DUK_LOCAL void duk__append_range_atom_matcher(duk_re_compiler_ctx *re_ctx, duk_small_uint_t re_op, const duk_uint16_t *ranges, duk_small_uint_t count) {
#if 0
DUK_ASSERT(re_op <= 0x7fUL);
DUK_ASSERT(count <= 0x7fUL);
duk__append_2bytes(re_ctx, (duk_uint8_t) re_op, (duk_uint8_t) count);
#endif
duk__append_reop(re_ctx, re_op);
duk__append_7bit(re_ctx, count);
duk__append_u16_list(re_ctx, ranges, count * 2);
}
DUK_LOCAL void duk__parse_disjunction(duk_re_compiler_ctx *re_ctx, duk_bool_t expect_eof, duk__re_disjunction_info *out_atom_info) {
duk_int32_t atom_start_offset = -1; /* negative -> no atom matched on previous round */
duk_int32_t atom_char_length = 0; /* negative -> complex atom */
duk_uint32_t atom_start_captures = re_ctx->captures; /* value of re_ctx->captures at start of atom */
duk_int32_t unpatched_disjunction_split = -1;
duk_int32_t unpatched_disjunction_jump = -1;
duk_uint32_t entry_offset = (duk_uint32_t) DUK__RE_BUFLEN(re_ctx);
duk_int32_t res_charlen = 0; /* -1 if disjunction is complex, char length if simple */
duk__re_disjunction_info tmp_disj;
DUK_ASSERT(out_atom_info != NULL);
if (re_ctx->recursion_depth >= re_ctx->recursion_limit) {
DUK_ERROR_RANGE(re_ctx->thr, DUK_STR_REGEXP_COMPILER_RECURSION_LIMIT);
}
re_ctx->recursion_depth++;
#if 0
out_atom_info->start_captures = re_ctx->captures;
#endif
for (;;) {
/* atom_char_length, atom_start_offset, atom_start_offset reflect the
* atom matched on the previous loop. If a quantifier is encountered
* on this loop, these are needed to handle the quantifier correctly.
* new_atom_char_length etc are for the atom parsed on this round;
* they're written to atom_char_length etc at the end of the round.
*/
duk_int32_t new_atom_char_length; /* char length of the atom parsed in this loop */
duk_int32_t new_atom_start_offset; /* bytecode start offset of the atom parsed in this loop
* (allows quantifiers to copy the atom bytecode)
*/
duk_uint32_t new_atom_start_captures; /* re_ctx->captures at the start of the atom parsed in this loop */
duk_lexer_parse_re_token(&re_ctx->lex, &re_ctx->curr_token);
DUK_DD(DUK_DDPRINT("re token: %ld (num=%ld, char=%c)",
(long) re_ctx->curr_token.t,
(long) re_ctx->curr_token.num,
(re_ctx->curr_token.num >= 0x20 && re_ctx->curr_token.num <= 0x7e) ?
(int) re_ctx->curr_token.num : (int) '?'));
/* set by atom case clauses */
new_atom_start_offset = -1;
new_atom_char_length = -1;
new_atom_start_captures = re_ctx->captures;
switch (re_ctx->curr_token.t) {
case DUK_RETOK_DISJUNCTION: {
/*
* The handling here is a bit tricky. If a previous '|' has been processed,
* we have a pending split1 and a pending jump (for a previous match). These
* need to be back-patched carefully. See docs for a detailed example.
*/
/* patch pending jump and split */
if (unpatched_disjunction_jump >= 0) {
duk_uint32_t offset;
DUK_ASSERT(unpatched_disjunction_split >= 0);
offset = unpatched_disjunction_jump;
offset += duk__insert_jump_offset(re_ctx,
offset,
(duk_int32_t) (DUK__RE_BUFLEN(re_ctx) - offset));
/* offset is now target of the pending split (right after jump) */
duk__insert_jump_offset(re_ctx,
unpatched_disjunction_split,
offset - unpatched_disjunction_split);
}
/* add a new pending split to the beginning of the entire disjunction */
(void) duk__insert_u32(re_ctx,
entry_offset,
DUK_REOP_SPLIT1); /* prefer direct execution */
unpatched_disjunction_split = entry_offset + 1; /* +1 for opcode */
/* add a new pending match jump for latest finished alternative */
duk__append_u32(re_ctx, DUK_REOP_JUMP);
unpatched_disjunction_jump = (duk_int32_t) DUK__RE_BUFLEN(re_ctx);
/* 'taint' result as complex */
res_charlen = -1;
break;
}
case DUK_RETOK_QUANTIFIER: {
if (atom_start_offset < 0) {
DUK_ERROR_SYNTAX(re_ctx->thr, DUK_STR_INVALID_QUANTIFIER_NO_ATOM);
}
if (re_ctx->curr_token.qmin > re_ctx->curr_token.qmax) {
DUK_ERROR_SYNTAX(re_ctx->thr, DUK_STR_INVALID_QUANTIFIER_VALUES);
}
if (atom_char_length >= 0) {
/*
* Simple atom
*
* If atom_char_length is zero, we'll have unbounded execution time for e.g.
