This is the mail archive of the
cygwin-developers
mailing list for the Cygwin project.
Re: [Cygwin64] dash segfault
On 2013-03-10 21:41, Corinna Vinschen wrote:
> On Mar 10 21:10, Peter Rosin wrote:
>> On 2013-03-10 20:38, Peter Rosin wrote:
>>> On 2013-03-10 20:20, Peter Rosin wrote:
>>>> Ok, here's a crash with the dll from release, still with home-built dash
>>>> w/o -O2.
>>>
>>> Below is another one, looking more similar to the ones I got with the
>>> dll from the install area...
>>>
>>> Every time that it has happened (or, I should say, every time I have
>>> checked), it has been a libtool script linking a dll (either as a
>>> library or as a module), that has crashed dash.
>>
>> I of course only needed to whisper that for it to happen elsewhere, this
>> time when running my not-overly-complicated build script. Still dash though.
>>
>> And now again in libtool --mode=install (not relinking).
>>
>> I have also seen a couple of instances of dash simply exiting w/o triggering
>> error_start (sadly wasn't sane enough to harvest an exit-code).
>>
>> Let me know if you want more backtraces, I get the feeling they're pretty
>> useless? I'd also appreciate further debugging tips.
>
> I don't know. I've tried myself but hadn't much time and fun to debug
> this more closely today. The most interesting snippet I got was another
> stackdump with my local non-optimized Cygwin DLL which again pointed to
> ptmalloc3.cc, line 792, so there seems to be some free on an invalid
> address. The rest of the information I could gather so far was not very
> helpful either. I have this on my plate for tomorrow and most of next
> week, but I would naturally appreciate if others would help debugging
> this, too. It seems a rather tricky one. If worst comes to the worst,
> I rip out ptmalloc3, and we can try again with the old malloc code.
I ripped out "CONFIG_SHELL=/bin/dash /bin/dash" from in front of my
configure invocations, and replaced /bin/sh with bash. I have now built
the project twice in a row w/o any crash. With dash, there seemed to
be at least one crash in every build attempt.
How sure are we that the dash memory allocations are correct? This
could easily be some use-after-free issue that was "safe" with the
old malloc...
Cheers,
Peter