X-Spam-Check-By: sourceware.org
Message-ID: <4473DF11.6080603@kleckner.net>
Date: Tue, 23 May 2006 21:20:33 -0700
From: Jim Kleckner <jek_subs1 AT kleckner DOT net>
User-Agent: Thunderbird 1.5 (Windows/20051230)
MIME-Version: 1.0
To: cygwin AT cygwin DOT com
Subject: Re: 1.15.19 dlopen() dies with no dlerror()
References: <Pine DOT GSO DOT 4 DOT 58 DOT 0604111406110 DOT 24175 AT sue> <4472AF3B DOT 2050300 AT kleckner DOT net> <4473B387 DOT 2020905 AT kleckner DOT net> <4473C5C9 DOT 50104 AT cygwin DOT com> <4473CF07 DOT 9090201 AT kleckner DOT net> <4473D20E DOT 5090507 AT cygwin DOT com>
In-Reply-To: <4473D20E.5090507@cygwin.com>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit
Mailing-List: contact cygwin-help AT cygwin DOT com; run by ezmlm
List-Subscribe: <mailto:cygwin-subscribe AT cygwin DOT com>
List-Archive: <http://sourceware.org/ml/cygwin/>
List-Post: <mailto:cygwin AT cygwin DOT com>
List-Help: <mailto:cygwin-help AT cygwin DOT com>, <http://sourceware.org/ml/#faqs>
Sender: cygwin-owner AT cygwin DOT com
Mail-Followup-To: cygwin AT cygwin DOT com
Delivered-To: mailing list cygwin AT cygwin DOT com



Larry Hall (Cygwin) wrote:
> Jim Kleckner wrote:
>>
>>
>> Larry Hall (Cygwin) wrote:
>>> Jim Kleckner wrote:
>>>> Jim Kleckner wrote:
>>>>> Michael McKerns wrote:
>>>>>> Yes, yes...  I've not given you enough information...
>>>>>> ...
>>>>>> See:
>>>>>> http://cygwin.com/cygwin-ug-net/dll.html
>>>>>> http://cygwin.com/faq.html#faq.programming.dll-relocatable
>>>>>>   
>>>>> I'm seeing a similar problem with python and 1.5.19 and also tried 
>>>>> the snapshot of 22-May.
>>>>>
>>>>> CYGWIN_NT-5.1 kleckner2 1.5.20s(0.155/4/2) 20060522 00:51:23 i686 
>>>>> Cygwin
>>>>>
>>>>> A simple test case doesn't fail in dlopen().
>>>>>
>>>>> My code is not simple but has been working prior to the most 
>>>>> recent update (which also updated python and other packages).
>>>>> A downrev of python does not make the problem go away.  If I 
>>>>> downrev cygwin, I get complaints about missing entry points.
>>>>>
>>>>> What do you recommend as the best way to isolate this?
>>>>
>>>> I tried using "prev" with setup.exe but that didn't make the 
>>>> problem go away.
>>>>
>>>> A simple test case with python access to a trivial function works 
>>>> fine (can supply if anyone wants).
>>>> The complex dll that used to work simply doesn't return from dlopen.
>>>>
>>>> I downloaded the 20060522 snapshot with debug symbols to get a 
>>>> backtrace with GDB.
>>>> GDB says there is a seg fault and somehow this is preventing any 
>>>> information from reaching dlerror().
>>>> Without the dlerror() info, it is hard to figure out what needs to 
>>>> change with the dll.
>>>> It appears that some constructors are having trouble.
>>>>
>>>> Let me know if there is some single stepping that could be helpful.
>>>> [snip]
>>>> (gdb) bt
>>>> #0  0x610b1ff8 in pthread_key_create (key=0x6622f8, destructor=0) at 
>>>                     ^^^^^^^^^^^^^^^^^^^
>>> Known issue already fixed in the Cygwin snapshot and in GDB's CVS.  
>>> This
>>> is not fatal.  Just continue until you stop seeing this complaint.
>>>
>>
>> As noted above, this was tested using snapshot 20060522. Should that 
>> snapshot have the fix you mention?  If it should, then this problem 
>> still exists in that snapshot.
>> If not, then which one should I test?
>
> The part of the fix that is Cygwin-specific is in the Cygwin snapshot you
> have.  But, like I said, there's another part of the fix that's only in
> GDB's CVS version right now.  If you want to be rid of the problem 
> right now,
> you need both changes and that means you'll need to grab GDB's source 
> from
> CVS and build it.  But whether you choose to do this or not should not
> inhibit your original investigation.  Depending on how many times your
> code path takes you through pthread_ket_create(), it may test test your
> tolerance level for the current work-around though. ;-)
Thanks for pointing me into the GDB and SIGSEGV discussions.
I didn't see the relationship to the dlopen() problem.

I didn't see discussion of a fix to python which has failing
dlopen() calls presumably because of initializations of mutex objects.
Does python need to do what GDB now does?

Is there a workaround/snapshot in the meantime?


--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Problem reports:       http://cygwin.com/problems.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/