GitLab

Update STM implementation for SMP builds

MERGE TO STABLE

Put back in a missing case for higher-rank types. When the
definition is
	a) non-recursive
	b) a function binding
	c) lacks a type signature
we want to *infer* a perhaps-higher-rank type for the RHS,
before making a monomorphically-typed Id for the LHS.

E.g. 	f = \(x :: forall a. a->a) -> (x True, x 'c')

This case got lost in the transition to 6.4

tc194 tests it

Fix Linker on Windows: there was one more place where we had to check
for ".ctors" sections.

Fix url

something very strange happened with previous commit; try again

Fix threaded RTS with older gcc versions. (merge rev. 1.215.2.10)

Move RTS_SUPPORTS_THREADS into RtsConfig.h

implement lockClosure properly

isObjectLinkable: don't return True for an empty linkable

If hsc_lang is HscNothing, produce an empty linkable.

Don't write the interface in JustTypecheck mode

Yet another wibble to the GHCi top-level interaction story; avoid dup error messages

typo

More details about the representation of types after type checking.

Simplify and improve the Capability-passing machinery for bound
threads.

The old story was quite complicated: if you find a thread on the run
queue which the current task can't run, you had to call
passCapability(), which set a flag saying where the next Capability
was to go, and then release the Capability.  When multiple
Capabilities are flying around, it's not clear how this story should
extend.

The new story is much simpler: each time around the scheduler loop,
the task looks to see whether it can make any progress, and if not, it
releases its Capability and wakes up a task which *can* make some
progress.  The predicate for whether we can make any progress is
encapsulated in the (inline) function ANY_WORK_FOR_ME(Condition).
Waking up an appropriate task is encapsulated in the function
threadRunnable() (previously it was in two places).

The logic in Capability.c is simpler, but unfortunately it is now more
closely connected with the Scheduler, because it inspects the run
queue.  However, performance when communicating between bound and
unbound threads might be better.

The concurrency tests still work, so hopefully this hasn't broken
anything.

Print only the result of a bind stmt

- scheduleDetectDeadlock() should invoke GarbageCollect() via
  scheduleDoGC(), which collects the Capabilities first.

- scheduleDoGC() doesn't need the Capability.  Also, it appears that
  there was an out-of-bounds array access in here.

- add printThreadQueue(), useful from gdb.

- fix a couple of warnings.

rename exportsFromAvail

Further GHCi wibbles

a) Don't print the value of a 'let'
b) Only one error message for 'print id'

Dead code (HEAD only)

Fix what "standard class" means

	MERGE to STABLE

The Report's definition of defaulting (Section 4.3.4) defines a "standard class" to be
one defined by the Prelude or a standard library.  GHC was omitting the "..or a standard
library" part; and it even omitted the Prelude classes Monad and Functor.
This commit fixes the bug by adding

	Random, RandomGen,
	Monad, MonadPlus, Functor

to the classes defined as "standard".

Comments and type sig only

Track new semantics of splitLongestPrefix; fixes -main-is bug

Clarify code for splitLongestPrefix; no effect on behaviour

needsCDecl: remove extra equation for CaseLabel, which was overlapped
(and wrong!).

Warning police (format strings, unused variables)

Warning police: Removed overlapped patterns. In general, taking
compiler warnings about unmatched patterns seriously when they come up
later might be better than writing catch-all patterns right from the
start. Otherwise readers are confused and wonder which patterns might
be missing when there are none.

Warning police: Removed an overlapped catch-all pattern (which could
really never match due to genprimopcode)

Warning police: Use non-deprecated form of "foreign import"

While acting as warning police, I tried to clean up the #ifdef chaos a
little bit, I hope nothing has been broken due to that... :-]

Bugfix: in rtsFatalInternalErrorFn vsnprintf function was used instead of
snprintf. The code is changed so that the error message will be displayed
even if it is longer that BUFSIZE.

Implement some more error checking to catch some cases where
registering a package will lead to a package database containing
conflicts, which would otherwise prevent GHC from being used without
any -hide-package options.

In 'update' mode, instead of complaining about conflicts, we now
attempt to hide any packages which would cause a conflict.  Previously
this was limited to just older versions of the current package, now it
applies to all packages which contain, or depend on packages which
contain, modules which conflict with any module belonging to the
current package or a dependency of it.

Unfortunately we still can't cope with conflicts that cross the
boundary between the user package databse and the global one.  We will
need some kind of white-out mechanism in order to be able to hide a
global package in the user database.

Documentation for new GHCi behaviour

Further wibble to preceding GHCi commit

Wibble to preceding GHCi commit

Improve the GHCi interaction

		Merge to STABLE?

This fix addresses Sourceforge #1156554 "GHCi: No instance for (Show (IO ()))",
and simultaneously improves the top-level interaction in two other ways:

- Only one error can show up (previously there could be two)

- If an I/O action gives a Showable result, the result is printed
  (provided it isn't ()).  So
	prompt> return 4
  prints 4, rather than nothing

- For command-line 'let' and 'x<-e' forms, if exactly one variable
  is bound, we print its value if it is Showable and not ()
	prompt> let x = 4
	4
	prompt> x <- return 5
	5

Liberalise the defaulting rules for GHCi

		Merge to STABLE?

The H98 defaulting rules are these.  Group constraints of the form (C v),
for some particular type variable v. Then default v if

a) v appears only in this group (no implicit params or D [v])
b) at least one of the C's is numeric
c) all the C's are standard

GHCi changed rules (b) to

(b') at least one of the C's is numeric, or Eq,Ord,Show

This commit further liberalises GHCi to change (c):

(c') any of the C's are standard

Otherwise the existence of a non-standard class (e.g. Random) is enough
to kill defaulting altogether.

Comments only

Fix locking when unblocking a thread in take/putMVar.  Due to CPP
nonsense the previous locking wasn't actually working, which lead to
deadlock problems.  It now turns out that I can call unblockOne
directly rather than needing unblockOneLocked (the lock on the MVar
means I have exclusive access to the threads on its queue).

- Move the call to threadPaused() from the scheduler into STG land,
  and put it in a new code fragment (stg_returnToSched) that we pass
  through every time we return from STG to the scheduler.  Also, the
  SAVE_THREAD_STATE() is now in stg_returnToSched which might save a
  little code space (at the expense of an extra jump for every return
  to the scheduler).

- SMP: when blocking on an MVar, we now wait until the thread has been
  made fully safe and placed on the blocked queue of the MVar before
  we unlock the MVar.  This closes a race whereby another OS thread could
  begin waking us up before the current TSO had been properly tidied up.

Fixes one cause of crashes when using MVars with SMP.  I still have a
deadlock problem to track down.