threaded perl-5.18.0 fails re/pat_rt_report_thr.t on i586 #12993
Comments
From @shlomifThis is a bug report for perl from shlomif@shlomifish.org, Threaded perl-5.18.0 fails the re/pat_rt_report_thr.t test script on [QUOTE] # This script reproduces a failure at the "make test" stage with perl-5.18.0 # Based on the Mageia's perl-5.18.0-1's .src.rpm's .spec. sh Configure -des \ [/QUOTE] [QUOTE] # This script reproduces a failure at the "make test" stage with perl-5.18.0 # Based on the Mageia's perl-5.18.0-1's .src.rpm's .spec. sh Configure -des \ [/QUOTE] This bug was discovered when building perl-5.18.0 for the Mageia development https://github.com/shlomif/build-perl-with-failure-on-i586 The test failures are: [QUOTE] Regards, -- Shlomi Fish Flags: Site configuration information for perl 5.16.3: Configured by Mageia at Sun Mar 17 18:51:26 UTC 2013. Summary of my perl5 (revision 5 version 16 subversion 3) configuration: Locally applied patches: @INC for perl 5.16.3: Environment for perl 5.16.3: Complete configuration data for perl 5.16.3: Author='' -- Shlomi Fish http://www.shlomifish.org/ I hope that you agree with me that 99.9218485921% of the users wouldn’t bother Please reply to list if it's a mailing list post - http://shlom.in/reply . |
From @jkeenanOn Sat May 25 09:36:11 2013, shlomif@shlomifish.org wrote:
I built with these arguments on dromedary and all tests passed. Now, That suggests that the test failures you witnessed are specific to the I would recommend trying other builds on this system, adding one Thank you very much. |
|
The RT System itself - Status changed from 'new' to 'open' |
From @shlomifHi James, On Sat May 25 17:19:15 2013, jkeenan wrote:
The fact that it fails on i586 was the whole point. All tests are Can't you set up an i586 Linux VM? See https://www.virtualbox.org/ .
Yes, maybe.
Yes, I already reduced the flags considerably and I will reduce them Regards, -- Shlomi Fish
|
From @shlomifHi all, On Sun May 26 01:11:10 2013, shlomif wrote:
I was able to reproduce it on Mageia Linux 3 i586 with this: https://github.com/shlomif/build-perl-with-failure-on-i586/blob/master/build-perl-with-failure.bash < QUOTE > Is there a need for further reduction? I am now going to try that on a Debian 7.x (Stable) i386 VM. Regards, -- Shlomi Fish |
From @shlomifHi all, On Sun May 26 01:48:36 2013, shlomif wrote:
With these flags on debian-7.0.0-i686 (installed from the LXDE CD at Regards, -- Shlomi Fish |
From @shlomifOn Sun May 26 02:41:06 2013, shlomif wrote:
And all tests are successful on Debian-7.0.0-i385 stable. Regards, -- Shlomi Fish
|
From @nwc10On Sun, May 26, 2013 at 02:48:36AM -0700, Shlomi Fish via RT wrote:
I can't replicate this on a 32-bit build on a Debian 6 system. Everything Nicholas Clark |
From @shlomifHi Nicholas, On Mon, 27 May 2013 15:11:51 +0100
Would it be possible for you to set up a Mageia 3 32-bit i586 VM so you can Regards, Shlomi Fish -- Shlomi Fish http://www.shlomifish.org/ Mephiqoleth: THERE ARE JEWS OF MANY SPECIES. Please reply to list if it's a mailing list post - http://shlom.in/reply . |
From @rjbs* Shlomi Fish <shlomif@shlomifish.org> [2013-05-27T13:03:30]
Have you tried contacting someone with a vested interest in having perl work on -- |
From @shlomifHi Ricardo, On Mon May 27 10:16:43 2013, perl.p5p@rjbs.manxome.org wrote:
Well, I am part of the Magiea community - being a packager for them and In any case, installing a VM of a system that causes problems is not Regards, -- Shlomi Fish |
From @shlomifHi all, I am now getting the same test failures after building a threaded perl Regards, -- Shlomi Fish |
From @tonycozOn Sat, May 25, 2013 at 09:36:12AM -0700, shlomif@shlomifish.org (via RT) wrote:
The failure on Megeia essentially comes down to: [tony@localhost perl]$ ./perl -Ilib -Mthreads -le 'sub foo { $s = "\x{1ff}" . "f" x 32000; my $r = $s =~ /\x{1ff}f*/gc; print $r; print pos($s) } foo()' Note that the pos() value is low even for smaller strings: [tony@localhost perl]$ ./perl -Ilib -Mthreads -le 'sub foo { $s = "\x{1ff}" . "f" x 32; my $r = $s =~ /\x{1ff}f*/gc; print $r; print pos($s) } my $t = threads->create(\&foo); $t->join' Tony |
From @tonycozOn Tue, May 28, 2013 at 07:07:07PM +1000, Tony Cook wrote:
The obvious thing to check: [tony@localhost perl]$ ./perl -Mre=debug -Ilib -Mthreads -le 'sub foo { $s = "\x{1ff}" . "f" x 32; my $r = $s =~ /\x{1ff}f*/gc; print $r; print pos($s) } my $t = threads->create(\&foo); $t->join' |
