GitLab

We should really delete the main thread in a deadlock situation,
otherwise if we somehow recover it might stay in the system
indefinitely.

-------------------------------
	Correct a nasty lub bug
	-------------------------------

	MERGE WITH STABLE BRANCH

Peter Sestoft discovered that lub wasn't commutative.
We certainly want that

	Err `lub` Abs = Lazy

This is a Real Bug.

Use elemRdrEnv

Get rid off about half a dozen symbols from RTS_MINGW_ONLY_SYMBOLS;
some unnecessary, some redundant now that 'kernel32' is on the
default DLL search list for 'std'.

std & mingw32: include 'kernel32' in extra_libraries

[btw, this config file needs a bit of work to have it work on a
i386-unknown-cygwin plat.]

new commands:

	:set args <arg> ...
	:set prog <progname>

to set the results of System.getArgs and System.getProgName as seen by
the program.

-fwarn-name-shadowing should check the global env as well as the local
env for names that could be shadowed (the docs don't say anything
about it applying to local names only).

merge from stable, rev 1.60.2.9:

  Check for duplicate symbols when loading objects, and give up (with
  suitably helpful error msg) if duplicates are found.

oops, need to look for source files in mkdependHS mode too.

Make SimplGently a bit less gentle

Default verbosity mode is always 1, and we don't print the
"Compiling ( foo.hs, foo.o )" message if we're in OneShot mode.

Fixes recently introduced problem where the "compilation IS NOT
required" message isn't printed at all in OneShot mode.

-------------------------------
	Correct a nasty existential bug
	-------------------------------

	MERGE WITH STABLE BRANCH

Thanks to Volker Stolz for finding this existential bug.
Again, it's amazing that it hasn't shown up before.
GHC 5.02 allows this utterly bogus program to get past
the type checker

  data DS = forall a. C (a -> Int)

  call (C f) arg = f arg

The existential-tyvar-escape check was wrong. Easily fixed, though.

	tcfail099 now tests for this

Describe how to use path searching in GHCi.

Changes to the finder algorithm: in non-compilation-manager modes
(eg. ghc -c) we now search for .hi files only, because a source file
is no good to us.  In compilation manager modes, we now don't search
for .hi files in the home package at all.

------------------------------------
	Allow foralls in newtype-with-record
	------------------------------------

	MERGE WITH STABLE BRANCH

A simple fix to a longstanding omission.  The parser rejected


 newtype CPS a = CPS { unCPS :: forall ans . (a -> ans) -> ans }

because the production for "newconstr" said "type" instead of "ctype".

import CLabe (isAsmTemp) for sparc/alpha (I guess someone removed the
import after a warning from -fwarn-unused-imports, so I've put the
import inside the relevant #ifdef this time).

Typo.

Clean mkDerivedConstantsHdr and mkNativeHdr.

Try to get mkLocalOcc right

Fix codegen globalisation for -split-objs

More notes about InstallShield. This is now a first draft of the
InstallShield notes; probably some bits have been left out, and I need
to fix make install to bring reality closer to document.

Say a bit about the way InstallShield should work. It doesn't yet, as
I have to finish fixing make install for Windows.

merge from stable, rev 1.3.2.2:

Fix some mistakes and omissions, plus remove some caveats that seem no
longer to hold.

merge from stable, rev 1.105.4.1:

  When not compiling via C, catch Casms in the typecheck and reject
  them in a civilised way rather than having the various back ends barf.

merge from stable, revs:

  1.191.4.1 +2 -2      fptools/ghc/compiler/Makefile
  1.7.4.2   +38 -13    fptools/ghc/compiler/ghci/ByteCodeFFI.lhs
  1.58.4.2  +4 -3      fptools/ghc/compiler/ghci/ByteCodeGen.lhs
  1.25.4.1  +40 -10    fptools/ghc/compiler/ghci/ByteCodeLink.lhs

  Make the bytecode generation machinery print a helpful message if
  it has to give up due to lack of 64-bit support.

