[Zlib-devel] HEADS UP: Apparent bad compilation under (just released) GCC 4.5.0

[Török Edwin]

On 05/04/2010 08:47 PM, Jesus Cea wrote:
> Now I have stranges crashes when using zlib 1.2.5, python and psyco (a
> kind of JIT for python). I have a crash everytime I do an import of any
> zipped library. 100% reproductible.
> 
> The problem is a memory corruption. Trying under GDB, the stack is very
> damaged.
> 
> The problem doesn't exist if I disable compiler optimizations when
> compiling zlib.
> 
> After days messing with GDB with little luck, and knowing that the
> problem seems to be related to optimization, I determine that the issue
> is in "inftrees.c". If I compile everything optimized EXCEPT this
> particular file, the crash vanishes.

Which optimization level breaks it? Does it work with -O2? -O1?
You can try to find out which individual optimization flag breaks it, by
decomposing -O2 into individual optimization flags, and using a binary
search to find out which flag is the culprit.

If this happens at -O3 only, my usual suspect is -ftree-vectorize.

For example you can decompose -O3 into individual flags by running this
command:
gcc-4.5 -O3 -Q --help=optimizers|grep enabled|cut -f1 -d\[|xargs echo

> 
> I compile zlib 1.2.3 and it crashes too!. zlib 1.2.3 is the version
> currently running in my system, and it works... compiled with old
> compiler. Compiling with GCC 4.5.0, it crashes.
> 
> I confirm this compiling zlib 1.2.5 under GCC 4.4.3, using other
> machine. This version works correctly.
> 
> So, GCC 4.5.0 seems to be doing something wrong.
> 
> The only other option is that psyco is messing things. Could be
> possible, since it generates x86 code directly (like a jit), so maybe it
> is messing the environment. I am using version 1.6, released in 2007, I
> think. Stable since.

Does the bug occur if you try to compress/uncompress the file using
minigzip?
Or can you find another way to reproduce the crash with python only
(without psyco)?

> 
> I am still trying to determine the exact scope. But even if the fault is
> not in zlib, I think it is useful you are aware of this particular
> issue, just in case you start getting other negative feedback, now that
> GCC 4.5.0 is out (two weeks ago).

Can you take a stacktrace, or disassemble the code around the crash (if
the stacktrace doesn't show useful line numbers)?

> 
> If you have any suggestion of how to debug this, please, let me know. I
> have experience with ASM, but not under x86 (could try).
> 
> I am under Solaris 10, x86, compiling zlib as 32 bits.

I was going to suggest running under valgrind, but I doubt valgrind
supports Solaris.

> If you think this messsage is inadecuate for zlib-devel, let me know.
>

I don't know, gcc-help at gcc.gnu.org would be a good place to seek help on
this too. I am not subscribed to that ML however.

Best regards,
--Edwin