* /()*x/.exec('x'). We can't just skip the match because it might have some
* side effects (for instance, if we allowed captures in simple atoms, the
* capture needs to happen). The simple solution below is to force the
* quantifier to match at most once, since the additional matches have no effect.
*
* With a simple atom there can be no capture groups, so no captures need
* to be reset.
*/
duk_int32_t atom_code_length;
duk_uint32_t offset;
duk_uint32_t qmin, qmax;
qmin = re_ctx->curr_token.qmin;
qmax = re_ctx->curr_token.qmax;
if (atom_char_length == 0) {
/* qmin and qmax will be 0 or 1 */
if (qmin > 1) {
qmin = 1;
}
if (qmax > 1) {
qmax = 1;
}
}
duk__append_u32(re_ctx, DUK_REOP_MATCH); /* complete 'sub atom' */
atom_code_length = (duk_int32_t) (DUK__RE_BUFLEN(re_ctx) - atom_start_offset);
offset = atom_start_offset;
if (re_ctx->curr_token.greedy) {
offset += duk__insert_u32(re_ctx, offset, DUK_REOP_SQGREEDY);
offset += duk__insert_u32(re_ctx, offset, qmin);
offset += duk__insert_u32(re_ctx, offset, qmax);
offset += duk__insert_u32(re_ctx, offset, atom_char_length);
offset += duk__insert_jump_offset(re_ctx, offset, atom_code_length);
} else {
offset += duk__insert_u32(re_ctx, offset, DUK_REOP_SQMINIMAL);
offset += duk__insert_u32(re_ctx, offset, qmin);
offset += duk__insert_u32(re_ctx, offset, qmax);
offset += duk__insert_jump_offset(re_ctx, offset, atom_code_length);
}
DUK_UNREF(offset); /* silence scan-build warning */
} else {
/*
* Complex atom
*
* The original code is used as a template, and removed at the end
* (this differs from the handling of simple quantifiers).
*
* NOTE: there is no current solution for empty atoms in complex
* quantifiers. This would need some sort of a 'progress' instruction.
*
* XXX: impose limit on maximum result size, i.e. atom_code_len * atom_copies?
*/
duk_int32_t atom_code_length;
duk_uint32_t atom_copies;
duk_uint32_t tmp_qmin, tmp_qmax;
/* pre-check how many atom copies we're willing to make (atom_copies not needed below) */
atom_copies = (re_ctx->curr_token.qmax == DUK_RE_QUANTIFIER_INFINITE) ?
re_ctx->curr_token.qmin : re_ctx->curr_token.qmax;
if (atom_copies > DUK_RE_MAX_ATOM_COPIES) {
DUK_ERROR_RANGE(re_ctx->thr, DUK_STR_QUANTIFIER_TOO_MANY_COPIES);
}
/* wipe the capture range made by the atom (if any) */
DUK_ASSERT(atom_start_captures <= re_ctx->captures);
if (atom_start_captures != re_ctx->captures) {
DUK_ASSERT(atom_start_captures < re_ctx->captures);
DUK_DDD(DUK_DDDPRINT("must wipe ]atom_start_captures,re_ctx->captures]: ]%ld,%ld]",
(long) atom_start_captures, (long) re_ctx->captures));
/* insert (DUK_REOP_WIPERANGE, start, count) in reverse order so the order ends up right */
duk__insert_u32(re_ctx, atom_start_offset, (re_ctx->captures - atom_start_captures) * 2);
duk__insert_u32(re_ctx, atom_start_offset, (atom_start_captures + 1) * 2);
duk__insert_u32(re_ctx, atom_start_offset, DUK_REOP_WIPERANGE);
} else {
DUK_DDD(DUK_DDDPRINT("no need to wipe captures: atom_start_captures == re_ctx->captures == %ld",
(long) atom_start_captures));
}
atom_code_length = (duk_int32_t) DUK__RE_BUFLEN(re_ctx) - atom_start_offset;
/* insert the required matches (qmin) by copying the atom */
tmp_qmin = re_ctx->curr_token.qmin;
tmp_qmax = re_ctx->curr_token.qmax;
while (tmp_qmin > 0) {
duk__append_slice(re_ctx, atom_start_offset, atom_code_length);
tmp_qmin--;
if (tmp_qmax != DUK_RE_QUANTIFIER_INFINITE) {
tmp_qmax--;
}
}
DUK_ASSERT(tmp_qmin == 0);
/* insert code for matching the remainder - infinite or finite */
if (tmp_qmax == DUK_RE_QUANTIFIER_INFINITE) {
/* reuse last emitted atom for remaining 'infinite' quantifier */
if (re_ctx->curr_token.qmin == 0) {
/* Special case: original qmin was zero so there is nothing
* to repeat. Emit an atom copy but jump over it here.
