Yampa documentation does not built automatically on hackage

[ivanperez-keera]

This is a follow up of #302 (comment)

[ivanperez-keera]

This is what I got from @gbaz

Looks like yampa-test was running too long, timing out, preventing the
builder from ever hitting yampa itself before the restart. I killed
the yampa-test builds manually. I suggest in the future you turn off
each version of yampa-test you build in "test settings" in the
maintainers corner when you upload. Would be nice if we had a patch to
do it once and forall for all versions, but ah well.

I still do not understand why yampa-test would be built before yampa. I wonder if there's a way to prevent this while still having meaningful tests and without any manual step required.

[gbaz]

the hackage builder runs in cycles, and each cycle assembles a reverse-chronological list of all packages it needs to build/test, then proceeds through them. a wrapper script will kill it if it runs too long without progress. so the reason yampa-test is built before yampa is just that it is uploaded, chronologically, directly after yampa. therefore yampa doesn't get built, and on each cycle, yampa-test times out, preventing yampa from being built. it would certainly be a nice improvement to hackage-builder if it could record timeouts as failures, but that would require at least a little rearchitecting -- perhaps added the timeout checking mechanism to the builder itself and not just the surrounding wrapper script.

[ivanperez-keera]

I see. It seems that, in the short term, there are only two quick possible actions on our side:

• Introduce an artificial delay between when we upload yampa and when we upload yampa-test.

• Reduce or optimize the tests in yampa-test so that they complete faster.