On Thu, Sep 27, 2007 at 09:18:02PM +1000, Allan McRae wrote:
> Essien Ita Essien wrote:
> > sounds interesting... very. Can you give s snippet from the resuling
> > PKGBUILD? sorry i'm a bit slow with _only_ textual descriptions :)
> >
> >   
> OK - see if this helps.
> 
> Firstly the case where one package is split-up.  For example the gcc 
> PKGBUILD would look something like (with much stealing from current 
> package builds):
> 
> build()
> {
> ....
> make || return 1
> }
> 
> package()
> {
> #main
> make -j1 DESTDIR=${startdir}/pkg/main install || return 1
> rm -f ${startdir}/pkg/main/usr/lib/lib*
> find ${startdir}/pkg/main -name libstdc++.mo -delete
> 
> #lib
> make -j1 DESTDIR=${startdir}/pkg/lib install-target-libstdc++-v3 
> install-target-libmudflap install-target-libgomp install-target-libssp 
> || return 1
> rm -rf ${startdir}/pkg/lib/usr/include
> rm -rf ${startdir}/pkg/lib/usr/lib/gcc
> rm -rf ${startdir}/pkg/lib/usr/bin
> rm -rf ${startdir}/pkg/lib/usr{,share}/man
> find ${startdir}/pkg/lib -name gcc.mo -delete
> 
> #obc
> ...
> }
> 
> Note the gcc package is installed in ${startdir}/pkg/main and the 
> gcc-libs is installed in ${startdir}/pkg/libs.  That is it.... no other 
> tricks!
> 
> 
> Now the slightly more difficult case of supporting different 
> configurations.  I'll use ettercap as an example as I have looked at it 
> recently (see http://bugs.archlinux.org/task/4051 ).  Basically, it can 
> have an optional GTK interface.  So you would end up with something like
> 
> build
> {
> cd $startdir/src/$origname-$origver
> 
> case "$1" in
>   "gtk" )
>     ./configure --prefix=/usr --sysconfdir=/etc --enable-plugins 
> --enable-https
>   ;;
>   * )
>     ./configure --prefix=/usr --sysconfdir=/etc --enable-plugins 
> --enable-https  --disable-gtk
>   ;;
> esac
> 
> make || return 1
> make DESTDIR=$startdir/pkg install
> 
> ...
> }
> 
> This would work with "makepkg" using the default case (*) and making 
> package ettercap.  "makepkg gtk" would make ettercap-gtk.  If we were 
> building foo-mysql and foo-prestgres then the default case would be 
> "return 1" or some other error.
> 
> Note that I have concerns that this way of providing multiple 
> configuration options may make things too complicated with needing case 
> statements for provides/conflicts, configure options, etc...  But then 
> again, it may still be better than providing multiple PKGBUILDS.  I'm 
> still trying to improve this so more suggestions are alway welcome! :)

One problem with both of these methods is that they're programatic.
There's no easy way to inspect the PKGBUILD to see what packages it builds
or see what configurations it can be compiled with without actually running
it and seeing what drops out.

It makes matching up PKGBUILDs to packages even more difficult.

I do like them very much though, one or two little tweaks, but they're
mostly bikeshed things.

Jason

Attachment: pgpq3KI5oMMOU.pgp
Description: PGP signature

_______________________________________________
arch mailing list
arch@archlinux.org
http://archlinux.org/mailman/listinfo/arch

Reply via email to