[FFmpeg-devel] CONFIG_W64_DEMUXER and dead-code elimination

Reimar Döffinger Reimar.Doeffinger at gmx.de
Wed May 11 21:15:11 CEST 2016


On 11.05.2016, at 17:00, Bruce Dawson <brucedawson-at-google.com at ffmpeg.org> wrote:

> Microsoft is aware of the issue and is working on it. They might fix it.
> They might not, however, because regressions in non-standard behavior are
> not as serious. We will see.
> 
> I'm committing a Chromium specific workaround because that allows us to
> test Microsoft's new optimizer - I want to be able to report bugs and make
> suggestions while it's not too late, and adding w64.c was a simple way to
> unblock this testing.

Wouldn't it be nicer to work around this (and the debug build issue) with a single file that provided dummy symbols for anything disabled?
That would be an option that avoids cluttering the main code, and something that those who need it can maintain on their own without affecting everyone else.
I admit I don't know for sure the end result would be any nicer than converting everything to #if, but it might be...


> 
> Anyway, thought the ffmpeg team should know.
> 
> On Wed, May 11, 2016 at 10:51 AM, Hendrik Leppkes <h.leppkes at gmail.com>
> wrote:
> 
>> On Wed, May 11, 2016 at 4:19 PM, Bruce Dawson
>> <brucedawson-at-google.com at ffmpeg.org> wrote:
>>> The error is:
>>> 
>>>    wavdec.obj : error LNK2001: unresolved external symbol
>> _ff_w64_guid_data
>>> 
>>> This happens because the compiler does not recognize early enough
>>> that _ff_w64_guid_data is never actually needed, so it creates a
>> reference
>>> to it.
>>> 
>> 
>> This sort of construct is used all the time in our code, so I would
>> suggest to also report this to MS as a regression.
>> Its important to keep in mind that they introduced a new optimizer
>> just recently in the pre-release builds, so regressions are bound to
>> happen, and hopefully disappear by the time it is released.
>> 
>> Testing Update 3 right now, and possibly reacting to any changes it
>> might require other then reporting regressions to MS, seems very
>> premature.
>> 
>> - Hendrik
>> _______________________________________________
>> ffmpeg-devel mailing list
>> ffmpeg-devel at ffmpeg.org
>> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>> 
> 
> 
> 
> -- 
> Bruce Dawson
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel


More information about the ffmpeg-devel mailing list