I actually don't like this change, for several reasons:
Bumping up some arbitrary limits is not a solution at all, the flakiness
already
came back (see our bots). Note that those tests regularly fail on our bots
for
months now, with more than half a dozen half-hearted repair attempts. This
is
really annoying for the whole team, because you constantly get "you broke
the
build" mails for no good reason.
The other reason why I don't like the change is that the tests are not "good
team players" in the unit test sense: Any unit test taking longer than 100ms
(probably even less) is very bad if you consider the runtime of the whole
suite.
If every test took 500ms and we would be able to perfectly distribute this
on 32
cores, the suite would take 2min35sec, on a 4 core Laptop almost 21min.
This is
not acceptable, running the whole suite should be in the range of 10-20sec,
otherwise people test even less than they do now.
The bug tracker has a proposed solution with an increasing timeout, this
should
really be implemented and it should be checked if the typical case is below
100ms. If this doesn't work out, the test should be done in a fundamentally
different way and if that's not possible we should drop the tests completely
from our suite, the annoyance-to-usefulness ration is a bit too high...
https://codereview.chromium.org/16280011/
--
--
v8-dev mailing list
[email protected]
http://groups.google.com/group/v8-dev
---
You received this message because you are subscribed to the Google Groups "v8-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/groups/opt_out.
