Removes the purge thread in favour of standard ScheduledThreadPoolExecutor APIs

[chggr]

The RXSchedulerPurge thread is currently enabled by default and runs every second
to call purge() on each executor in the pool. This is causing significant issues
for battery powered devices (e.g. mobile phones), because it needs to periodically
wake up the CPU to perform purging. The RXSchedulerPurge thread could be completely
removed in favour of using the standard setRemoveOnCancelPolicy() API on the
ScheduledThreadPoolExecutor which became available in Java 7 and offers removal
of cancelled tasks at the moment they are cancelled in O(1).

[akarnokd]

The question is, how does this affect compatibility and desugaring?

/cc @JakeWharton

[JakeWharton]

I do not recall the minimum Android version we decided on for 3.x. It doesn't seem explicitly documented anywhere.

The API in question (setRemoveOnCancelPolicy) is available from Android API 21 and newer (which corresponds to Android 5.0 as a user-facing version). API 21 is a relatively safe minimum, but I suspect prior to this you were able to use 3.x on even lower API levels.

It's not unheard of for libraries to bump their minimum-supported Android API level in minor releases. So I'm supportive for the next minor version provided that the minimum is also documented somewhere.

[akarnokd]

Okay then, from 3.0.14, we require API 21. I'll add this to the readme.

[codecov]

Codecov Report

Merging #7293 (331b50c) into 3.x (c5883dc) will increase coverage by 0.01%.
The diff coverage is 100.00%.

@@             Coverage Diff              @@
##                3.x    #7293      +/-   ##
============================================
+ Coverage     99.53%   99.54%   +0.01%     
+ Complexity     6756     6744      -12     
============================================
  Files           747      747              
  Lines         47391    47347      -44     
  Branches       6382     6376       -6     
============================================
- Hits          47170    47133      -37     
+ Misses          103       96       -7     
  Partials        118      118              

Impacted Files	Coverage Δ
...va/io/reactivex/rxjava3/schedulers/Schedulers.java	100.00% <ø> (ø)
...ava3/internal/schedulers/SchedulerPoolFactory.java	100.00% <100.00%> (ø)
...l/operators/observable/ObservableFlatMapMaybe.java	91.54% <0.00%> (-6.34%)	⬇️
.../operators/observable/ObservableFlatMapSingle.java	94.44% <0.00%> (-2.39%)	⬇️
...operators/flowable/FlowableFlatMapCompletable.java	98.71% <0.00%> (-1.29%)	⬇️
...a/io/reactivex/rxjava3/subjects/ReplaySubject.java	99.16% <0.00%> (-0.42%)	⬇️
...3/internal/operators/flowable/FlowableGroupBy.java	84.73% <0.00%> (-0.30%)	⬇️
...ternal/operators/observable/ObservableFlatMap.java	98.22% <0.00%> (ø)
...ernal/operators/flowable/FlowableFlatMapMaybe.java	98.44% <0.00%> (+0.51%)	⬆️
...rnal/operators/flowable/FlowableFlatMapSingle.java	94.18% <0.00%> (+0.58%)	⬆️
... and 6 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update c5883dc...331b50c. Read the comment docs.

[JakeWharton]

Sorry by minor I meant for a 3.1.0 rather than patch release in the 3.0.x series. At least then someone might look a bit more closely at the changes than a patch release where they expect only fixes and not new functionality / changes.

Ultimately you can decide though since it's not explicitly documented anywhere and this is really about making it explicit.

[akarnokd]

Indeed 3.1.0 would be more appropriate.