*/
duk__append_u32(re_ctx, DUK_REOP_JUMP);
duk__append_jump_offset(re_ctx, atom_code_length);
duk__append_slice(re_ctx, atom_start_offset, atom_code_length);
}
if (re_ctx->curr_token.greedy) {
duk__append_u32(re_ctx, DUK_REOP_SPLIT2); /* prefer jump */
} else {
duk__append_u32(re_ctx, DUK_REOP_SPLIT1); /* prefer direct */
}
duk__append_jump_offset(re_ctx, -atom_code_length - 1); /* -1 for opcode */
} else {
/*
* The remaining matches are emitted as sequence of SPLITs and atom
* copies; the SPLITs skip the remaining copies and match the sequel.
* This sequence needs to be emitted starting from the last copy
* because the SPLITs are variable length due to the variable length
* skip offset. This causes a lot of memory copying now.
*
* Example structure (greedy, match maximum # atoms):
*
* SPLIT1 LSEQ
* (atom)
* SPLIT1 LSEQ ; <- the byte length of this instruction is needed
* (atom) ; to encode the above SPLIT1 correctly
* ...
* LSEQ:
*/
duk_uint32_t offset = (duk_uint32_t) DUK__RE_BUFLEN(re_ctx);
while (tmp_qmax > 0) {
duk__insert_slice(re_ctx, offset, atom_start_offset, atom_code_length);
if (re_ctx->curr_token.greedy) {
duk__insert_u32(re_ctx, offset, DUK_REOP_SPLIT1); /* prefer direct */
} else {
duk__insert_u32(re_ctx, offset, DUK_REOP_SPLIT2); /* prefer jump */
}
duk__insert_jump_offset(re_ctx,
offset + 1, /* +1 for opcode */
(duk_int32_t) (DUK__RE_BUFLEN(re_ctx) - (offset + 1)));
tmp_qmax--;
}
}
/* remove the original 'template' atom */
duk__remove_slice(re_ctx, atom_start_offset, atom_code_length);
}
/* 'taint' result as complex */
res_charlen = -1;
break;
}
case DUK_RETOK_ASSERT_START: {
duk__append_u32(re_ctx, DUK_REOP_ASSERT_START);
break;
}
case DUK_RETOK_ASSERT_END: {
duk__append_u32(re_ctx, DUK_REOP_ASSERT_END);
break;
}
case DUK_RETOK_ASSERT_WORD_BOUNDARY: {
duk__append_u32(re_ctx, DUK_REOP_ASSERT_WORD_BOUNDARY);
break;
}
case DUK_RETOK_ASSERT_NOT_WORD_BOUNDARY: {
duk__append_u32(re_ctx, DUK_REOP_ASSERT_NOT_WORD_BOUNDARY);
break;
}
case DUK_RETOK_ASSERT_START_POS_LOOKAHEAD:
case DUK_RETOK_ASSERT_START_NEG_LOOKAHEAD: {
duk_uint32_t offset;
duk_uint32_t opcode = (re_ctx->curr_token.t == DUK_RETOK_ASSERT_START_POS_LOOKAHEAD) ?
DUK_REOP_LOOKPOS : DUK_REOP_LOOKNEG;
offset = (duk_uint32_t) DUK__RE_BUFLEN(re_ctx);
duk__parse_disjunction(re_ctx, 0, &tmp_disj);
duk__append_u32(re_ctx, DUK_REOP_MATCH);
(void) duk__insert_u32(re_ctx, offset, opcode);
(void) duk__insert_jump_offset(re_ctx,
offset + 1, /* +1 for opcode */
(duk_int32_t) (DUK__RE_BUFLEN(re_ctx) - (offset + 1)));
/* 'taint' result as complex -- this is conservative,
* as lookaheads do not backtrack.
