"Trivial" program results in "thread blocked indefinitely in an MVar operation"

[ocharles]

The following:

module Main where

import Control.FRPNow

main :: IO ()
main = runNowMaster (return never)

When compiled (with ghc --make -threaded) results in

Test: thread blocked indefinitely in an MVar operation

when executed.

[atzeus]

This is due to nothing actually happening. GHC runtime will detect that and then throw this error. I'll see if I can prevent it.

[atzeus]

Fixed! This will print the following in the next release:

FRP System has become inactive because nothing else is going to happen.We could keep on waiting forever, but instead we will just exit the FRP system now...Bye!

[Xandaros]

I was interested in how you implemented this, but I can't actually find the commit. Where was this done?

[atzeus]

Whoops, forgot to push. The code is at work, I'll push it tomorrow. Essentially, I catch an Indefinitly blocked on MVar exception.

[int-index]

@atzeus What if I want to catch it myself? I think throwing an exception is a sensible behavior, it just needs to be documented.

[atzeus]

Hmm, you're right. I propose throwing a FRPNowInactive exception, and documenting it. Agreed?

[int-index]

@atzeus Agreed, if you're sure that you can't accidentally catch exceptions thrown due to user-defined MVars (as opposed to MVars used by FRPNow internally).

[atzeus]

Sadly, I can't guarantee that, since IndefinitlyBlockedOnMVar exception does not tell which MVar is involved :( I propose I document that as well..

[int-index]

Throwing FRPNowInactive when the thread is blocked on a user-defined MVar isn't better than throwing IndefinitelyBlockedOnMVar when the FRP system is inactive.

I think it's better not to mess with this and just document that when the FRP system becomes inactive, an MVar exception may be thrown.

[atzeus]

Agreed. I propose I just dump a message to standard error saying that the following exception may be due to the FRP system becoming inactive and (re)throw the indefinitly blocked on MVar exception. Agreed?

[int-index]

I don't like libraries doing IO without my consent or explicit command. Messages are something the user can see, but this message is targeted at the developer.

But that's just me.

[atzeus]

I just realized that it is actually possible to guarantee that a FRPSystemInactive exception comes from the FRPNow MVar instead of a user defined MVar, so I will just do that :)

[int-index]

Perfect! :)

[atzeus]

Commited!