GitLab

Fixed bootstrap failure on linux for --target=mingw32
Signed-off-by: Sergei Trofimovich <slyfox@gentoo.org>

Since I added the SomeAsyncException class, we weren't calling the
StackOverflowHook() properly for stack overflows, because this bit of
code was not expecting the extra layer in the hierarchy.

Trac #876 is the oldest ticket I have fixed in a long time.
I finally figured out how to make foldr behave in a non
space-leaky way for length.

Thanks to Andy for re-opening.

This is because the new kind-polymorphic Typeable seems to
require optimisation to deal with typeRep operations on
(extremely) large types.

This patch makes the Data.Typeable.Typeable class work with arguments of any
kind. In particular, this removes the Typeable1..7 class hierarchy, greatly
simplyfing the whole Typeable story. Also added is the AutoDeriveTypeable
language extension, which will automatically derive Typeable for all types and
classes declared in that module. Since there is now no good reason to give
handwritten instances of the Typeable class, those are ignored (for backwards
compatibility), and a warning is emitted.

The old, kind-* Typeable class is now called OldTypeable, and lives in the
Data.OldTypeable module. It is deprecated, and should be removed in some future
version of GHC.

This is caused because ensureIOManagerIsRunning is repeatedly called and this was initializing new IO managers on each call. Fixed so that a new manager is not created if one already exists.

On OS X, set haveOneShot = False and always wakeup the manager after a fd is registered.

kevent64() is only available on OS X. It appears to be buggy, so we use kevent() instead, which is also available on FreeBSD.

ioManagerCapabilitiesChanged takes no arguments and instead gets the current enabled number of capabilities by executing getNumCapabilities after it acquires a lock on the IO managers array.

Use ordinary Bool value to distinguish between backends supporting oneShot mode and eliminate several uses of conditional compilation in favor of ordinary conditional expressions using this Bool value. The benefit of this change is that more of the code compiles on more of the platforms.

Undo recent change to the type of GHC.Event.Thread.getSystemEventManager and update the commentary on this function.

Move callback invocation to after close. close must be run after the all backends are updated. Therefore the sequence is to update the backends, in the process getting the callbacks to invoke (actually just getting an IO action which when executed will execute the callbacks), call close, and finally update the Managers' callback tables for the fd and execute the callbacks.

In closeFdWith moved closing of fd after calling closeFd_ on each capabilities manager, ensuring that the fd is still valid when the fd is deregistered with the backend.

This makes closeFd follow the patten of closeFd_.

The file must not be closed when unregistering interest in the file with the backend, because otherwise the backend will likely encounter an error (invalid fd). Therefore, the close fd command is executed after the backend is updated.

This missing wakeup may have affected the IO manager when used with the poll backend.

Making this comment stand alone and not reference the comment on closeFd.