*/
res_charlen = -1;
break;
}
case DUK_RETOK_ATOM_PERIOD: {
new_atom_char_length = 1;
new_atom_start_offset = (duk_int32_t) DUK__RE_BUFLEN(re_ctx);
duk__append_u32(re_ctx, DUK_REOP_PERIOD);
break;
}
case DUK_RETOK_ATOM_CHAR: {
/* Note: successive characters could be joined into string matches
* but this is not trivial (consider e.g. '/xyz+/); see docs for
* more discussion.
*/
duk_uint32_t ch;
new_atom_char_length = 1;
new_atom_start_offset = (duk_int32_t) DUK__RE_BUFLEN(re_ctx);
duk__append_u32(re_ctx, DUK_REOP_CHAR);
ch = re_ctx->curr_token.num;
if (re_ctx->re_flags & DUK_RE_FLAG_IGNORE_CASE) {
ch = duk_unicode_re_canonicalize_char(re_ctx->thr, ch);
}
duk__append_u32(re_ctx, ch);
break;
}
case DUK_RETOK_ATOM_DIGIT:
case DUK_RETOK_ATOM_NOT_DIGIT: {
new_atom_char_length = 1;
new_atom_start_offset = (duk_int32_t) DUK__RE_BUFLEN(re_ctx);
duk__append_u32(re_ctx,
(re_ctx->curr_token.t == DUK_RETOK_ATOM_DIGIT) ?
DUK_REOP_RANGES : DUK_REOP_INVRANGES);
duk__append_u32(re_ctx, sizeof(duk_unicode_re_ranges_digit) / (2 * sizeof(duk_uint16_t)));
duk__append_u16_list(re_ctx, duk_unicode_re_ranges_digit, sizeof(duk_unicode_re_ranges_digit) / sizeof(duk_uint16_t));
break;
}
case DUK_RETOK_ATOM_WHITE:
case DUK_RETOK_ATOM_NOT_WHITE: {
new_atom_char_length = 1;
new_atom_start_offset = (duk_int32_t) DUK__RE_BUFLEN(re_ctx);
duk__append_u32(re_ctx,
(re_ctx->curr_token.t == DUK_RETOK_ATOM_WHITE) ?
DUK_REOP_RANGES : DUK_REOP_INVRANGES);
duk__append_u32(re_ctx, sizeof(duk_unicode_re_ranges_white) / (2 * sizeof(duk_uint16_t)));
duk__append_u16_list(re_ctx, duk_unicode_re_ranges_white, sizeof(duk_unicode_re_ranges_white) / sizeof(duk_uint16_t));
break;
}
case DUK_RETOK_ATOM_WORD_CHAR:
case DUK_RETOK_ATOM_NOT_WORD_CHAR: {
new_atom_char_length = 1;
new_atom_start_offset = (duk_int32_t) DUK__RE_BUFLEN(re_ctx);
duk__append_u32(re_ctx,
(re_ctx->curr_token.t == DUK_RETOK_ATOM_WORD_CHAR) ?
DUK_REOP_RANGES : DUK_REOP_INVRANGES);
duk__append_u32(re_ctx, sizeof(duk_unicode_re_ranges_wordchar) / (2 * sizeof(duk_uint16_t)));
duk__append_u16_list(re_ctx, duk_unicode_re_ranges_wordchar, sizeof(duk_unicode_re_ranges_wordchar) / sizeof(duk_uint16_t));
break;
}
case DUK_RETOK_ATOM_BACKREFERENCE: {
duk_uint32_t backref = (duk_uint32_t) re_ctx->curr_token.num;
if (backref > re_ctx->highest_backref) {
re_ctx->highest_backref = backref;
}
new_atom_char_length = -1; /* mark as complex */
new_atom_start_offset = (duk_int32_t) DUK__RE_BUFLEN(re_ctx);
duk__append_u32(re_ctx, DUK_REOP_BACKREFERENCE);
duk__append_u32(re_ctx, backref);
break;
}
case DUK_RETOK_ATOM_START_CAPTURE_GROUP: {
duk_uint32_t cap;
new_atom_char_length = -1; /* mark as complex (capture handling) */
new_atom_start_offset = (duk_int32_t) DUK__RE_BUFLEN(re_ctx);
cap = ++re_ctx->captures;
duk__append_u32(re_ctx, DUK_REOP_SAVE);
duk__append_u32(re_ctx, cap * 2);
duk__parse_disjunction(re_ctx, 0, &tmp_disj); /* retval (sub-atom char length) unused, tainted as complex above */
duk__append_u32(re_ctx, DUK_REOP_SAVE);
duk__append_u32(re_ctx, cap * 2 + 1);
break;
}
case DUK_RETOK_ATOM_START_NONCAPTURE_GROUP: {
new_atom_start_offset = (duk_int32_t) DUK__RE_BUFLEN(re_ctx);
duk__parse_disjunction(re_ctx, 0, &tmp_disj);
new_atom_char_length = tmp_disj.charlen;
break;
}
case DUK_RETOK_ATOM_START_CHARCLASS:
case DUK_RETOK_ATOM_START_CHARCLASS_INVERTED: {
/*
* Range parsing is done with a special lexer function which calls
* us for every range parsed. This is different from how rest of
* the parsing works, but avoids a heavy, arbitrary size intermediate
* value type to hold the ranges.
