Exceptions 0.10.0 #61
Conversation
the previous implementation was just delegating to 'mask', and was thus interruptible
| unwrap (f $ \m -> (wrap $ \s' -> io_restore (unwrap m s'))) s | ||
| where | ||
| wrap g = GhcT $ GHC.GhcT $ \s -> MTLAdapter (g s) | ||
| unwrap m = unMTLA . GHC.unGhcT (unGhcT m) |
There was a problem hiding this comment.
I also took the opportunity to fix the implementation of uninterruptibleMask. It was previously defined as uninterruptibleMask = mask, so calling uninterruptibleMask body was misleadingly running body in interruptible mode, not in uninterruptible mode. The difference is that some calls, like threadDelay, are intentionally marked as being interruptible even when running inside mask. As a result,
> timeout 1000000 $ mask_ $ threadDelay 3000000
terminates after 1 second because the timeout successfully interrupts the threadDelay after 1 second, while
> timeout 1000000 $ uninterruptibleMask_ $ threadDelay 3000000
terminates after 3 seconds, because the timeout has to wait until the uninterruptibleMask computation terminates before aborting the computation.
Without this fix,
> timeout 1000000 $ runGhcT ... $ mask_ $ lift $ threadDelay 3000000
and
> timeout 1000000 $ runGhcT ... $ uninterruptibleMask_ $ lift $ threadDelay 3000000
both terminate after 1 second, whereas with this fix, the second terminates after 3 seconds, as desired.
| (\a -> unwrap (body a) s) | ||
| where | ||
| wrap g = GhcT $ GHC.GhcT $ \s -> MTLAdapter (g s) | ||
| unwrap m = unMTLA . GHC.unGhcT (unGhcT m) |
There was a problem hiding this comment.
This is the main fix: exceptions-0.9 introduced a new method named generalBracket, whose type I fixed in exceptions-0.10, so in order to upgrade from exceptions-0.8 to exceptions-0.10, we have to implement this method. If we wanted, we could add an #ifdef around this method definition in order to support both.
There was a problem hiding this comment.
What about depending on exceptions >= 0.8.0?
There was a problem hiding this comment.
Do you mean >= 0.10.0, or did you change your mind and now want an #ifdef in order to support both 0.8 and 0.10?
There was a problem hiding this comment.
I indeed meant >= 0.10.0 - thanks for spotting that. I would not bother supporting 0.8 and earlier, especially given how it's a v0, and how we're likely one of the last libraries to add support for the new exceptions versions.
Fixes #54