upper_bound = -1;
+
+ beg_interval = p - 1;
+
+ if (p == pend)
+ {
+ if (syntax & RE_NO_BK_BRACES)
+ goto unfetch_interval;
+ else
+ FREE_STACK_RETURN (REG_EBRACE);
+ }
+
+ GET_UNSIGNED_NUMBER (lower_bound);
+
+ if (c == ',')
+ {
+ GET_UNSIGNED_NUMBER (upper_bound);
+ if (upper_bound < 0) upper_bound = RE_DUP_MAX;
+ }
+ else
+ /* Interval such as `{1}' => match exactly once. */
+ upper_bound = lower_bound;
+
+ if (lower_bound < 0 || upper_bound > RE_DUP_MAX
+ || lower_bound > upper_bound)
+ {
+ if (syntax & RE_NO_BK_BRACES)
+ goto unfetch_interval;
+ else
+ FREE_STACK_RETURN (REG_BADBR);
+ }
+
+ if (!(syntax & RE_NO_BK_BRACES))
+ {
+ if (c != '\\') FREE_STACK_RETURN (REG_EBRACE);
+
+ PATFETCH (c);
+ }
+
+ if (c != '}')
+ {
+ if (syntax & RE_NO_BK_BRACES)
+ goto unfetch_interval;
+ else
+ FREE_STACK_RETURN (REG_BADBR);
+ }
+
+ /* We just parsed a valid interval. */
+
+ /* If it's invalid to have no preceding re. */
+ if (!laststart)
+ {
+ if (syntax & RE_CONTEXT_INVALID_OPS)
+ FREE_STACK_RETURN (REG_BADRPT);
+ else if (syntax & RE_CONTEXT_INDEP_OPS)
+ laststart = b;
+ else
+ goto unfetch_interval;
+ }
+
+ /* If the upper bound is zero, don't want to succeed at
+ all; jump from `laststart' to `b + 3', which will be
+ the end of the buffer after we insert the jump. */
+ if (upper_bound == 0)
+ {
+ GET_BUFFER_SPACE (3);
+ INSERT_JUMP (jump, laststart, b + 3);
+ b += 3;
+ }
+
+ /* Otherwise, we have a nontrivial interval. When
+ we're all done, the pattern will look like:
+ set_number_at <jump count> <upper bound>
+ set_number_at <succeed_n count> <lower bound>
+ succeed_n <after jump addr> <succeed_n count>
+ <body of loop>
+ jump_n <succeed_n addr> <jump count>
+ (The upper bound and `jump_n' are omitted if
+ `upper_bound' is 1, though.) */
+ else
+ { /* If the upper bound is > 1, we need to insert
+ more at the end of the loop. */
+ unsigned nbytes = 10 + (upper_bound > 1) * 10;
+
+ GET_BUFFER_SPACE (nbytes);
+
+ /* Initialize lower bound of the `succeed_n', even
+ though it will be set during matching by its
+ attendant `set_number_at' (inserted next),
+ because `re_compile_fastmap' needs to know.
+ Jump to the `jump_n' we might insert below. */
+ INSERT_JUMP2 (succeed_n, laststart,
+ b + 5 + (upper_bound > 1) * 5,
+ lower_bound);
+ b += 5;
+
+ /* Code to initialize the lower bound. Insert
+ before the `succeed_n'. The `5' is the last two
+ bytes of this `set_number_at', plus 3 bytes of
+ the following `succeed_n'. */
+ insert_op2 (set_number_at, laststart, 5, lower_bound, b);
+ b += 5;
+
+ if (upper_bound > 1)
+ { /* More than one repetition is allowed, so
+ append a backward jump to the `succeed_n'
+ that starts this interval.
+
+ When we've reached this during matching,
+ we'll have matched the interval once, so
+ jump back only `upper_bound - 1' times. */
+ STORE_JUMP2 (jump_n, b, laststart + 5,
+ upper_bound - 1);
+ b += 5;
+
+ /* The location we want to set is the second
+ parameter of the `jump_n'; that is `b-2' as
+ an absolute address. `laststart' will be
+ the `set_number_at' we're about to insert;
+ `laststart+3' the number to set, the source
+ for the relative address. But we are
+ inserting into the middle of the pattern --
+ so everything is getting moved up by 5.
+ Conclusion: (b - 2) - (laststart + 3) + 5,
+ i.e., b - laststart.
+
+ We insert this at the beginning of the loop
+ so that if we fail during matching, we'll
+ reinitialize the bounds. */
+ insert_op2 (set_number_at, laststart, b - laststart,
+ upper_bound - 1, b);
+ b += 5;
+ }
+ }
+ pending_exact = 0;
+ beg_interval = NULL;
+ }
+ break;
+
+ unfetch_interval:
+ /* If an invalid interval, match the characters as literals. */
+ assert (beg_interval);
+ p = beg_interval;
+ beg_interval = NULL;
+
+ /* normal_char and normal_backslash need `c'. */
+ PATFETCH (c);
+
+ if (!(syntax & RE_NO_BK_BRACES))
+ {
+ if (p > pattern && p[-1] == '\\')
+ goto normal_backslash;
+ }
+ goto normal_char;
- goto fail;
-
-
- /* on_failure_keep_string_jump is used to optimize `.*\n'. It
- pushes NULL as the value for the string on the stack. Then
- `pop_failure_point' will keep the current value for the
- string, instead of restoring it. To see why, consider
- matching `foo\nbar' against `.*\n'. The .* matches the foo;
- then the . fails against the \n. But the next thing we want
- to do is match the \n against the \n; if we restored the
- string value, we would be back at the foo.
-
- Because this is used only in specific cases, we don't need to
- check all the things that `on_failure_jump' does, to make
- sure the right things get saved on the stack. Hence we don't
- share its code. The only reason to push anything on the
- stack at all is that otherwise we would have to change
- `anychar's code to do something besides goto fail in this
- case; that seems worse than this. */
- case on_failure_keep_string_jump:
- DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
-
- EXTRACT_NUMBER_AND_INCR (mcnt, p);
- DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
-
- PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
- break;
+ goto fail;
+
+
+ /* on_failure_keep_string_jump is used to optimize `.*\n'. It
+ pushes NULL as the value for the string on the stack. Then
+ `pop_failure_point' will keep the current value for the
+ string, instead of restoring it. To see why, consider
+ matching `foo\nbar' against `.*\n'. The .* matches the foo;
+ then the . fails against the \n. But the next thing we want
+ to do is match the \n against the \n; if we restored the
+ string value, we would be back at the foo.
+
+ Because this is used only in specific cases, we don't need to
+ check all the things that `on_failure_jump' does, to make
+ sure the right things get saved on the stack. Hence we don't
+ share its code. The only reason to push anything on the
+ stack at all is that otherwise we would have to change
+ `anychar's code to do something besides goto fail in this
+ case; that seems worse than this. */
+ case on_failure_keep_string_jump:
+ DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
+
+ EXTRACT_NUMBER_AND_INCR (mcnt, p);
+ DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
+
+ PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
+ break;