[FFmpeg-devel] [PATCH] configure: add check for pkg-config return value

[Martin Storsjö]

On Mon, 28 Feb 2011, M?ns Rullg?rd wrote:

> Stefano Sabatini <stefano.sabatini-lala at poste.it> writes:
> 
> > On date Thursday 2011-02-24 22:36:27 +0000, M?ns Rullg?rd encoded:
> >> Stefano Sabatini <stefano.sabatini-lala at poste.it> writes:
> >> 
> >> > Add check_pkgconfig_flags() function. This ensures that the result of
> >> > pkg-config is correct. Avoid spurious success during configuration,
> >> > which may lead to compilation errors.
> >> >
> >> > In particular prevents the compilation error I was getting with
> >> > --enable-librtmp --enable-libopencv, which was due to a missing
> >> > libssl.
> >> > ---
> >> >  configure |   15 +++++++++++----
> >> >  1 files changed, 11 insertions(+), 4 deletions(-)
> >> 
> >> This is more convoluted than it needs to be.  I'd also like to know
> >> exactly why the link test didn't fail for you before adding workarounds.
> >
> > Simplified and updated.
> >
> > Note: I'm not sure /dev/null is portable (what about MinGW/CygWin?).
> >
> > I already explained on irc the reason of the failure, it was due to a
> > pkg-config limitation (pkg-config flags only reflect the condition of
> > the system when a package was installed), but an additional check
> > can't hurt too much and it saves the time wasted on debugging the
> > consequential misconfiguration/compilation problems.
> 
> No, you never explained how the same link command can succeed in
> configure and fail during make.

If I understood the situation corretly, there never was such a case. Only 
a case where missing dependency libs were brought in by another library 
transitively. When linking to this dependency was disabled, the other 
linking dependency broke. So he's not trying to fix any concrete breakage, 
only wanting to notice breakage more clearly early on.

Or that's how I've percieved the discussion at least.

// Martin