[vlc-devel] [vlc-commits] picture: factor freeing picture_t

[Steve Lhomme]

The issue is this line:
http://git.videolan.org/?p=vlc.git;a=blob;f=src/misc/picture.c;h=bcab74622d2f39f1b5df8c1993adb56ffdbe40b8;hb=HEAD#l406

It free() the clone. Which is then free() again in the caller of 
picture_DestroyClone(), that's
picture_Destroy().

Maybe it works for you if you don't use codecs using clones...

On 14/12/2018 09:03, Steve Lhomme wrote:
> On 13/12/2018 16:50, Rémi Denis-Courmont wrote:
>> Le torstaina 13. joulukuuta 2018, 11.16.35 EET Steve Lhomme a écrit :
>>> This crashes on Windows. The commits after that don't fix it.
>>> It crashes with software and hardware chromas when freeing the decoded
>>> picture.
>> There are only two ways that this changeset can crash: either it's a 
>> trivial-
>> to-fix double free, or it's some latent/preexisting bug.
>
> We'll see when there is a fix. I had a look and it didn't look like a 
> double free. The free() crashes with a memset() inside, probably to 
> write 0xdeadbeef in debug builds. But the memory has proper values 
> before the call.
> Also it's not the first free() that crashes but after a few frames are 
> decoded.
>
> The fix may be trivial but it's not trivial to find.
>
>>
>> Either way, I don't see the need to make a huge deal on vlc-devel.
>
> This is how we signal the author that a commit has an issue. That's 
> why vlc-commit mails have a reply-to going to vlc-devel.
> Examples:
> https://mailman.videolan.org/pipermail/vlc-devel/2018-October/121682.html
> https://mailman.videolan.org/pipermail/vlc-devel/2018-July/120197.html
> https://mailman.videolan.org/pipermail/vlc-devel/2018-July/120196.html
> https://mailman.videolan.org/pipermail/vlc-devel/2018-June/119543.html
> https://mailman.videolan.org/pipermail/vlc-devel/2018-June/119481.html
> https://mailman.videolan.org/pipermail/vlc-devel/2018-June/119467.html
>
>
>>
>>> Can you send your patches on the ML when you have big changes like that
>>> that you are not sure about ?
>> I do. And typically, they get ignored for more than a week, if not 
>> forever.
>
> It's still worth a try.
>
>> Obivously I will not be extending this to patches that I am sure 
>> about, like
>> this one.
>>
>> And *this* is not a big change in terms of either subjective concept or
>> objective size.
>>
>>> We all do.
>> Ahaha, very funny.
>>
>>> I assume this was tested before pushing.
>> Of course. This was manually tested and make-checked with sanitizers.
>>
>> -- 
>> 雷米‧德尼-库尔蒙
>> http://www.remlab.net/
>>
>>
>>
>> _______________________________________________
>> vlc-devel mailing list
>> To unsubscribe or modify your subscription options:
>> https://mailman.videolan.org/listinfo/vlc-devel
>
> _______________________________________________
> vlc-devel mailing list
> To unsubscribe or modify your subscription options:
> https://mailman.videolan.org/listinfo/vlc-devel