Message210558
In the end I think Charles-Francois is right -- the rounding up in selectmodule.c (for epoll) and selectors.py (for poll and, redundantly[*], for epoll) should be sufficient to avoid the busy-wait behavior with delays under 1ms.
We're not promising real-time guarantees at any level, so both returning too early and too late must be acceptable. If a test fails frequently despite relaxed constraints, that's a problem with the test.
So I think Victor should apply his first patch (remove_granularity.patch) to both CPython and Tulip and then we can adjust any tests that still cause occasional red buildbots.
[*] It's not redundant when using Tulip on Python 3.3, so I prefer to keep it. |
|
Date |
User |
Action |
Args |
2014-02-07 21:22:21 | gvanrossum | set | recipients:
+ gvanrossum, vstinner, neologix, python-dev |
2014-02-07 21:22:21 | gvanrossum | set | messageid: <1391808141.34.0.407980346092.issue20505@psf.upfronthosting.co.za> |
2014-02-07 21:22:21 | gvanrossum | link | issue20505 messages |
2014-02-07 21:22:21 | gvanrossum | create | |
|