Nim sets don't support negative ID in thief ID checks

[mratsim]

From @liquid600pgm:

/path/to/.nimble/pkgs/weave-#master/weave/scheduler.nim(260) worker_entry_fn
/path/to/.nimble/pkgs/weave-#master/weave/scheduler.nim(225) schedulingLoop
/path/to/.nimble/pkgs/weave-#master/weave/workers.nim(40) runTask
/path/to/.nimble/pkgs/weave-#master/weave/parallel_for.nim(221) weaveTask_ParallelFor_
/path/to/.nimble/pkgs/weave-#master/weave/parallel_for.nim(38) weaveParallelForSection
/path/to/.nimble/pkgs/weave-#master/weave/runtime.nim(114) loadBalance
/path/to/.nimble/pkgs/weave-#master/weave/victims.nim(326) shareWork
/path/to/.nimble/pkgs/weave-#master/weave/victims.nim(302) distributeWork
/path/to/.nimble/pkgs/weave-#master/weave/victims.nim(246) splitAndSend
/path/to/.nimble/pkgs/weave-#master/weave/instrumentation/contracts.nim(86) evalSplit
/path/to/.choosenim/toolchains/nim-#devel/lib/system/fatal.nim(49) sysFatal
Error: unhandled exception: value out of range: -1 notin 0 .. 65535 [RangeError]

[mratsim]

This is now "reliably" reproducible in CI for #107 for example: https://dev.azure.com/numforge/Weave/_build/results?buildId=404&view=logs&jobId=5336da5d-2772-5db2-cb26-68743e5bcd30&j=5336da5d-2772-5db2-cb26-68743e5bcd30&t=76edda6a-eee8-5183-4ec0-7505684fb3bb

[mratsim]

Analysis:

With the following compilation command

nim c -d:WV_debugSplit -d:WV_LazyFlowvar --verbosity:0 --hints:off --warnings:off --threads:on --outdir:build -r weave/parallel_for.nim

After running it a couple times you will get a stacktrace similar to:

Worker 15: 44 steps left (start: 156, current: 156, stop: 200, stride: 1, 6 thieves)
Worker 14: loop task 0x54504389 (iterations [140, 142)) waiting for the remainder
Worker 30: Finished loop task 0x54504389 (iterations [144, 145)) (futures: 0x00000000)
Worker 14: loop task 0x54504389 (iterations [140, 142)) complete
Worker  4: Sending [127, 129) to worker 10 (2 steps) (hasFuture: 1, dependsOnFutures: 0x00000000)
Worker  4: Finished loop task 0x54504389 (iterations [126, 127)) (futures: 0x3001d800)
Worker 10: has 0 steal requests queued
Worker 10: workSharing split, thiefID 22, total subtree thieves 3, left{id: 21, waiting: 1, requests: 1}, right{id: 22, waiting: 1, requests: 2}
Worker 10: 2 steps left (start: 127, current: 127, stop: 129, stride: 1, 3 thieves)
/home/beta/Programming/Nim/weave/weave/scheduler.nim(260) worker_entry_fn
/home/beta/Programming/Nim/weave/weave/scheduler.nim(225) schedulingLoop
/home/beta/Programming/Nim/weave/weave/workers.nim(40) runTask
/home/beta/Programming/Nim/weave/weave/parallel_for.nim(233) weaveTask_ParallelForAwaitable_
/home/beta/Programming/Nim/weave/weave/parallel_for.nim(77) weaveParallelForAwaitableSection
/home/beta/Programming/Nim/weave/weave/runtime.nim(114) loadBalance
/home/beta/Programming/Nim/weave/weave/victims.nim(326) shareWork
/home/beta/Programming/Nim/weave/weave/victims.nim(302) distributeWork
/home/beta/Programming/Nim/weave/weave/victims.nim(246) splitAndSend
/home/beta/Programming/Nim/weave/weave/instrumentation/contracts.nim(86) evalSplit
/home/beta/.choosenim/toolchains/nim-1.2.0/lib/system/fatal.nim(49) sysFatal
Error: unhandled exception: value out of range: -1 notin 0 .. 65535 [RangeError]
Worker 10: Sending [128, 129) to worker 22 (1 steps) (hasFuture: 1, dependsOnFutures: 0x00000000)
Worker 10: Finished loop task 0x54504389 (iterations [127, 128)) (futures: 0xd400c200)
Worker 10: Finished loop task 0x54504389 (iterations [127, 128)) (futures: 0xd400c200)
Worker 10: loop task 0x54504389 (iterations [127, 128)) waiting for the remainder
Worker  0: Sending [123, 126) to worker 1 (3 steps) (hasFuture: 1, dependsOnFutures: 0x

The first thing to notice is that this is not a Weave error. Weave does not throw RangeError everything is an AssertionError this is further confirmed by the fact that all Weave errors prepend the Worker thread ID to ease debugging.

The second thing is that int16 has 65536 values and sets are only allowed to hold int16.

And we happen to have a set there:

weave/weave/victims.nim

Lines 203 to 206 in 052ae40

	if workSharing:
	# The real splitting will be done by the child worker
	# We need to send it enough work for its own children and all the steal requests pending
	ascertain: req.thiefID in {myWorker().left, myWorker().right}

However I'm pretty sure that sets in the past could hold negative values from -32768 to 32767 instead of 0 to 65535. it seems like not but my particular construct wasn't properly checked.

[mratsim]

Probably linked to nim-lang/Nim#13764