asyncdispatch.poll() eats cpu; poll is not blocking

[enthus1ast]

import asyncdispatch
proc foo() {.async.} = 
  while true:
    echo "in foo"
    await sleepAsync(1000)

asyncCheck foo()
while true:
  echo "."
  poll(500)

results in 100% cpu load on one core. Tested on linux

edit: fixed typo

[dom96]

Proves we need to move to proper timers :)

[cheatfate]

All latest changes to poll() procedure causes this bug... And changing behavior of sleepAsync() to use hardware timers is not a solution for this problem.

[cheatfate]

After some investigation, i have found that this bug is regression, made by somebody who introduced this branch (https://github.com/nim-lang/Nim/blob/devel/lib/pure/asyncdispatch.nim#L1234).
As i mentioned in other issues, this happens just because of policy of rash and untested 'improvements'.

@enthus1ast, remove this line, and everything will be fine.

@dom96, while trying to fix this #4262 via 5bf1643, you have introduced this one.

[dom96]

Are you sure this is a regression? This line was always in there: https://github.com/nim-lang/Nim/blame/000b8afd26fa16684a116d9afe798ea94df9c270/lib/upcoming/asyncdispatch.nim#L1222

[cheatfate]

@dom96, i'm sure that exactly this check must be removed, but this will enable #4262 again, but #4262 must be fixed in other way. This check is actually disables timeouts processing, when there no descriptors registered.

[enthus1ast]

this is the output of the above test with
asyncdispatch.nim#L1234). removed:

.
.
in foo
.
.
.
.
.
.
.
in foo
.
.
in foo
.
.
in foo
.
.
.
.
.
.
.
.
.
.
.
in foo
.
.
in foo
.

[cheatfate]

@enthus1ast, looks like work as expected, isn't it?

[enthus1ast]

@cheatfate it works! (i just have no clue how asyncdispatch works inside, i guess...)

[dom96]

You were right all along @cheatfate :)

I decided to finally investigate how to fix #7758/#7886 and ended up doing exactly what you suggest here. I can't reproduce #4262 anymore though.