git-ubuntu regressions in edge

[Robie Basak]

A few times recently we've had edge end up broken. Each time, the
self-test of the snap failed, but we don't currently have any means of
running the self-test in our CI pipeline. For the time being, I've
turned off automatic publication to edge (in fact to any channel) to
avoid discouraging users from using edge. I intend to push to edge after
testing CI builds manually for now.

Details:

The underlying problem seems to be that even though we get a CI pass
(following a CI snap build and integration test using it), when we push
to master, Launchpad builds the snap again and the result is different.
Launchpad currently provides no mechanism to test the resulting snap
before pushing to edge, and is blocked on snapcraft support to do so[1].

Since edge was being automatically build and uploaded to the snap store
every time git master changed, this caused the regressions to hit edge.

Nish fixed the first regression by specifying stricter ordering in
snapcraft.yaml (AIUI, snapcraft is currently not deterministic in
determining build ordering where it isn't stricly defined).

We now have some second regression that I haven't pinned down yet. I
started by addressing these edge regressions first.

[1] https://forum.snapcraft.io/t/launchpad-post-build-pre-upload-testing/5545?u=rbasak
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <https://lists.ubuntu.com/archives/ubuntu-distributed-devel/attachments/20180604/40ca2f8c/attachment.sig>