Re: Reexecuting do_fetch after do_unpack failed with reference is not a tree

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

On Wed, 2013-04-24 at 10:46 +0200, Martin Jansa wrote:
> Hi,
> 
> I guess this will be hard to fix correctly, but please comment if I've
> overlooked something:
> 
> 1) update foo git repository with new commit ABCD, but forget to push it
> 2) bump SRCREV in foo.bb to ABCD
> 3) run bitbake foo, it will run do_fetch to catch new changes in remote
>    repo, but ABCD is not there, so do_unpack will fail like this:
> 
>    NOTE: Running task 2289 of 4227 (ID: 1185, foo.bb, do_unpack)
>    NOTE: recipe foo-1.0+gitr2+ABCD: task do_unpack: Started
>    ERROR: Function failed: Fetcher failure: Fetch command failed with exit code 128, output:
>    fatal: reference is not a tree: ABCD
> 
> 4) realize missing "git push" in step 1) and push it now
> 5) run bitbake foo again, but do_unpack is still failing, because
>    do_fetch checksums is the same (nothing changed in metadata) so stamp
>    is still valid.
> 
> 6) fix it with "bitbake -c fetch -f foo", but that could be complicated
>    if bump SRCREV was pushed already and possibly many builders already
>    executed do_fetch, so maybe push PR bump or something like that.
> 
> Only better solution than 6) I can think of, is to detect 
> "fatal: reference is not a tree" in do_unpack and if it's git repository
> then remove do_fetch stamp and hope it was fixed in remote git repo.

I'd say do_fetch should detect that the specified revision isn't present
and fail there and then in do_fetch.

Cheers,

Richard