Re: [PATCH v5 1/2] ltp: Update to 20200120

[Richard Purdie <richard.purdie@linuxfoundation.org>]

On Sat, 2020-02-29 at 13:42 +0000, Richard Purdie wrote:
> On Sat, 2020-02-29 at 13:22 +0000, Richard Purdie wrote:
> > On Sat, 2020-02-29 at 07:46 +0100, Petr Vorel wrote:
> > > At least we know, that once we solve this problem, we can remove
> > > 0002-kernel-controllers-Link-with-libfts-explicitly-on-mu.patch
> > > 
> > > Please, how to reproduce this build locally? Is it:
> > > . oe-init-build-env
> > > TCLIBC=musl bitbake ltp
> > 
> > Yes, that should be enough.
> > 
> > > Or do I need to put something into config files conf/local.conf
> > > or conf/bblayers.conf ?
> > > 
> > > I need to double check if musl-fts is really built and whether
> > > the
> > > order of
> > > -lfts is correct. But all this has been solved even for static
> > > build
> > > for Buildroot,
> > > so I guess it's some issue related to openembedded.
> > > Unfortunately,
> > > I'm not much
> > > experienced in openembedded (stepped in, because nobody else has
> > > updated LTP
> > > package yet).
> > > 
> > > Khem, any idea, what can be wrong?
> > 
> > FWIW I'll run a build here and see if I can spot anything but I'm
> > trying to do other things today!
> 
> I was able to confirm that fts is in the recipe sysroot and that the
> compile step is running "make -j 88 LIBC=musl LIBS=-lfts", i.e. its
> being passed in to ltp. Beyond that I'm a little out my depth to know
> how/why its not ending up where it should unfortunately. At least the
> issue does appear to reproduce though...

Grep suggests LIBS is a configure option, not a makefile one. I tried:

diff --git a/meta/recipes-extended/ltp/ltp_20200120.bb b/meta/recipes-extended/ltp/ltp_20200120.bb
index 847f267b0b7..5be9489a756 100644
--- a/meta/recipes-extended/ltp/ltp_20200120.bb
+++ b/meta/recipes-extended/ltp/ltp_20200120.bb
@@ -16,7 +16,8 @@ LIC_FILES_CHKSUM = "\
 
 DEPENDS = "attr libaio libcap acl openssl zip-native"
 DEPENDS_append_libc-musl = " fts "
-EXTRA_OEMAKE_append_libc-musl = " LIBC=musl LIBS=-lfts "
+EXTRA_OEMAKE_append_libc-musl = " LIBC=musl "
+EXTRA_OECONF_append_libc-musl = " LIBS=-lfts "
 
 # since ltp contains x86-64 assembler which uses the frame-pointer register,
 # set -fomit-frame-pointer x86-64 to handle cases where optimisation

and this does seem to build for musl. Might be worth runtime testing
that on the autobuilder?

Cheers,

Richard