On MIPS N32 ABI, LIBDIR is lib32. I noticed that although popt's .so's were installed to /usr/lib32, its pkgconfig .pc file was installed to /usr/lib/pkgconfig. To reproduce: mkdir -p popt-test/install cd popt-test tar -xvzf /usr/portage/distfiles/popt-1.16.tar.gz cd popt-1.16 ./configure --prefix=`pwd`/../install --libdir=`pwd`/../install/usr/lib32/ make make install Now, see that in ../install/usr/lib32 the .so's have been installed but the .pc file is in ../install/lib/pkgconfig/. Strangely enough, the .pc file is correct with information like libdir=/home/mattst88/popt-test/popt-1.16/../install/usr/lib32
Created attachment 257910 [details, diff] patch to fix hard-coded 'lib/pkgconfig' Tested working, according to previously stated test.
just patch the Makefile.in (to avoid timestamp issues). or run sed on it. either way, change looks fine if you wish to commit.
This change should be necessary only for the 'install' step? Then I'd use (to avoid messing with Makefile.* for timestamp/autotools: emake DESTDIR="${D}" pkgconfigdir="/usr/$(get_libdir)/pkgconfig" install || die #349558 and mail your patch to upstream :)
Sent the patch upstream. I like Samuli's fix for now. Should I revbump to 1.16-r1 and drop back to ~arch?
trouble with the install step is that when the prefix guys want to screw with things, we'll probably have to tweak that further. which is why i like the makefile/configure logic. and because what we send upstream is what we actually use in the ebuild.
Committed with vapier's suggestion to just modify Makefile.in. Patch sent upstream: http://marc.info/?t=129317116500001&r=1&w=2 Unclear whether the patch has been committed upstream.