GitLab

64 bit fix

Use stgMallocBytesRWX for allocating info tables, since the memory
needs to be executable (not sure how this is working on OpenBSD right
now, but it definitely breaks on x86_64/Linux).

Move mallocBytesRWX into RtsUtils, rename it to stgMallocBytesRWX, and
export it.

Add some comments on why we don't use posix_memalign() right now.

declare stg_returnToSchedNotPaused (forgot to commit this yesterday)

After some experiments, it seems like we're stealing too many registers from
newer GCCs on SPARC, leading to

   "unable to find a register to spill in class `GENERAL_REGS'"

errors. The fix is to leave l6 and l7 to GCC. Tested with a full 2-stage
bootstrap (including OpenGL/GLUT packages) on SPARC Solaris 8 with GCC 3.4.4.

A test case for this (which I'm too lazy/tired to commit) is:

   module Blah ( foo ) where
   import Foreign.Ptr ( FunPtr )
   type Bar = Int -> Double -> Double -> Double -> IO ()
   foreign import ccall unsafe "dynamic" foo :: FunPtr Bar -> Bar

SimonM: MERGE TO STABLE (if nobody yells)

Avoid calling threadPaused() on exit from STG land if we're just
switching to the interpreter, and conversely call threadPaused() in
the interpreter if we're returing to the scheduler for anything other
than switching to STG.

This will probably fix the recent slowdown in GHCi (ioref001 test, for
example).  It was broken when we moved the threadPaused() call into
STG from the scheduler, so it only affects the HEAD.

make_constr_itbls: 64-bit fix (size of info table was hardcoded)

byte code info tables for x86_64 (same as i386)

Turn on GHCi for x86_64

GHCi staggers to its feet on x86_64.

The problem is that in the small code model on x86_64, all symbol
relocations are 32-bit, because the program is assumed to fit into
2Gb.  However, shared libraries are linked outside the 2Gb range, and
the linker arranges that references to shared library symbols either
go via a jump table (for code references) or are copied into the 2Gb
area (for data references).

In GHCi's linker, We can use a jump table for the function symbols,
but we can't copy the data symbols because they've already been
relocated to point to the shared library.  What's more, we can't tell
whether a symbol reference is to code or data - this information is
only available from the shared library itself.

Fortunately, at least HSbase.o doesn't have any data references to
shared libraries (at least on the Linux distro I'm using), so I can at
least get GHCi up and limping.

awakenBlockedQueue(): check for NULL.  This is a hack to workaround a
bug that occurs when async exceptions, unsafePerformIO, and
interruptible operations are used together.  See the comment in
Exceptions.cmm and bug #1235728 for more details.

Wibbles to unifyFunTy error messages

Improvements to speakN, define speakNOf, move plural from TcSimplify

mingw32: ignore isDoesNotExistError failure from
GHC.ConsoleHandler.flushConsole

Submitted-by: Brian Smith <brianlsmith at gmail.com>

Improve the error message from unifyFunTys.  Previously we got a really
horrible message from this:
	  t = ((\Just x -> x) :: Maybe a -> a) (Just 1)
    
Try.hs:1:6:
    Couldn't match the rigid variable `a' against `t -> t1'
      Expected type: a
      Inferred type: t -> t1

Now it's much better:

Try.hs:14:6:
    The lambda expression `\ Just x -> ...' has two arguments,
    but its type `Maybe a -> a' has only one
    In the expression: (\ Just x -> x) :: Maybe a -> a


tcfail140 tests some cases

Fix a bug in the renamer for parallel list comprehensions
	MERGE TO STABLE

It's surprinsingly tricky to combine 
  a) The parallel scopes for par-list-comps
with
  b) The general form of the renamer types, whereby
     scoped constructs work like
	 rnPat :: Pat -> RnM (thing,FreeVars)
	       -> RnM ((Pat,thing), FreeVars)
     This general shape neatly allows rnPat to
     extend the envt, report unused variables from
     the 'thing' inside, and return the correct set
     of free variables

But combining (a) and (b) is tricky, and was plain wrong before.

Add speakN, and the ability to set printing depth

enable splitting on x86_64

x86_64: Pass -fno-asynchronous-unwind-tables to gcc, which eliminates
some unnecessary junk from the via-C generated code and allows
-split-objs to work.

Fixed & simplified previous commit.

Silence the error message when ld doesn't understand the -x option.

Solaris' sed obviously doesn't like nested \(\), so do the gcc version
detection magic in two steps instead of one.

Fix building on IA64.

Please merge to stable.

Revert accidental commit (merge to stable if bogus commit was merged)

MERGE TO STABLE (once I"ve checked it works)

The dictionary-floating code in Specialise wasn't taking 
variables free in the RULES into account, which resulted in
such variables perhaps going out of scope.

MERGE TO STABLE

Add a check for Haskell-98 mode, to check that there is no type
synonym in an instance declaration.

tcfail139 tests this case

genOutputFilenameFunc: fix output filename generated for non-Haskell
compilations with -odir.

Clean .hi-boot and .o-boot files too

small performance fix: in via-C mode we previously always created a
switch instead of an conditional-tree for a multi-branch case.  Refine
this slightly so that 2-branch switches turn into conditionals again,
since gcc doesn't do a good job of optimising the equivalent switch.

DocBook validation police

ghc -e should exit(1) if the statement fails to compile.

Fixes #1233146 (this was already partially fixed, in that ghc -e would
exit(1) if the modules fail to compile, but if the expression itself
was invalid it would still exit(0)).

Fix ffi on OpenBSD by dynamically allocating the obscure_ccall_ret_code.
Also merge in SimonM's wrapper over inline asm patch.

setContextAfterLoad: include the Prelude if the top module is compiled
(this broke at some point in the HEAD, 6.4 works fine).

Vital wibble to last commit!

MERGE TO STABLE

Fix a long-lurking renamer bug, concerning the reporting of ambiguous
name occurrences.

	(the merge may require a little fiddling)

Make it compile on Alpha.

Add a comment that grabbing all the capabilities before GC causes the
thread->capability hash table to acquire multiple entries for the
current thread, which seems to be harmless.

Disable the assertion that checks for multiple entries with the same
key in the hash table.  It's useful to be able to do this sometimes.

MERGE TO STABLE

Fix a typechecker bug, which made the typechecker loop under certain
circumstances, notably when we have
	type Foo a = a
and try to unify
	b :=: Foo b

typecheck/should_compile/tc195 tests this case now.