*
* Another complication is the handling of character ranges when
* case insensitive matching is used (see docs for discussion).
* The range handler callback given to the lexer takes care of this
* as well.
*
* Note that duplicate ranges are not eliminated when parsing character
* classes, so that canonicalization of
*
* [0-9a-fA-Fx-{]
*
* creates the result (note the duplicate ranges):
*
* [0-9A-FA-FX-Z{-{]
*
* where [x-{] is split as a result of canonicalization. The duplicate
* ranges are not a semantics issue: they work correctly.
*/
duk_uint32_t offset;
DUK_DD(DUK_DDPRINT("character class"));
/* insert ranges instruction, range count patched in later */
new_atom_char_length = 1;
new_atom_start_offset = (duk_int32_t) DUK__RE_BUFLEN(re_ctx);
duk__append_u32(re_ctx,
(re_ctx->curr_token.t == DUK_RETOK_ATOM_START_CHARCLASS) ?
DUK_REOP_RANGES : DUK_REOP_INVRANGES);
offset = (duk_uint32_t) DUK__RE_BUFLEN(re_ctx); /* patch in range count later */
/* parse ranges until character class ends */
re_ctx->nranges = 0; /* note: ctx-wide temporary */
duk_lexer_parse_re_ranges(&re_ctx->lex, duk__generate_ranges, (void *) re_ctx);
/* insert range count */
duk__insert_u32(re_ctx, offset, re_ctx->nranges);
break;
}
case DUK_RETOK_ATOM_END_GROUP: {
if (expect_eof) {
DUK_ERROR_SYNTAX(re_ctx->thr, DUK_STR_UNEXPECTED_CLOSING_PAREN);
}
goto done;
}
case DUK_RETOK_EOF: {
if (!expect_eof) {
DUK_ERROR_SYNTAX(re_ctx->thr, DUK_STR_UNEXPECTED_END_OF_PATTERN);
}
goto done;
}
default: {
DUK_ERROR_SYNTAX(re_ctx->thr, DUK_STR_UNEXPECTED_REGEXP_TOKEN);
}
}
/* a complex (new) atom taints the result */
if (new_atom_start_offset >= 0) {
if (new_atom_char_length < 0) {
res_charlen = -1;
} else if (res_charlen >= 0) {
/* only advance if not tainted */
res_charlen += new_atom_char_length;
}
}
/* record previous atom info in case next token is a quantifier */
atom_start_offset = new_atom_start_offset;
atom_char_length = new_atom_char_length;
atom_start_captures = new_atom_start_captures;
}
done:
/* finish up pending jump and split for last alternative */
if (unpatched_disjunction_jump >= 0) {
duk_uint32_t offset;
DUK_ASSERT(unpatched_disjunction_split >= 0);
offset = unpatched_disjunction_jump;
offset += duk__insert_jump_offset(re_ctx,
offset,
(duk_int32_t) (DUK__RE_BUFLEN(re_ctx) - offset));
/* offset is now target of the pending split (right after jump) */
duk__insert_jump_offset(re_ctx,
unpatched_disjunction_split,
offset - unpatched_disjunction_split);
}
#if 0
out_atom_info->end_captures = re_ctx->captures;
#endif
out_atom_info->charlen = res_charlen;
DUK_DDD(DUK_DDDPRINT("parse disjunction finished: charlen=%ld",
(long) out_atom_info->charlen));
re_ctx->recursion_depth--;
}
