Trouble using until() about nanonext HOT 8 CLOSED

So it looks like until() decrements the cv even when it hits the timeout. So I can create an even smaller reprex:

library(nanonext)
#> Warning: package 'nanonext' was built under R version 4.3.1
cv <- cv()
cv_value(cv)
#> [1] 0
cv_signal(cv)
cv_signal(cv)
cv_value(cv)
#> [1] 2
until(cv, 10) # decrements the cv
cv_value(cv)
#> [1] 1

^{Created on 2023-10-03 with reprex v2.0.2}

@shikokuchuo, is the decrement always guaranteed to happen on every call to until(). If so, I might be able to work with this by defining my own version:

crew_until <- function(condition, seconds_timeout) {
  on.exit(nanonext::cv_signal(condition))
  nanonext::until(cv = condition, msec = 1000 * seconds_timeout)
}

I will test this.

from nanonext.

Hmm..., it looks like nanonext::until() cannot decrement the cv any further if it is already 0, so then cv_signal(condition) would be incorrect. And writing if (cv_value(condition) > 0L) cv_signal(condition) is a race condition because what if a task completes after if() but before cv_signal()?

On my end, the only other solution I can think of is to preemptively call cv_signal() a few thousand times before I submit any tasks at all, but that seems really messy, and I am not sure what edge cases that would cause.

So then would it be possible for until() to not signal the cv in the first place?

from nanonext.

Hmm... maybe I am misunderstanding the intended use cases of wait() and until(). I can see now that this is the intended behavior in the docs. In addition, I notice that until() only waits if cv_value(cv) is 0. Any positive value will not cause a wait:

library(nanonext)
cv <- cv()
cv_signal(cv)
until(cv, 10000) # returns control instantly

This creates a challenge for me when using mirai::nextget("cv"). I would like to be able to wait on different kinds of conditions. Mainly:

• Wait until a cv value changes, regardless of its current value, without changing its value further.
• Wait until cv_value(cv) reaches a certain fixed number. controller$wait() and controller$map() could be simplified a lot with something like nanonext::until(cv, msec, value_wake_up = 7L, decrement_value = FALSE).

from nanonext.

And I think a manual cv_decrement() could also add a stronger meaning to nextget("cv"). Currently, cv_value(nextget("cv")) is the number of resolved tasks in the lifetime of the crew controller, regardless of whether they have been returned with pop(). If nanonext were to support a manual way to decrement a cv, then I could decrement it on each controller$pop(). Now, the cv would indicate the number of resolved tasks currently available for collection, which is much more powerful and would allow me to make controller$wait() simpler and more efficient.

from nanonext.

until(cv, 1) decrements your cv safely.

from nanonext.

until() is correct and behaves as documented. If it is something else you are after please open a new feature request. It will not be implemented as a change to until(). Thanks.

from nanonext.

Sorry, that was a lot of thinking out loud on my part.

from nanonext.

No worries! My comment just meant that any new functionality should belong in a new function. These synchronisation primitives are basic building blocks and it would not be right to amend them however convenient it may seem.

from nanonext.