  Add various bits of supporting infrastructure for 64-bit values
  in the bytecode generator.  Making it all work is beyond the scope
  of a patchlevel release, so these are unused right now.


  1.25.4.2  +27 -7     fptools/ghc/compiler/ghci/ByteCodeLink.lhs

  Print a civilised and helpful error message if the bytecode linker
  should encounter a link failure.


  1.58.4.3  +6 -8      fptools/ghc/compiler/ghci/ByteCodeGen.lhs
  1.25.4.3  +1 -1      fptools/ghc/compiler/ghci/ByteCodeLink.lhs

  Also give civilised messages for interactive FFI link failures.


  1.25.4.4  +2 -1      fptools/ghc/compiler/ghci/ByteCodeLink.lhs

  Refine the runtime-link-failure msg a bit.

merge from stable revs:

  1.60.2.7  +29 -8     fptools/ghc/rts/Linker.c
  Teach the runtime linker to ignore stabs-style debugging info on x86-ELF.

  1.60.2.8  +15 -3     fptools/ghc/rts/Linker.c
  PEi386 fixes for ignoring stabs-style (GNU, -g) debugging info.

merge from stable revs:
  1.121.4.1 +7 -6      fptools/ghc/rts/GC.c
  1.9.4.1   +4 -1      fptools/ghc/rts/GCCompact.c
  1.17.4.1  +4 -3      fptools/ghc/rts/StoragePriv.h

  SimonM's fixes to deal with GHCi and CAFs properly in the compacting
  collector.

merge from stable rev 1.54.4.1:

  CLEAN_FILES += gmp/libgmp.a

----------------------------------------------
	The CoreTidy/CorePrep/CoreToStg saga continues
	[actually, this commit mostly completes the job]
	----------------------------------------------

			DO NOT MERGE!

* CorePrep injects implicit bindings, not the type checker,
  nor CgConTbls.   (This way, all the code generators see
  them, so no need to fiddle with the byte code generator.)

  As a result, all bindings in the module are for LocalIds,
  at least until CoreTidy.   This is a Big Win.

  Hence remove nasty isImplicitId test in update_bndr in
  SimplCore and DmdAnal

* hasNoBinding is no longer true of a dataConId (worker).
  There's an implicit curried binding for it.

* Remove yukky test in exprIsTrivial that did not regard
  a hasNoBinding Id as trivial; similarly in SimplUtils.tryEtaReduce

* In CoreTidy, get the names to avoid from the type env.
  That way it includes implicit bindings too.

* CoreTidy set the Arity of a top-level Id permanently;
  it's up to the rest of the compiler to respect it.
  Notably, CorePrep uses etaExpand to make the manifest arity
  match the claimed arity.

* As a result, nuke CgArity, so that CgInfo now contains only
  CafInfo.  The CafInfo is knot-tied as before.


Other things

* In Simplify.simplLazyBind, be a bit keener to float bindings
  out if it's a top-level binding.

Tidy up maybeGlobaliseId

Complete the removal of nullAddr# primop (dont merge)

Correct bug in todays bug-fix to DmdAnal

Add stuff about variables

Oops, don't try to load the rts or gmp packages at all.

Add a lightweight arena allocation scheme, and use it to speed up
allocation of cost centres and cost-centre stacks in the profiler.

Make catch work with older compilers, and avoid clash between
Exception.catch and PrelException's catch.

Use 'long long' types for the various memory allocation counters in
cost centres and cost centre stacks, as these are prone to overflowing.

Back out part of the previous commit that shouldn't have gone in, and
tidy up a little.

Fix two minor bugs in DmdAnal, and add comments.
The bugs were both in dmdTransform, the dataConId case

  * The test for saturation should be against call_depth
    not agaainst (length ds).

  * The arg_ds computation for k=Keep should be
    with 'both' not 'lub'.

Remove old rubbish and put some new in instead.