I recently found a strange problem involving the gitsm fetcher and our PREMIRRORS and I was able to reproduce it using poky. > git clone git://git.yoctoproject.org/poky poky1 > cd poky1 > . oe-init-build-env > bitbake ovmf The ovmf recipe has git submodules inside git submodules. This step works fine. I then exposed the downloads directory over http (http://<host>/downloads) and the downloads/git2 directory over http with git-http-backend enabled (http://<host>/git). I noticed that there are shallow clone tarballs in downloads that match the git repos in downloads/git2. For example: downloads/git2/boringssl.googlesource.com.boringssl/ downloads/git2_boringssl.googlesource.com.boringssl.tar.gz Since these are duplicates of the git repos, I deleted the tar.gz files > rm -f downloads/git2_* <new shell> > git clone git://git.yoctoproject.org/poky poky2 > cd poky2 > . oe-init-build-env > cat >> conf/local.conf <<EOF WRS_MIRROR_HOST = "<host>" BB_ALLOWED_NETWORKS = "${WRS_MIRROR_HOST}" BB_NO_NETWORK = '0' BB_FETCH_PREMIRRORONLY = '1' PREMIRRORS_append = " \ .*://.*/.* http://${WRS_MIRROR_HOST}/downloads/ \n \ git://.*/.* git://${WRS_MIRROR_HOST}/git/MIRRORNAME;protocol=http \n \ gitsm://.*/.* git://${WRS_MIRROR_HOST}/git/MIRRORNAME;protocol=http \n \ " CONNECTIVITY_CHECK_URIS = "" EOF > bitbake ovmf This fails with: ERROR: ovmf-edk2-stable201905-r0 do_unpack: gitsm: submodule unpack failed: UnpackError Unpack failure for URL: 'gitsm://github.com/openssl/openssl;protocol=https;name=CryptoPkg/Library/OpensslLib/openssl;subpath=CryptoPkg/Library/OpensslLib/openssl;bareclone=1;nobranch=1'. No up to date source found: clone directory not available or not up to date: /ala-lpggp22/kscherer/gitsm/poky2/build/downloads/git2/github.com.openssl.openssl; shallow clone not enabled If I leave the tarballs in the PREMIRROR the build succeeds. The other recipes that use git are fetched properly. The logs don't have any warnings before the error and I verified that the objects are indeed present in the repo. Are my PREMIRROR settings correct? Is this a bug in the gitsm fetcher?
Mark, Do you think you'll work on this in the next few weeks?
It looks like a feature rather than a bug. BB_FETCH_PREMIRRORONLY functionality is kind of broken in the current bitbake (see https://bugzilla.yoctoproject.org/show_bug.cgi?id=13353 https://bugzilla.yoctoproject.org/show_bug.cgi?id=9061 https://bugzilla.yoctoproject.org/show_bug.cgi?id=13233). Trying to run git clone (and ls-remote) from upstream if premirror has been specified is a actual bug while erroring out if BB_FETCH_PREMIRRORONLY has been specified but mirror doesn't contain revision is a proper behaviour to not break reproducible builds.
Fixes for mentioned bugs are merged and behaviour should be consistent now. If BB_FETCH_PREMIRRORONLY = '1' has been specified and revision not in the PREMIRROR the build will fail (with NetworkAccess error). Fetchers will not try upstream for git fetcher anymore. https://git.openembedded.org/bitbake/commit/?id=b47ecab3e3aad5c5c376ec023aa82a51aa0f3b86