[oe] Missing includes in STAGING_INCDIR

[Michael 'Mickey' Lauer]

On Tuesday 04 March 2008 13:20:04 Tom Cooksey wrote:
> > > I am writing a recipie for Qt/Embedded 4.4 beta (using qt4 recipies as
> > > a base). At the moment Qt's configure test fails on dbus. I have added
> > > -I${STAGING_INCDIR}/dbus-1.0 to the configure flags so the configure
> > > test now picks up dbus/dbus.h correctly.
> > >
> > > However, dbus.h itself includes dbus/dbus-arch-deps.h, which is not in
> > > ${STAGING_INCDIR}/dbus-1.0/*, but _is_ in the dbus work dir. This means
> > > I have to add the dbus work dir to the include list, which feels wrong.
> >
> > Surely that's wrong. Actually dbus-arch-deps.h is where it belongs to,
> > ${libdir}/dbus-1.0/include/. On my favourite system this is:
> >
> > tmp/staging/arm-angstrom-linux-gnueabi/lib/dbus-1.0/include/dbus/dbus-arc
> >h-deps.h
>
> Ok, yes, I have this too. Why the two different include paths? Surely
> ${STAGING_INCDIR}/dbus-1.0/ is useless without the other?

Don't ask me, this is what upstream decided to do.

> > > I have had the same problem with gstreamer too. I think something has
> > > broken recently, as I've tried building the existing qtopia core
> > > recipies and had the same failures.
> >
> > Just rely on pkgconfig and it will do the job. The relevant excerpt for
> > dbus is:
>
> Ah... we disable pkgconfig when cross-compiling. ;-) It usually uses the
> host's includes and libraries due to most toolchains shipping with broken
> .pc files. We thought the compiler spitting out "Can't find ..." is better
> than "... binary is incompatable". It looks like OE runs .pc files through
> sed, is this to fix the same issue?

Exactly. We think pkgconfig has a lot of value and rather attempted to fix it 
(by sed'ing the .pc files) than to circumvent it. I advise you to do the 
same.

> I'll add ${libdir}/dbus-1.0/include to the configure line. Is libdir or
> STAGING_LIBDIR the best to use?

This would be ${STAGING_LIBDIR}/dbus-1.0/include.

> Ideally this would go into the mkspec file
> for the cross-compiler. Otherwise, it will be passed to the host compiler
> too (Not that the host compiler's used much with qmake & friends built
> seperatly).
>
> I might look into changing the current "patch the mkspecs/common/*" way of
> doing things to actually installing a proper mkspec for the OE cross
> compiler. Would you have any objection to this?

Not at all -- if it's done in a sane way that doesn't break our existing 
applications using qmake.

> Also, as cross-compiling is supported on Qt/Embedded, I'm doing things "the
> qt way" rather than using the cross-compile patch the other qt4 packages
> use. Don't want to patch the source if I don't need to.

Sounds good. We disabled Qt/E thinking it cross-compiles because the configure 
script was doing more harm than good in that case. If you can fix it, we are 
happy.

:M:

-- 
Dr. Michael 'Mickey' Lauer | IT-Freelancer | http://www.vanille-media.de