As a temporary stop-gap, explicitly deny the noisiest denials until bug #1849753 is fixed.
There is no chance of regression for this since snap-confine is already
denying them. We could consider making these allowed rather than
explicit deny, but people haven't been complaining about these classic
chromium content api snaps not working, so I'd like to continue denying
for now.
Merge pull request #7770 from bboozzoo/bboozzoo/mock-command-abs-path
testutil, many: make MockCommand() create prefix of absolute paths
Followup on #7755. We have some code that mocks commands at absolute paths. Normally this would require the caller to create the prefix manually. The patches extend MockCommand() to create prefix.
Merge pull request #7727 from stolowski/improve-prereq-test
tests: improve TestDoPrereqRetryWhenBaseInFlight to fix occasional flakiness.
The test could fail very sporadically (e.g. after 2 hours when run in a loop) with the prerequsites task in "Done" state instead of expected "Doing". It's relatively easy to make this test fail fast (e.g. after just a few iterations when run in a loop by modifying the wait times inside the test loop, e.g. from 1 ms to 1s).
This was caused by delicate time dependencies in the test. The main task status-probing loop probably assumed that every Ensure+Wait execution picked exactly one task at a time, but what I found by a bit of printf-debugging in the test and in the taskrunner code was that every iteration executed both tasks - most likely because there is no blocking (as WaitFor) dependency between the tasks; that meant that on the second iteration of the loop, if link-snap was picked first (meaning it finished the job that prereq waits for), then prerequsites would also run and transition to DoneStatus too, and therefore the expectation that Ensure loop ends with link-snap in DoneStatus and prerequisites in DoingStatus would not hold true.
The fix here moves the check for prerequisites being in "DoingStatus" into the fake link-snap handler. It also cleans the test a little bit and reduces all wait-times to the minimum as they are not ciritcial anymore to drive the test.
Merge pull request #7774 from pedronis/seed-20-optional-snaps-support
seed: proper support for optional snaps for Core 20 models
Optional model snaps get included if they are specified explicitly via SetOptionsSnaps, that means --snap to prepare-image/ubuntu-image.
To enable this this had to relax the checks around invoking SetOptionsSnaps vs dangerous. The allowed features are now using the later checks, that were already in place, but the corresponding test had to be modified.
As a temporary stop-gap, explicitly deny the noisiest denials until bug #1849753 is fixed.
There is no chance of regression for this since snap-confine is already
denying them. We could consider making these allowed rather than
explicit deny, but people haven't been complaining about these classic
chromium content api snaps not working, so I'd like to continue denying
for now.
