👋 Hello!

I am trying to understand what the intended behavior is supposed to be when acquiring a semaphore and changing the max at the same time to a lower number. Here is an example:

# First acquire should work
iex(1)> Semaphore.acquire(:a, 5)
true

iex(2)> Semaphore.count(:a)     
1

# Increasing the max and acquire again
iex(3)> Semaphore.acquire(:a, 7) 
true

iex(4)> Semaphore.count(:a)     
2

# Acquire with the same max
iex(5)> Semaphore.acquire(:a, 7)
true

iex(6)> Semaphore.count(:a)     
3

# Now acquire with a lower max, shouldn't this return false?
iex(7)> Semaphore.acquire(:a, 2)
true

# This count feels a little disingenuous because of the previous statement returning true
iex(8)> Semaphore.count(:a)     
2

What is the behavior supposed to be when lowering the max?

At first my intuition was that the count would be reset to 0, and that's why the iex(7) succeeded, but then when I look at the count I see 2, which is quite confusing. The 2 would make sense if the return value of iex(7) was false (i.e. the semaphore is still full but we now have a lower limit). Another possibility would be that changing the max via the acquire does not alter the count.

Additionally, I am not sure what the use case is for adjusting the max while acquiring at the same time. Would it makes things more clear to have a separate set_max function?

See this following repro:

{pid, ref} = spawn_monitor fn ->
  Semaphore.call(
    :foo,
    5,
    fn ->
      spawn_link(
        fn ->
          exit(:ded)
        end
      )
      Process.sleep(50)
    end
  )
end
receive do
  {:DOWN, ^ref, :process, ^pid, reason} ->
    count = Semaphore.count(:foo)
    IO.puts("Process down for reason #{inspect reason} - semaphore leaked? #{count}")
end

When ran, it will output:

Process down for reason :ded - semaphore leaked? 1

It should output:

Process down for reason :ded - semaphore leaked? 0

  def call_linksafe(name, max, func) do
    if acquire(name, max) do
      safe_key = {name, self()}
      inserted = ETS.insert_new(@call_safe_table, [safe_key])

What if the process killed before ETS.insert_new(@call_safe_table, [safe_key]) ?
Seems like we still have problem, although time window for this is rather small.

just wanted to mention that your code should always expect that additional fields can be added to any object at any time, that should not break your code

Originally posted by @advaith1 in discord/discord-api-docs#3769 (comment)