libcrypto core dump in 64bit

Any hints?

/pro/3gl/LINUX/git-1.5.4.rc5/t 154 > sh t5302-pack-index.sh
*   ok 1: setup
*   ok 2: pack-objects with index version 1
*   ok 3: pack-objects with index version 2
*   ok 4: both packs should be identical
*   ok 5: index v1 and index v2 should be different
*   ok 6: index-pack with index version 1
*   ok 7: index-pack with index version 2
*   ok 8: index-pack results should match pack-objects ones
*   ok 9: index v2: force some 64-bit offsets with pack-objects
t5302-pack-index.sh[67]: 3156 Illegal instruction(coredump)
* FAIL 10: index v2: verify a pack with some 64-bit offsets
        git verify-pack -v "test-3-${pack3}.pack"
FATAL: Unexpected exit with code 0

HP gdb 5.7 for HP Itanium (32 or 64 bit) and target HP-UX 11.2x.
Copyright 1986 - 2001 Free Software Foundation, Inc.
Hewlett-Packard Wildebeest 5.7 (based on GDB) is covered by the
GNU General Public License. Type "show copying" to see the conditions to
change it and/or distribute copies. Type "show warranty" for warranty/support.
..
Core was generated by `git'.
Program terminated with signal 4, Illegal instruction.
ILL_REGNAT - Register NaT Consumption
#0  0xc0000000033c8940:0 in sha1_block_asm_host_order+0x22e0 ()
   from /usr/local/ssl/lib/libcrypto.so
(gdb) where
#0  0xc0000000033c8940:0 in sha1_block_asm_host_order+0x22e0 ()
   from /usr/local/ssl/lib/libcrypto.so
#1  0xc0000000033cb680:0 in SHA1_Update () at ../md32_common.h:476
#2  0x40000000000ef560:0 in verify_pack (p=0x6000000000037e40, verbose=1)
    at pack-check.c:168
#3  0x40000000000d5130:0 in cmd_verify_pack (argc=2, argv=<not available>,
    prefix=<NaT>) at builtin-verify-pack.c:43
#4  0x40000000000440e0:0 in handle_internal_command (argc=3,
    argv=0x9fffffffffffee88) at git.c:257
#5  0x40000000000446a0:0 in main (argc=3, argv=0x9fffffffffffee88) at git.c:447
(gdb)

Hi,

On Mon, 11 Feb 2008, H.Merijn Brand wrote:

Yes.

This seems to be an OpenSSL issue, probably in its Itanium-optimised code (since Itanium is not _all_ that common, it is quite likely that no many people exercise this part of the code).

Unfortunately, I am not at all an expert in Itanium's assembler, otherwise I'd try to help...

Sorry, Dscho

On Mon, 11 Feb 2008 22:42:16 +0000 (GMT), Johannes Schindelin <Johannes.Schindelin@gmx.de> wrote:

Now tried with HP C-ANSI-C -O3, -O2, -O1, and -O0 and with GNU gcc 4.2.1 -O3
All give the same failure
 

Hello,

H.Merijn Brand <h.m.brand <at> xs4all.nl> writes:

yes, this is OpenSSL issue. I had to recompile OpenSSL myself to make it work (with no-asm option)

I'm using only HP cc for building git (with all the toolchain). (No gcc here)

# uname -mrs HP-UX B.11.23 ia64 # cc --version cc: HP C/aC++ B3910B A.06.14 [Feb 22 2007]

Michal

On Tue, 12 Feb 2008 15:26:16 +0000 (UTC), Michal Rokos <michal@rokos.cz> wrote:

OK, I did the same now, and the docs tell me to turn down optimization to +O2 when you choose to do so.

Apart from that, when you create shared libs, the default build creates a conflicting situation, so you have to throw away the files that cause dup names: aes-ia64.o rc4-ia64.o bn-ia64.o from libcrypto.a

t5302-pack-index.sh now passes

HP-UX 11.23/64 U rx1620/64 Itanium 2/1600(2) ia64 2037 Mb cc: HP C/aC++ B3910B A.06.15 [May 16 2007]

Did you get the full test suite to pass?

Hello,

On Feb 12, 2008 5:38 PM, H.Merijn Brand <h.m.brand@xs4all.nl> wrote:

No - vanilla git's master won't pass (since HPUX vsnprintf is broken (it returns -1 when n is exceeded).

With strbuf workarounds test suite passes up to t4200-rerere.sh.

I did not try to investigate further or continue with the rest of the tests.

MR

On Wed, 13 Feb 2008 10:04:23 +0100, "Michal Rokos" <michal@rokos.cz> wrote:

diff -purN git-1.5.4/strbuf.c git-1.5.4p/strbuf.c --- git-1.5.4/strbuf.c 2008-02-02 05:09:01 +0100 +++ git-1.5.4p/strbuf.c 2008-02-13 11:26:56 +0100