I haven't looked at this, but probably the test could be adjusted to first build the needed parts with only the original cache? Alex On Wed, 8 Dec 2021 at 09:27, Stefan Herbrechtsmeier < stefan.herbrechtsmeier-oss@weidmueller.com> wrote: > Am 08.12.2021 um 09:10 schrieb Alexander Kanavin: > > On Wed, 8 Dec 2021 at 08:45, Stefan Herbrechtsmeier > > > > wrote: > > > > I will update my patch series but need some time to test it locally. > > The > > devtool test build a lot of recipe on every run because it use a > > private > > sstate cache. How does the autobuilder overcome this problem? > > > > > > If you don't rebase the branch used for testing, and set the cache to a > > directory outside of build directory (e.g. /home/stefan/sstate ) it will > > reuse that cache between selftest invocations. > > The devtool test use its own PRIVATE sstate cache: > > def setUpClass(cls): > ... > cls.original_sstate = bb_vars['SSTATE_DIR'] > cls.devtool_sstate = os.path.join(bb_vars['TOPDIR'], 'sstate_devtool') > cls.sstate_conf = 'SSTATE_DIR = "%s"\n' % cls.devtool_sstate > cls.sstate_conf += ('SSTATE_MIRRORS += "file://.* file:///%s/PATH"\n' > % cls.original_sstate) > > And remove it after the test run: > > def tearDownClass(cls): > ... > runCmd('rm -rf %s' % cls.devtool_sstate) > > Even a keep builddir doesn't help to keep the private sstate cache. > > > The autobuilder uses a single writeable cache for all builds and > > builders, mounted as read-write NFS partition. > > This doesn't help in this case. The devtool test clean parts of the > sstate cache and therefore use its private sstate cache and the original > sstate cache as mirror. This is a problem if a test triggers a build of > dependencies. > > Regards > Stefan >