regexp-exec critcial section errors

regexp-exec critcial section errors

[Kevin Ryde]

[-- Attachment #1: Type: text/plain, Size: 777 bytes --]

In regexp-exec in 1.8, a nul in the input string, or a bad flags arg,
results in a throw from a critical section, which calls abort().  Eg.

	(regexp-exec (make-regexp "x") "\0")
	=> abort

	(regexp-exec (make-regexp "x") "hello" 0 'badflags)
	=> abort

What's the reason for the critical section?  Do we think regexec() is
not reentrant?  It looks like a conversion from SCM_DEFER_INTS in past
versions, if that suggests anything about it.

I struck this when running a regexp search on some file data that
slightly unexpectedly had nuls in it.  I think I was using 1.6 last
time I did the same thing.  I guess that version silently ignores past
the nul.  Maybe that should get a check, though there's probably a
zillion libc calls that would really ought to have the same.



[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: regex-posix.c.critical.diff --]
[-- Type: text/x-diff, Size: 636 bytes --]

--- regex-posix.c.~1.78.2.1.~	2006-02-14 08:59:02.000000000 +1100
+++ regex-posix.c	2007-01-13 12:53:32.000000000 +1100
@@ -245,7 +245,6 @@
      whole regexp, so add 1 to nmatches. */
 
   nmatches = SCM_RGX(rx)->re_nsub + 1;
-  SCM_CRITICAL_SECTION_START;
   matches = scm_malloc (sizeof (regmatch_t) * nmatches);
   c_str = scm_to_locale_string (substr);
   status = regexec (SCM_RGX (rx), c_str, nmatches, matches,
@@ -269,7 +268,6 @@
 				   scm_from_long (matches[i].rm_eo + offset)));
     }
   free (matches);
-  SCM_CRITICAL_SECTION_END;
 
   if (status != 0 && status != REG_NOMATCH)
     scm_error_scm (scm_regexp_error_key,

[-- Attachment #3: Type: text/plain, Size: 143 bytes --]

_______________________________________________
Guile-devel mailing list
Guile-devel@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-devel

Re: regexp-exec critcial section errors

[Han-Wen Nienhuys]

Kevin Ryde escreveu:
> In regexp-exec in 1.8, a nul in the input string, or a bad flags arg,
> results in a throw from a critical section, which calls abort().  Eg.
> 
> 	(regexp-exec (make-regexp "x") "\0")
> 	=> abort
> 
> 	(regexp-exec (make-regexp "x") "hello" 0 'badflags)
> 	=> abort
> 
> What's the reason for the critical section?  Do we think regexec() is
> not reentrant?  It looks like a conversion from SCM_DEFER_INTS in past
> versions, if that suggests anything about it.
> 
> I struck this when running a regexp search on some file data that
> slightly unexpectedly had nuls in it.  I think I was using 1.6 last
> time I did the same thing.  I guess that version silently ignores past
> the nul.  Maybe that should get a check, though there's probably a
> zillion libc calls that would really ought to have the same.

Ehm,

I get

**
Running reader.test
Running receive.test
Running regexp.test
ERROR: Unbound variable: exception:string-contains-nul
FAIL: check-guile
==================================
**


it might be a good idea to have an autobuild daemon that compiles and 
checks the test-suite.

-- 
 Han-Wen Nienhuys - hanwen@xs4all.nl - http://www.xs4all.nl/~hanwen



_______________________________________________
Guile-devel mailing list
Guile-devel@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-devel

Re: regexp-exec critcial section errors

[Kevin Ryde]

Han-Wen Nienhuys <hanwen@xs4all.nl> writes:
>
> ERROR: Unbound variable: exception:string-contains-nul

Oops, not enough checkin, fixed now.


_______________________________________________
Guile-devel mailing list
Guile-devel@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-devel