| Message ID | aeffa34acdcb8da840639ba67cf8984d2bd343d6.1686530834.git.nabijaczleweli@nabijaczleweli.xyz |
|---|---|
| State | New |
| Headers | show |
| Series | [v7,1/3] posix: regcomp(): clear RE_DOT_NOT_NULL | expand |
On 6/11/23 20:47, наб wrote: > The POSIX API always stops at first NUL so there's no change for that. > > The BSD REG_STARTEND API, with its explicit range, can include NULs > within that range, and those NULs are matched with . and [^]. > > Heretofor, for a string of "a\0c", glibc would match "[^q]c", but not > ".c". This is both inconsistent and nonconformant to BSD REG_STARTEND. > > With this patch, they're identical like you'd expect, and the > tst-reg-startend.c: ..c: a^@c: no match$ > failure is removed. > > Another approach would be to remove it from _RE_SYNTAX_POSIX_COMMON, > but it's unclear to me what the custody chain is like for that and what > other regex APIs glibc offers that could be affected by this. > > Signed-off-by: Ahelenia Ziemiańska <nabijaczleweli@nabijaczleweli.xyz> These changes are being made to sources shared between gnulib and glibc. As the files are listed in SHARED-SOURCES we cannot easily accept changes to them since they should be shared with gnulib. Would you be willing to disclaim these changes or assign copyright? > --- > No changes (rebased cleanly); full resend. > > posix/regcomp.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/posix/regcomp.c b/posix/regcomp.c > index 12650714c0..a928ef6c2d 100644 > --- a/posix/regcomp.c > +++ b/posix/regcomp.c > @@ -462,7 +462,7 @@ regcomp (regex_t *__restrict preg, const char *__restrict pattern, int cflags) > { > reg_errcode_t ret; > reg_syntax_t syntax = ((cflags & REG_EXTENDED) ? RE_SYNTAX_POSIX_EXTENDED > - : RE_SYNTAX_POSIX_BASIC); > + : RE_SYNTAX_POSIX_BASIC) & ~RE_DOT_NOT_NULL; > > preg->buffer = NULL; > preg->allocated = 0;
diff --git a/posix/regcomp.c b/posix/regcomp.c index 12650714c0..a928ef6c2d 100644 --- a/posix/regcomp.c +++ b/posix/regcomp.c @@ -462,7 +462,7 @@ regcomp (regex_t *__restrict preg, const char *__restrict pattern, int cflags) { reg_errcode_t ret; reg_syntax_t syntax = ((cflags & REG_EXTENDED) ? RE_SYNTAX_POSIX_EXTENDED - : RE_SYNTAX_POSIX_BASIC); + : RE_SYNTAX_POSIX_BASIC) & ~RE_DOT_NOT_NULL; preg->buffer = NULL; preg->allocated = 0;
The POSIX API always stops at first NUL so there's no change for that. The BSD REG_STARTEND API, with its explicit range, can include NULs within that range, and those NULs are matched with . and [^]. Heretofor, for a string of "a\0c", glibc would match "[^q]c", but not ".c". This is both inconsistent and nonconformant to BSD REG_STARTEND. With this patch, they're identical like you'd expect, and the tst-reg-startend.c: ..c: a^@c: no match$ failure is removed. Another approach would be to remove it from _RE_SYNTAX_POSIX_COMMON, but it's unclear to me what the custody chain is like for that and what other regex APIs glibc offers that could be affected by this. Signed-off-by: Ahelenia Ziemiańska <nabijaczleweli@nabijaczleweli.xyz> --- No changes (rebased cleanly); full resend. posix/regcomp.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)