GitLab

Make ghc-pkg independent of hslibs

Correct comments

Yet another correction to earlier commit..
*now* those rank-2 pattern matches should work

		MERGE TO STABLE

Remove bogus warning

Tidy up error message for "funny global things"

Put debugging output in #ifdef DEBUG

MERGE TO STABLE

Remove redundant debug printing

Fix a(nother) bug in the new THUNK_SELECTOR code.  Hopefully should
fix the crashes seen in the last few nightly builds.

Add notes on how to build a .a file to the docs, as requested by Koen
Claessen.

Sigh! forgot one case!  MERGE TO STABLE

Cosmetics only

Correction to earlier commit.. *now* those rank-2 pattern matches should work

		MERGE TO STABLE

The body of a for-all should be of kind *

		MERGE TO STABLE

--------------------------------
	Fix rank-2 pattern-match failure
	--------------------------------

This fixes the failure when you have a rank-2 type sig
matching a data type pattern.  Thus

	data T a = T1 | T2 a

	f :: (forall x. T x) -> Int
	f T1 = ...

This crashes GHC 5.04

--------------------------------------
	Attach inline pragmas to class methods
	--------------------------------------

This fix makes INLINE pragmas on method bindings (in class
or instance decls) work properly.

It seems to have been hanging around in my tree for some time.
To be on the safe side, let's not merge this into 5.04.1, although
it should be fine (an an improvement).

Formatting only

Formatting only

Make sure we get the compiler's own Pretty library when we #include
../utils/ghc-pkg/Package.hs.

Urk, the 'import Pretty' in this module was a bit of a hack; when
compiled as part of the standalone ghc-pkg tool it referred to the
Pretty library from the util package, but when compiled as part of the
compiler it referred to the compiler's own Pretty module.  Hack around
this.

Build without hslibs on GHC >= 5.04

Build without hslibs on GHC >= 5.04

Build without hslibs on GHC >= 5.04

Tentative fix for the recent blkcnt_t trouble

Nuke the unsupported hslibs version of HOpenGL on the HEAD, too. Now
we are only left with two versions, which I think is OK: A stable one
(not in the fptools repository, but on my web site) and a cool, but
*very* incomplete one (currently only GLUT) for bleeding edge people
in the hierarchical libraries parts of the repository.

Disallow ForeignObj as well as ForeignPtr FFI arguments

Finally separate the compiler from hslibs.

Mainly import wibbles, and use the new POSIX library when
bootstrapping.

Partial rewrite of the POSIX library.

The main purpose of this sweep is to remove the last dependencies of
the compiler on hslibs.  When I've committed the associated compiler
changes, only the 'base' package will be required to bootstrap the
compiler.  Additionally to build GHCi, the 'readline' and 'unix'
packages will be required.

The new POSIX library lives mostly in libraries/unix, with a few bits
required for compiler bootstrapping in libraries/base.  The 'base'
package is mostly free of hsc2hs code to make bootstrapping from HC
files easier, but the 'unix' package will use hsc2hs liberally.

The old POSIX library continues to provide more-or-less the same
interface as before, although some of the types are more correct now
(previously lots of POSIX types were just mapped to Int).  The new
interface is largely the same as the old, except that some new
functionality from the latest POSIX spec has been added (eg. symbolic
links).

So far, the new POSIX library has signal support, directory/file
operations and lots of stuff from unistd.h.  The module names are:

  System.Posix
	The main dude, exports everything

  System.Posix.Types
	All the POSIX types, using the same naming scheme as
        Foreign.C.Types, Eg. CUid, COff, etc.  Many of these types
        were previously exported by GHC.Posix.

        Additionally exports the "nicer" names used by the old POSIX
	library for compatibility (eg. ProcessID == CPid, FileMode ==
	CMode, etc.)

	All reasonable instances are derived for these types.

  System.Posix.Signals
	Signal support, contains most of which was in PosixProcPrim before.
	The RTS interface to the signal handling support has been
	rationalised slightly.

  System.Posix.Directory
 	Directory support, most were in PosixFiles before.

  System.Posix.Files
	File operations, most were in PosixFiles before.

  System.Posix.Unistd
	(for want of a better name) Miscellaneous bits that mostly come
	from the unistd.h header file.  PosixProcEnv before.

The rest of the library should pan out like so:

  System.Posix.IO
  System.Posix.Error   (maybe)
  System.Posix.Process
  System.Posix.Terminal

(I've no doubt broken Win32 support, but I'm checking the build at the moment).

Remove ForeignObj and ForeignPtr keys

Remove foreignPtrTyCon and foreignObjTyCon

Disallow ForeignPtr as an FFI argument type.

Disallow 'ForeignPtr' as an FFI argument type.  It has been deprecated
for some time (withForeignPtr is the approved way to pass a
ForeignPtr).  The changes I'm about to make to ForeignPtr will stop it
working anyhow.

Selector Thunk Fix, take II.

The previous version didn't deal well with selector thunks which point
to more selector thunks, and on closer inspection the method was
flawed.  Now I've introduced a function

	StgClosure *eval_selector_thunk( int field, StgClosure * )

which evaluates a selector thunk returning its value, in from-space,
if possible.  It blackholes the thunk during evaluation.  It might
recursively evaluate more selector thunks, but it does this in a
bounded way and updates the thunks with indirections (NOT forwarding
pointers) after evaluation.

This cleans things up somewhat, and I believe it deals properly with
both types of selector-thunk loops that arise.

MERGE TO STABLE

perl made uglier to work around the perl 5.7/5.8 bug documented at
http://bugs6.perl.org/rt2/Ticket/Display.html?id=1760 and illustrated
by the seg fault of perl -e '("x\n" x 5000) =~ /(.*\n)+/'

MERGE TO STABLE

Fix for infinite loop when there is a THUNK_SELECTOR which eventually
refers to itself, such as might be generated by code like

	let x = (fst x, snd x) in ...

At the same time, I re-enabled the code to traverse multiple selector
thunks with bounded depth, because I believe it now works.

MERGE TO STABLE (but test thoroughly in the HEAD first, this is
fragile stuff)

In code style, print negative floating point literals in parentheses
to avoid interacting with surrounding syntax.

Fixes SourceForge bug #604849

MERGE TO STABLE

Remove RtsAPIDeprec.c, since this is causing grief.  The upshot (I
think) is that you won't be able to do foreign import "wrapper" with
an Addr in the type.

Disallow 'foreign import stdcall "wrapper"' when compiling via the
NCG.

MERGE TO STABLE

Import fixups

Exception => Control.Exception

Recent changes to simplify PrimRep had introduced a bug: the heap
check code was assuming that anything with PtrRep representation was
enterable.  This isn't the case for the unpointed primitive types
(eg. ByteArray#), resulting in the ARR_WORDS crash in last night's
build.

This bug isn't in STABLE.