[FFmpeg-devel] Longevity Test Report

[Benjamin Larsson]

Michael Niedermayer wrote:
> On Wed, Feb 04, 2009 at 01:08:23PM +0100, Reimar D?ffinger wrote:
>> On Wed, Feb 04, 2009 at 12:52:57PM +0100, Michael Niedermayer wrote:
>>> On Wed, Feb 04, 2009 at 12:13:24PM +0100, Reimar D?ffinger wrote:
>>>> On Wed, Feb 04, 2009 at 12:00:53PM +0100, Benjamin Larsson wrote:
>>>>> I know at least one ticket where this kind of tool would have helped 
>>>>> alot. Or we could use memwatch.
>>>> Neither your suggestions nor memwatch allow to get a backtrace AFAICT.
>>> what backtrace do you want that you dont get?
>>> valgrind can provide pretty backtraces when you compile without
>>> -fomit-frame-pointer and possibly -g<large number>
>> Uh, his suggestions were av_leak_malloc(size,__FILE__,__LINE) and
>> memwatch, valgrind was mine. I was arguing that the suggested solutions
>> are worse than the existing once, including gdb.
> 
> sorry i was tired and mixed things up ...
> 
> 
>> And I have seen more cases where removing -fomit-frame-pointer hindered
>> debugging (due to hiding bugs) than where it helped, so I consider that really
>> bad advice. Inlining is far more of a issue for debugging IMO.
>> -g3 is advisable though.
> 
> if you need a full backtrace, and your tool doesnt provide it without
> -fomit-frame-pointer, adding -fomit-frame-pointer is a solution, finding
> a better tool is another of course.
> 

Ok, I've implemented what I meant, now I'm just wondering if it would be
accepted into svn. Yes there are more advance tools that do this, but it
might not be available on the platform where the problem occurs. I
remember some leak in the ts demuxer that only occured on win32 using
the libs. With this tool it should have been quite easy to find it. But
now instead it took a while before it was found. That's pretty much my
only argument.

MvH
Benjamin Larsson