Note that the command was `./pants`, running v1 ta...
# development
Note that the command was
, running v1 tasks...
👍 2
Pants V1 still runs V2 code for build file parsing and target creation
It got past that and started running tests
👍 1
but those tests do generate BUILD files and such
so, i repro this for
./pants test contrib/go/tests/python/pants_test/contrib/go/tasks
, but it looks specific to the test harness. it’s the “inner” pants that is under test that experiences this, and not the outer pants. and moreover, if you hit
on the subprocess, you can see that it is a plain jane “we’ve opened hundreds of Schedulers for some reason” case, and we do actually hit a 10k limit about halfway through the test.
so… that’s an upside. but i don’t see why
would be leaking schedulers, so.
Huh. Remember that Alex used to hit this all the time? The solution was to make the scheduler set up be per class, rather than per test
that test target contains about 8 files and 10 total instances of
. but 10 schedulers shouldn’t be enough to hit these limits.
(we don’t open anywhere near a thousand files per scheduler)
aaaaanyway. too many yaks today.