use slots for ParkingLot and Event

[belm0]

Event instances can be very numerous in a Trio program, especially considering idioms like "create a new event on every set()". Event contains ParkingLot and neither were using slots, which is a source of garbage objects.

closes #1943

[codecov]

Codecov Report

Merging #1944 (ef3f94e) into master (6754c74) will decrease coverage by 0.08%.
The diff coverage is 100.00%.

❗ Current head ef3f94e differs from pull request most recent head a21abe0. Consider uploading reports for the commit a21abe0 to get more accurate results

@@            Coverage Diff             @@
##           master    #1944      +/-   ##
==========================================
- Coverage   99.58%   99.50%   -0.09%     
==========================================
  Files         114      114              
  Lines       14617    14617              
  Branches     1116     1116              
==========================================
- Hits        14557    14544      -13     
- Misses         43       51       +8     
- Partials       17       22       +5     

Impacted Files	Coverage Δ
trio/_core/_parking_lot.py	100.00% <100.00%> (ø)
trio/_sync.py	100.00% <100.00%> (ø)
trio/_core/tests/tutil.py	96.61% <0.00%> (-3.39%)	⬇️
trio/_core/_multierror.py	98.87% <0.00%> (-1.13%)	⬇️
trio/_core/tests/test_ki.py	99.13% <0.00%> (-0.87%)	⬇️
trio/_core/tests/test_multierror.py	99.27% <0.00%> (-0.73%)	⬇️
trio/tests/test_ssl.py	99.57% <0.00%> (-0.29%)	⬇️
trio/tests/test_testing.py	99.76% <0.00%> (-0.24%)	⬇️

[njsmith]

LGTM. TBH we should probably rewrite Event at some point to use a plain set and wait_task_rescheduled, instead of all the heavyweight ParkingLot machinery. (I was very into ParkingLot when first writing trio's synchronization primitives, but my excitement has dropped over time.) If you're worried about Event's memory usage, do you want to do that here?

[belm0]

merge blocked by #1938

[belm0]

TBH we should probably rewrite Event at some point to use a plain set and wait_task_rescheduled, instead of all the heavyweight ParkingLot machinery.

The parts of ParkingLot used by Event() don't look heavy-- I guess Event could use a simple list of tasks and just iterate that on set(), vs. ParkingLot's OrderedDict and popitem() 🤷‍♂️

[njsmith]

It would be a `set` of tasks, not a `list`, because cancellation needs to be able to delete arbitrary items efficiently. But yeah, that's what I mean.

…

On Sun, Apr 4, 2021 at 11:28 PM John Belmonte ***@***.***> wrote: TBH we should probably rewrite Event at some point to use a plain set and wait_task_rescheduled, instead of all the heavyweight ParkingLot machinery. The parts of ParkingLot used by Event() don't look heavy-- I guess Event could use a simple list of tasks and just iterate that on set(), vs. ParkingLot's OrderedDict and popitem() 🤷‍♂️ — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#1944 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAEU42DOVMTS5D57W6YDEGLTHFKA5ANCNFSM42MAQ2YQ> .

-- Nathaniel J. Smith -- https://vorpus.org <http://vorpus.org>

[belm0]

closing in favor of #1948