After removing collation for [a-z] ranges in r302512, do it here too.
Instead of trying to expand whole range at regcomp() stage as we do, GNU regex allocates separate ranges [start,end] set each character is checked against, so collation is possible and turned on for ranges here. When something like that will be implemented or our obsoleted regex code will be replaced to something like TRE, and in case we decide to use collation in [a-z] ranges, all changes related to r302512 can be backed out, but now we need consistency.
This commit is contained in:
parent
dbaa6b077d
commit
a66b222850
@ -2664,7 +2664,11 @@ build_range_exp (bitset_t sbcset, bracket_elem_t *start_elem,
|
||||
return REG_ECOLLATE;
|
||||
cmp_buf[0] = start_wc;
|
||||
cmp_buf[4] = end_wc;
|
||||
#ifdef __FreeBSD__
|
||||
if (wcscmp (cmp_buf, cmp_buf + 4) > 0)
|
||||
#else
|
||||
if (wcscoll (cmp_buf, cmp_buf + 4) > 0)
|
||||
#endif
|
||||
return REG_ERANGE;
|
||||
|
||||
/* Got valid collation sequence values, add them as a new entry.
|
||||
@ -2706,8 +2710,13 @@ build_range_exp (bitset_t sbcset, bracket_elem_t *start_elem,
|
||||
for (wc = 0; wc < SBC_MAX; ++wc)
|
||||
{
|
||||
cmp_buf[2] = wc;
|
||||
#ifdef __FreeBSD__
|
||||
if (wcscmp (cmp_buf, cmp_buf + 2) <= 0
|
||||
&& wcscmp (cmp_buf + 2, cmp_buf + 4) <= 0)
|
||||
#else
|
||||
if (wcscoll (cmp_buf, cmp_buf + 2) <= 0
|
||||
&& wcscoll (cmp_buf + 2, cmp_buf + 4) <= 0)
|
||||
#endif
|
||||
bitset_set (sbcset, wc);
|
||||
}
|
||||
}
|
||||
|
@ -3964,8 +3964,13 @@ check_node_accept_bytes (const re_dfa_t *dfa, int node_idx,
|
||||
{
|
||||
cmp_buf[0] = cset->range_starts[i];
|
||||
cmp_buf[4] = cset->range_ends[i];
|
||||
#ifdef __FreeBSD__
|
||||
if (wcscmp (cmp_buf, cmp_buf + 2) <= 0
|
||||
&& wcscmp (cmp_buf + 2, cmp_buf + 4) <= 0)
|
||||
#else
|
||||
if (wcscoll (cmp_buf, cmp_buf + 2) <= 0
|
||||
&& wcscoll (cmp_buf + 2, cmp_buf + 4) <= 0)
|
||||
#endif
|
||||
{
|
||||
match_len = char_len;
|
||||
goto check_node_accept_bytes_match;
|
||||
|
Loading…
Reference in New Issue
Block a user