[FFmpeg-devel] [PATCH] configure: fix module dependencies on zlib

[James Almer]

On 11/20/2017 4:25 PM, Michael Niedermayer wrote:
> On Mon, Nov 20, 2017 at 02:58:58PM -0300, James Almer wrote:
>> On 11/20/2017 2:34 PM, Michael Niedermayer wrote:
>>> On Sun, Nov 19, 2017 at 04:00:11PM -0300, James Almer wrote:
>>>> select should not be used with external libraries. It's mean to soft
>>>> enable internal modules/features.
>>>>
>>>> Signed-off-by: James Almer <jamrial at gmail.com>
>>>> ---
>>>>  configure | 48 ++++++++++++++++++++++++++----------------------
>>>>  1 file changed, 26 insertions(+), 22 deletions(-)
>>>
>>> btw, is all the *suggest= zlib stuff correct ?
>>>
>>> at least in the past configure could not handle a mix of
>>> dependancies which turned targets on and which turned sources off
>>> as the order in which they where executed resilted in different
>>> outcomes
>>
>> 'foo_suggest="bar"' nowadays is basically the same as doing 'enabled foo
>> && enable_weak bar && enabled bar && append foo_extralibs
>> $bar_extralibs'. It does not disable foo under any circumstances.
>>
>> enable_weak can't enable what was hard disabled, and zlib being
>> autodetected will be either hard enabled or hard disabled by the
>> automated check or by a command line option by the time dependency
>> checks are made.
> 
> IIRC the issue i remember was:
> 
> 'foo_suggest="bar"' or 'foo_select="bar"'
> AND
> 'dot_deps="bar"'
> 
> had in the past unpredictable behavior because if foo_suggest/select is
> evaluated first dot will be enabled and if dot_deps is evaluated first
> dot would stay disabled.

That sounds like it would only be possible with _select, and I'm not
even sure if even then at all.
_deps of course doesn't try to enable anything, and _suggest can't
enable what was hard disabled, and external libraries (autodetected or
not) get hard disabled long before dependencies are resolved.

As long as bar was hard disabled by the time dependencies are resolved,
it shouldn't matter which module gets checked first in your scenario, if
dot or foo.

> 
> That is the dependency solver would produce different results when
> one would not expect it. On top of that the dependency solver recursivly
> evaluted things making it a tad bit unpredictable what would get
> evaluated first.
> 
> I remember thinking back then that all dependancies should be clear
> one way, eiter only deps or only selects/suggest. (which was the case
> back then for nearly all things as i rememer checking)
> 
> But its a while ago that i looked at this so i might misremember a detail
> and there where various changes since then in configure.

My bet is that yeah, so many changes have been made to configure that by
now such scenarios are probably not a problem anymore.

> 
> but after a bit of looking now, i found my grep command i used to check
> for this issue:
> 
> egrep '(suggest|select)=' configure | sed 's/.*=//;s/"//g' | tr ' ' '\n' | sort | uniq > select.tmp
> egrep '(deps|deps_any)=' configure | sed 's/.*=//;s/"//g' | tr ' ' '\n' | sort | uniq >dep.tmp
> cat select.tmp dep.tmp |sort | uniq -d
> 
> [...]
> 
> 
> 
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>