[Buildroot] [PATCH 3/3] support/download: detect and abort when using a git branch by name

[ricardo.martincoski at gmail.com]

Hello,

On Sun, Aug 12, 2018 at 05:48 PM, Yann E. MORIN wrote:

> On 2018-08-12 22:41 +0200, Thomas Petazzoni spake thusly:
>> > > The other way is by using this series:
>> > > http://patchwork.ozlabs.org/project/buildroot/list/?series=44009
>> > > current master:
>> > > https://gitlab.com/RicardoMartincoski/buildroot/pipelines/27301317
>> > > current master + this patch:
>> > > https://gitlab.com/RicardoMartincoski/buildroot/pipelines/27301358
>> > > It doesn't get to show that special-ref is broken because the tests run
>> > > sequentially (to avoid overpopulate the gitlab CI), but it shows (in the
>> > > -build.log) that also the download of a sha1 tip of a branch (search for
>> > > "git-sha1-branch-head" in the log) would be broken with this patch.
>> > > This can be reproduced locally:
>> > > $ make defconfig
>> > > $ ./utils/config --set-str BR2_BACKUP_SITE ""
>> > > $ BR2_DL_DIR=$(mktemp -d) make tremor-dirclean tremor-source
>> > > ...
>> > > Commit '7c30a66346199f3f09017a09567c6c8a3a0eedc8' is a branch name.
>> > > Using a branch name is not supported.
>> 
>> Yann: this one I don't understand.
>> 7c30a66346199f3f09017a09567c6c8a3a0eedc8 is a regular commit SHA1, so
>> it should not be considered as a branch. Why do we have this failure
>> for a commit SHA1 ?
> 
> That's because of our special refs support, for which we do [0]:
> 
>     git fetch origin "${cset}:${cset}"

But only removing this code will not make 'git show-ref' to work as expected.
There are git caches out there with the mentioned branch.
If for any reason the tarball is not there (like we do test in autobuilder) the
download will fail for the wrong reason, and it would require a manual fix.

And IMO we should not start removing all refs from the git caches otherwise we
can decrease the cache hit ratio (due to git gc), specially for linux trees.

So the best approach IMO is to ask "given we want a named ref, is that ref a
branch in the remote?". That obviously lead to using git ls-remote and check
the second column:
$ git ls-remote https://git.xiph.org/tremor.git
7c30a66346199f3f09017a09567c6c8a3a0eedc8        HEAD
293fd1c04f9d4489be6d4b2b1ca8698f2f902e8e        refs/heads/lowmem
7c30a66346199f3f09017a09567c6c8a3a0eedc8        refs/heads/master
f946f9acbf7aced75d3810a52c878e47680a18f6        refs/heads/nobyte
3175ff964c7d85d070df823189997f6b753cfef7        refs/heads/tremolo
0bcded806a0327c86d9246703724b45037d1bbaa        refs/heads/tremolo_mips

And the corner case is when there is a branch and a tag with the same name.
We must choose whether that case fails or the tag is used. I would use the tag.

Regards,
Ricardo