From @tonycozA bit of "printf" debugging, and I believe it's a bug in perl, not in S_regrepeat() sets max to I32_MAX when the supplied value is if (max == REG_INFTY) so at this point max is half our address space. If the character we're matching ('f' in this case) is invariant (it In the simple branch, then: if (utf8_target && scan + max < loceol) { Our test regexp is UTF-8, if scan were under 0x80000000 this code Trying to match EXACT \x66 vs \x66 (@b690046a eol 36900469 regeol b690048a) (scan = b690046a, loceol = 36900469, PL_regeol = b690048a) so our C<< scan + max < loceol >> condition is true, and the pointer The scan loop: while (scan < loceol && UCHARAT(scan) == c) { then terminates early since loceol starts out less than scan. Tony |
From @shlomifHi Tony, thanks for your help in this diagnosis. It's very appreciated. Regards, Shlomi Fish. On Tue, 28 May 2013 20:05:32 +1000
-- Shlomi Fish http://www.shlomifish.org/ <mst> I find it’s usually safe to assume that whatever shlomif’s doing, there Please reply to list if it's a mailing list post - http://shlom.in/reply . |
From @tonycozOn Tue, May 28, 2013 at 08:05:32PM +1000, Tony Cook wrote:
The attached patch fixes this particular instance of the bug, but from Tony |
From @tonycoz0001-avoid-overflowing-a-pointer-for-repeated-EXACT-nodes.patchFrom 932efe1a848ca53f7a355589d74c5a0a50e6063f Mon Sep 17 00:00:00 2001
From: Tony Cook <tony@develop-help.com>
Date: Tue, 28 May 2013 20:43:28 +1000
Subject: [PATCH] avoid overflowing a pointer for repeated EXACT nodes
there may be many other similar bugs in S_regrepeat
---
regexec.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/regexec.c b/regexec.c
index 4590a81..5f4da4b 100644
--- a/regexec.c
+++ b/regexec.c
@@ -6730,7 +6730,7 @@ S_regrepeat(pTHX_ regexp *prog, char **startposp, const regnode *p,
* can use UTF8_IS_INVARIANT() even if the pattern isn't UTF-8, as it's
* true iff it doesn't matter if the argument is in UTF-8 or not */
if (UTF8_IS_INVARIANT(c) || (! utf8_target && ! is_utf8_pat)) {
- if (utf8_target && scan + max < loceol) {
+ if (utf8_target && loceol - scan > max) {
/* We didn't adjust <loceol> because is UTF-8, but ok to do so,
* since here, to match at all, 1 char == 1 byte */
loceol = scan + max;
--
1.8.1.5
|
From @tonycozOn Tue May 28 03:06:14 2013, tonyc wrote:
Last night I spent some time trying to figure out if I can run the The glibc supplied with Mageia appears to switch to mmap() for allocation The only other way I can see to do that would be a custom malloc(), but Tony |
From @tonycozOn Tue May 28 03:49:19 2013, tonyc wrote:
I've applied this fix, and a few similar changes to S_regrepeat as I think this is a candidate for maint. Tony |
From @rjbs* Tony Cook via RT <perlbug-followup@perl.org> [2013-06-03T08:32:13]
Would some other committer please have a look at this code and comment on its Thanks. -- |
From @nwc10On Wed, May 29, 2013 at 05:22:05PM -0700, Tony Cook via RT wrote:
I suspect that Perl's malloc can be (ab)used so that it always allocates Nicholas Clark |
From @cpansproutOn Fri Jun 07 08:35:09 2013, perl.p5p@rjbs.manxome.org wrote:
’Fraid I’m not familiar with that code at all. -- Father Chrysostomos |
From @hvdsRicardo Signes <perl.p5p@rjbs.manxome.org> wrote: The change is essentially four variants of this: - if (utf8_target && scan + max < loceol) { I think this is correct only if scan and loceol are guaranteed to be I think the former is intended to be the case, but the docs and comments I wonder whether it would make all this significantly clearer if I don't think the change makes anything worse, in the sense that I don't For the record, in this case I also think one commit would have been Hugo |
From @tonycozOn Sat, Jun 08, 2013 at 08:43:18AM +0100, hv@crypt.org wrote:
The difference isn't required to fit into I32 - C's promotion rules scan and loceol are pointers in the string being matched against -
It's clearer with davem's changes in blead where PL_regeol has become
Neither version has a problem with strings larger than 2GB, but
Please propose an example.
I mostly wanted to make the fixes match the tests added (for the two Tony |
From @iabynOn Mon, Jun 03, 2013 at 05:32:13AM -0700, Tony Cook via RT wrote:
Now cherry-picked into maint-5.18 as -- |
|
@iabyn - Status changed from 'open' to 'resolved' |
Migrated from rt.perl.org#118175 (status was 'resolved')
Searchable as RT118175$
The text was updated successfully, but these errors were encountered: