[FFmpeg-devel] [RFC] full paths in #include directives

[Diego Biurrun]

On Thu, May 08, 2008 at 09:35:21AM +0100, M?ns Rullg?rd wrote:
> Diego Biurrun <diego at biurrun.de> writes:
> 
> > On Thu, May 08, 2008 at 01:36:11AM +0100, M?ns Rullg?rd wrote:
> >> Diego Biurrun <diego at biurrun.de> writes:
> >> 
> >> > I would like to propose using full paths in all #include directives in
> >> > FFmpeg, i.e.
> >> >
> >> > #include "libavutil/common.h"
> >> >
> >> > instead of
> >> >
> >> > #include "common.h"
> >> >
> >> > The reasons are manyfold:
> >> >
> >> > 0) We already expect this from external users.
> >> >
> >> > 1) Many header names are not at all unique.  The above-mentioned
> >> > common.h comes to mind, as do others.  Depending on the order of -I
> >> > flags the right header may or may not be included.
> >> >
> >> > This has bitten me several times in MPlayer already.  We now use full
> >> > paths for FFmpeg #includes in MPlayer, but still -I flags are needed so
> >> > that the FFmpeg headers themselves can find their dependencies.  But
> >> > with -Ilibavcodec on the gcc command line '#include <png.h>' picks up
> >> > the FFmpeg version instead of the libpng header.  Build system kludges
> >> > can work around this, but the system remains brittle, future FFmpeg
> >> > changes may necessitate more or updated kludges.
> >> >
> >> > 2) The need for -I flags would be reduced internally as well as
> >> > externally.
> >> >
> >> > Attached is an example patch for libpostproc.  If the idea is accepted,
> >> > I will do the rest of the work myself.
> >> 
> >> I say go ahead.  Headers in the same directory as the file containing
> >> the #include directive probably don't need the full paths.  The C99
> >> standard leaves it implementation-defined where header files are
> >> searched for.  POSIX does, however, specify that the directory of the
> >> source file be searched first when the #include directive uses
> >> double-quotes.
> >
> > So it may be slightly less portable to use full paths everywhere?
> 
> I think you mean the opposite.

Yes, of course.

> > I say we go with full paths everywhere then.
> 
> I wouldn't bother.  I've never heard of a compiler that didn't search
> the source directory first.

I can't help but think that if we don't use full paths for everything we
will quickly have a mix of both systems again.  Thus, I prefer to be
consistent.

Diego