GitLab

Splittable class comes from GlaExts now

PrelGHC ==> GHC.Prim, for what it's worth (this documentation is quite
out of date).

PrelGHC ==> GHC.Prim (comments only)

PrelGHC ==> GHC.Prim (in comments only)

PrelGHC ==> GHC.Prim

update mention of PrelGHC

Import wibbles

Take the old strictness analyser out of #ifdef DEBUG and put it
instead in #ifdef OLD_STRICTNESS.  DEBUG was getting a bit slow.

No need to compile Printf.lhs with -fvia-c now.

The foreign declaration for snprintf was wrong, because we need to
give the type of the C function after applying C's type promotion
rules.  Hence the Float argument should be a Double, because it
occurs in the vararg section of the function type.

Back out previous fix, it was wrong.

Misc cleanup: remove the iface pretty-printing style, and clean up
bits of StringBuffer that aren't required any more.

undo accidental commits

Documentation about type system extensions

Remove the interface file parser, and move .hi-boot parsing into the
main parser.  The syntax of .hi-boot files is now greatly improved in
terms of readability; here's an example:

	module M where
	data T
	f :: T -> GHC.Base.Int

note that
	(a) layout can be used
	(b) there's no explcit export list; everything declared
	    is implicitly exported
	(c) Z-encoding of names is no longer required
	(d) Any identifier not declared in the current module must
	    still be quailified with the module which originally
	    defined it (eg. GHC.Base.Int above).

We'd like to relax (d), but that will come later.

------------------------
	Change
		GlobalName --> ExternalName
		LocalName  ->  InternalName
	------------------------

For a long time there's been terminological confusion between

	GlobalName vs LocalName	 (property of a Name)
	GlobalId vs LocalId	 (property of an Id)

I've now changed the terminology for Name to be

	ExternalName vs InternalName

I've also added quite a bit of documentation in the Commentary.

Lots of stuff about external and internal names

Turn "return e" into "return $! e" in several places to improve
performance.

Remove -fall-strict

Update the documentation to reflect new commands since 5.02.2:

	:browse
	:set args
	:set prog
	:show bindings
	:show modules

and the new syntax for :module.

Fix the -fasm/-fvia-C auto-selection again: now we're back to the
proper behaviour, namely -fasm is the default unless you specify -O or
-fvia-C, but only on arch's that support it.

I screwed it up in the previous commit, but it exposed a bug in the
NCG so it wasn't all bad :)

Cope with VoidRep fields in a static constructor.

It *is* possible to get VoidRep fields in a constructor, but their
representation should be empty (not a zero word).

Fix panic message.

GHC version wibble

Fix a bug in the native code generator's handling of floating-point
arguments to ccalls: it wasn't correctly promoting float arguments to
doubles.  Amazingly this has gone undetected for quite a while.

Fixes bug #523996.

Forced commit: note that the previous revision also improved the error
message for a compiler stack overflow slightly.

eek!  Somehow, I managed to make -fvia-C the default on the HEAD and
nobody noticed for two months.

------------------------
	Fix a type-invariant bug
	------------------------

We need to call Type.mkGenTyConApp from Type.mkAppTy, in
case there's a partially applied type synonym.  Explanation
with Type.mkAppTy.  All part of GHC's rather liberal treatment
of type synonyms.

Shown up by a program from Ralf Laemmel:

	type Generic i o = forall x. i x -> o x
	type Id x = x
	comb :: Generic Id Id

Test is typecheck/should_compile/tc149.hs

Add a magic number to the beginning of interface files (0x1face :-) to
avoid trying to read old text-style interfaces as binary (the usual
result is an attempt to allocate a ByteArray larger than the available
memory size and an obscure crash).

If we are in a deadlock situation but there are signal handlers
installed, pause() instead of reporting deadlock.  One of the signal
handlers may allow the process to continue after being triggered.

Fixes bug #525772.

I've made a half-hearted attempt to clean up the deadlock code in the
scheduler, it now looks slightly better.

Main threads are now not kept alive artificially, so it is possible
for a main thread to be sent the BlockedOnDeadMVar exception.  Main
threads are no longer GC roots.

This involved cleaning up the weak pointer processing somewhat, and
separating the processing of real weak pointers from the processing of
the all_threads list (which can be thought of as "weaker pointers": a
finalizer can keep a blocked thread alive, but not vice-versa).  The
new story is described in a detailed comment in GC.c.

One interesting consequence is that it's much harder to get a Deadlock
exception now - many deadlock situations involving main threads will
turn into BlockedOnDeadMVar situations instead.  For example, if there
are a group of threads in a circular deadlock, then they will all be
sent BlockedOnDeadMVar simultaneously, whereas before if one of them
was the main thread it would be sent Deadlock.  It's really hard to
get Deadlock now - you have to somehow keep an MVar independently
reachable, eg. by using a StablePtr.

When deciding whether the source for a module is unchanged, don't look
at the file referred to in the argument of '-o' unless it really is
the object file for this compilation (it might be the binary instead).

MERGE TO STABLE

- Fix brokenness in foreign export.
- Remove some unused imports.

- Add quotes around invocation of ${WithGhc-ghc}, in case the pathname
  has spaces in (bug #528810).

- remove now-unused definition of is_gcc_v3.

-----------
	Fix mkSynTy
	-----------

Fix a nasty and long-lived bug in mkSynTy which meant that in:

	newtype A a = A a
	type B = A
	f :: B Double

the type (B Double) wasn't regarded properly as (A Double).  This bug
has lasted for a long time because the type inference engine is fairly
forgiving about missing the invariant that a TyConApp always looks
like one right at the top level.  In fact, it's so forgiving that I
don't know how to tickle this bug into showing up.  (It showed up in
5.02, becuase the context-simplification for 'deriving' is done by a
different engine as all the other context simplification.)

Still, the invariant should hold, and this fix makes it so.

Comments

import openFileEx and IOModeEx(..)

Add a warning about booting from .hc files on GHC 5.xx, and recommend
using 4.08.2 instead.

Re-instate ifeq "$(way)" "" around the rule for recursively descending
into SUBDIRS that I removed a while back.  This will fix a bug in
nofib whereby if NoFibWays is set to anything then we do O(n^2)
traversals of the tree, but it also means you can't sit at the top of
the tree and say 'make way=p' to just build profiling libraries (which
was the reason I made the change in the first place).  We'll have to
find another way to do this.