[MPlayer-dev-eng] [PATCH] directfb -> pkg-config

[Rich Felker]

On Tue, Feb 21, 2006 at 12:56:39AM +0100, Aurelien Jacobs wrote:
> On Mon, 20 Feb 2006 18:27:13 -0500
> Rich Felker <dalias at aerifal.cx> wrote:
> 
> > On Mon, Feb 20, 2006 at 08:55:02PM +0100, Enrico Weigelt wrote:
> > > * Ivan Kalvachev <ikalvachev at gmail.com> schrieb:
> > > 
> > > <snip>
> > > 
> > > > The only problem if we try to detect library when pkg-config fails
> > > > is that the configure would be a little bit bigger and slower.
> > > > I don't see this as problem as long as there are cases where
> > > > pkg-config could fail and detection would work.
> > > 
> > > In other words, you suggest an multi-step detection. This is okay,
> > > if it can be easily disabled, ie. "--disable-autodetects" or maybe 
> > > "--pkg-config-only" prevents running the builtin tests and only
> > > relying on pkg-config. This is sometimes necessary (ie. in sysroot
> > > environments) to prevent false-positives. 
> > > 
> > > For example, imagine we check for directfb by looking for 
> > > /usr/include/directfb.h. The host system has directfb installed,
> > > but the (sysroot'ed) build environment has not. You can easily 
> > > see, that we get an false positive, which ends up in an broken 
> > > build. (the compiler won't find direcfb.h, since its outside
> > > of its sysroot).
> > 
> > I'm not sure how sysroot works or what it is, but IMO it's incorrrect
> > to read any headers from /usr/include if you're building in a
> > sandbox/isolated/crosscompiling environment.
> 
> Absolutely agree. But unfortunately, that's what is done by MPlayer's
> (supposedly decent design) configure all over the place !

No it's not. That's what's done by your compiler. You need a properly
configured toolchain to look in just a sandbox for headers/libs/etc.

Rich