I just released a new version of the scheduler that should now correctly
schedule tests in experimental. The scheduler was run, and new jobs have
alreadyT been submitted.
In particular, the following will now trigger tests:
* Upload of foo_1.0~exp1 to experimental will trigger tests in an
unstable+experimental environment, with foo pinned to foo_1.0~exp1.
This means
-> foo_1.0~exp1 is tested, with all dependencies from unstable
(or experimental, if they don't yet exist in unstable)
-> All reverse dependencies of foo_1.0~exp1 are tested with their
versions from unstable
* With bar_1.0 in unstable, upload of one of its dependencies (say,
gcc-13) to experimental will trigger a test of bar_1.0 in unstable
with the dependency pinned to experimental
Curiously enough, this required a massive overhaul of the scheduler. The
scheduler was conceived as a very rough prototype, and had grown
organically to the point where implementing the above was just not
feasible due to some early design decisions.
This version also fixes a subtle bug [1] that could lead to the wrong
version of a dependency triggering a test, e.g. gcc-13 triggering for
bookworm (gcc-13 does not exist there). Oddly, it seems that this hasn't
manifested itself in an obvious way so far.
Finally, I removed tracking linux-signed-amd64 from all tests. This
triggered tests on uploads of linux-signed-amd64, but that didn't make
sense because tests would need to wait for the new kernel to become
active in the testbed, which would frequently be too late even for daily
rebuilt QEMU worker images. I'll have to think of a new process for that.