[Bug libstdc++/21554] New: [4.0 Regression] ext/array_allocator/2.cc execution test fails on hppa64-hpux

[Bug libstdc++/21554] New: [4.0 Regression] ext/array_allocator/2.cc execution test fails on hppa64-hpux

[jsm28 at gcc dot gnu dot org]

The failure
FAIL: ext/array_allocator/2.cc execution test
appeared on 4.0 branch on hppa64-hpux with the patch
http://gcc.gnu.org/ml/gcc-cvs/2005-05/msg00433.html

2005-05-09  Benjamin Kosnik  <bkoz@redhat.com>

        * docs/html/test.html: Update.
        * testsuite/printnow.c: Remove.
        * scripts/check_survey.in: Remove.

        * testsuite/abi_check.cc: To...
        * testuite/testsuite_abi_check.cc: ...here.
        * testsuite/testsuite_abi.h (compare_symbols): Return int.
        * testsuite/testsuite_abi.cc: Same.
        
        * testsuite/testsuite_hooks.h: Move character related bits to...
        * testsuite/testsuite_character.h: ...here.
        * testsuite/testsuite_character.cc: ... and here.
        [...]

-- 
           Summary: [4.0 Regression] ext/array_allocator/2.cc execution test
                    fails on hppa64-hpux
           Product: gcc
           Version: 4.0.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: libstdc++
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: jsm28 at gcc dot gnu dot org
                CC: bkoz at gcc dot gnu dot org,gcc-bugs at gcc dot gnu dot
                    org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21554

[Bug libstdc++/21554] [4.0 Regression] ext/array_allocator/2.cc execution test fails on hppa64-hpux

[jsm28 at gcc dot gnu dot org]

------- Additional Comments From jsm28 at gcc dot gnu dot org  2005-05-13 19:27 -------
What was the purpose of applying this patch to 4.0 branch?  Did it fix a
regression?  I can't find this patch (looking for the ChangeLog entry test) in
my gcc-patches folders.

As the patch changes the testsuite harness it changes what each test tests so
the appearance of a FAIL might or might not strictly be a regression (depending
on whether the same test and harness combination fail or pass with older
compiler versions) but it seems undesirable on a release branch.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21554

[Bug libstdc++/21554] [4.0 Regression] ext/array_allocator/2.cc execution test fails on hppa64-hpux

[pcarlini at suse dot de]

------- Additional Comments From pcarlini at suse dot de  2005-05-13 20:16 -------
Given our current basic_string implementation (which I'm maintaining) I can
tell you that this testcase is very brittle: see 19495, that lead to the
xfail for powerpc, for additional details. Benjamin's changes cannot have
directly caused the fail and, IMHO, represent a very welcome clean-up.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21554

[Bug libstdc++/21554] [4.0 Regression] ext/array_allocator/2.cc execution test fails on hppa64-hpux

[joseph at codesourcery dot com]

------- Additional Comments From joseph at codesourcery dot com  2005-05-13 20:39 -------
Subject: Re:  [4.0 Regression] ext/array_allocator/2.cc
 execution test fails on hppa64-hpux

On Fri, 13 May 2005, pcarlini at suse dot de wrote:

> Given our current basic_string implementation (which I'm maintaining) I can
> tell you that this testcase is very brittle: see 19495, that lead to the
> xfail for powerpc, for additional details. Benjamin's changes cannot have
> directly caused the fail and, IMHO, represent a very welcome clean-up.

So this should be treated as a known latent bug, a testcase which fails or 
passes at random on targets requiring strict alignment?  In which case 
there should be an effective-target keyword for strict alignment and the 
testcase should be xfailed on all such targets?



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21554

[Bug libstdc++/21554] [4.0 Regression] ext/array_allocator/2.cc execution test fails on hppa64-hpux

[pcarlini at suse dot de]

------- Additional Comments From pcarlini at suse dot de  2005-05-13 21:13 -------
> So this should be treated as a known latent bug, a testcase which fails or 
> passes at random on targets requiring strict alignment?  In which case 
> there should be an effective-target keyword for strict alignment and the 
> testcase should be xfailed on all such targets?

Yes, I think the testcase should be simply xfaild on targets requiring strict 
alignment, that is, requiring alignment strictly > alignment(char) for memory
involved in atomic operations. Currently (*) basic_string doesn't rebind the
passed allocator to one sufficiently aligned for atomic memory, sadly.

(*) Since we don't want to break the library ABI anytime soon, probably I
will try again to fix this long standing bug within v6 (while also keep on
working on the next basic_string in v7).

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21554

[Bug libstdc++/21554] [4.0 Regression] ext/array_allocator/2.cc execution fails

[ebotcazou at gcc dot gnu dot org]

------- Additional Comments From ebotcazou at gcc dot gnu dot org  2005-06-01 08:33 -------
Confirmed on SPARC/Solaris 64-bit.


-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |ebotcazou at gcc dot gnu dot
                   |                            |org
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
 GCC target triplet|                            |hppa64-hpux, sparc64-solaris
   Last reconfirmed|0000-00-00 00:00:00         |2005-06-01 08:33:42
               date|                            |
            Summary|[4.0 Regression]            |[4.0 Regression]
                   |ext/array_allocator/2.cc    |ext/array_allocator/2.cc
                   |execution test fails on     |execution fails
                   |hppa64-hpux                 |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21554

[Bug libstdc++/21554] [4.0 Regression] ext/array_allocator/2.cc execution fails

[pcarlini at suse dot de]

------- Additional Comments From pcarlini at suse dot de  2005-06-01 08:43 -------
Eric, if you are willing to prepare a patchlet xfailing the testcase on
"strict-alignment" targets, whatever that means (*), I would be grateful.

(*) Actually, in case of hppa64 the alignment is insufficient for the
size_t length and capacity fields. In other cases (powerpc) can be the
atomic memory for the reference-count. In any case, I don't think we
can fix this problem without changing basic_string in a binary
incompatible way vs other so.6 releases, unfortunately. Certainly, not
for 4.0.x.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21554