diff options
author | 2021-04-12 22:29:23 -0700 | |
---|---|---|
committer | 2021-04-18 11:29:05 +0100 | |
commit | 89d675efd633b495daa4a3a57420b9c309497035 (patch) | |
tree | 7b03c8eded0c2d731211258b5e1cd879599750d0 /meta/recipes-devtools/libtool/libtool-2.4.6.inc | |
parent | 4479f810c1a3ab2badf4f9610c309bc0e23e2a5f (diff) | |
download | openembedded-core-contrib-89d675efd633b495daa4a3a57420b9c309497035.tar.gz |
kmod: do not symlink config.guess/config.sub during autoreconf
I was encountering the following race condition on poky:
- automake-native does do_install.
- automake-native does do_populate_sysroot. This hardlinks config.guess
and config.sub into ${D}.
- kmod-native does do_configure. This runs `autoreconf`, which runs
`automake --add-missing` (symlinks config.guess/config.sub from
recipe-sysroot-native to build dir), then runs `gnu-configize` (copies
_its own_ config.guess/config.sub _on top_ of the already existing
ones). Since the destinations already had symlinks, the copy would
overwrite config.guess/config.sub in recipe-sysroot-native, which
would in turn overwrite the same in ${D} due to being hardlinked.
- automake-native does do_package. The outhash is thus calculated on the
clobbered config.guess/config.sub files.
With hash equivalency enabled, the different outhash produced a
different unihash, which kept me from reusing sstate between my laptop
and my build server. This race condition would happen only on the build
server (BB_NUMBER_THREADS = 32) but never on my laptop
(BB_NUMBER_THREADS = 6).
I didn't see the --install and --symlink flags being used by any other
recipe, so I removed them, and that fixed the issue.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Diffstat (limited to 'meta/recipes-devtools/libtool/libtool-2.4.6.inc')
0 files changed, 0 insertions, 0 deletions