GitLab

Make dmdFix not loop forever

Fix compilation with -DDEBUG.

MERGE TO STABLE

Improve the error message for duplicate or invalid binders
in a binding group.

	[Consequence: rnfail034 should not be an expected failure
		      any more]

	MERGE WITH STABLE BRANCH

----------------------------------------
	Make dict funs and default methods
	into LocalIds only at their binding site
	----------------------------------------
        [part of 3 related commits]

There's a long comment about this with MkId.mkDefaultMethodId,
which I reproduce below.

While I was at it, I renamed setIdNoDiscard to setIdLocalExported.
Which is hardly an improvement, I'm afraid.  This renaming touches
	Var.lhs, Id.lhs, SimplCore.lhs
in a trivial way.

	---------------------

Dict funs and default methods are *not* ImplicitIds.  Their definition
involves user-written code, so we can't figure out their strictness etc
based on fixed info, as we can for constructors and record selectors (say).

We build them as GlobalIds, but when in the module where they are
bound, we turn the Id at the *binding site* into an exported LocalId.
This ensures that they are taken to account by free-variable finding
and dependency analysis (e.g. CoreFVs.exprFreeVars).   The simplifier
will propagate the LocalId to all occurrence sites.

Why shouldn't they be bound as GlobalIds?  Because, in particular, if
they are globals, the specialiser floats dict uses above their defns,
which prevents good simplifications happening.  Also the strictness
analyser treats a occurrence of a GlobalId as imported and assumes it
contains strictness in its IdInfo, which isn't true if the thing is
bound in the same module as the occurrence.

It's OK for dfuns to be LocalIds, because we form the instance-env to
pass on to the next module (md_insts) in CoreTidy, afer tidying
and globalising the top-level Ids.

BUT make sure they are *exported* LocalIds (setIdLocalExported) so
that they aren't discarded by the occurrence analyser.

-----------------------------------
	Pin on accurate strictness info for
	record and dictionary selectors
	-----------------------------------
        [part of 3 related commits]

This fixes a long-standing infelicity.  Sometimes selectors aren't
inlined until after strictness analysis, so if we don't have decent
strictness info on them we get bad strictness results.

For record selectors, the unboxing-strict-fields stuff makes it hard
to figurwe out the correct strictness, so we just invoke the demand
analyser to work it out.

------------------------
	Fix the demand analyser
	------------------------

A spiffy new domain for demands, and definitions for lub/both
which are actually monotonic.   Quite a bit of related jiggling
around.

One of the original motivations was to do with functions like:

	sum n []     = n
	sum n (x:xs) = sum (n+x) xs

Even though n is returned boxed from the first case, we don't want
to get strictness
	S(L)V -> T
because that means we pass the box for n, and that is TERRIBLE.
So the new version errs on the side of unboxing, more like the forwards
analyser, and only passes the box if it is *definitely* needed, rather
than if it *may* be needed.

Omit unnecessary import

-------------------
	Newtypes and ccalls
	     [addendum]
	-------------------

	MERGE WITH STABLE BRANCH

I accidentally omitted these two wibbles from my previous commit.
I've added PrelNames.unitTyConKey, and used it in TcType and DsCCall.

-------------------
	Newtypes and ccalls
	-------------------

	MERGE WITH STABLE BRANCH

Yet another bit of newtype-squashing that hadn't been
synced with reality.  In desugaring ccalls, we can still
see newtypes, if they are recursive, and we must generate
appropriate coerces.

Fixes a bug in cg011.

Fix the code to *really* do the following:

On the Alpha we can only handle <= 4 integer arguments with foreign export
dynamic.  Following the example of the corresponding Sparc hack, we detect
when we're being asked to do something we can't and refuse.

MERGE TO STABLE BRANCH

document -fext-core flag and set pointers to ext-core docs/tools (MERGE to STABLE pleeeeeease.)

Import wibbles

Change a couple more Int32s into CTimes, which they are really.

Convert some CInts into CTimes.

MERGE TO STABLE.

add pointer to primops.txt documentation feature

change ext-core output file suffix from .core (too easy to delete accidentally) to .hcr (MERGE TO STABLE)

Fix cpp-inspired sparc breakage.  Sigh.

Fix 32-bit breakage.  Int32 is the wrong thing; CInt is probably
right.  I assume that Alpha is LP64, so that on Alpha
   sizeof(int) = 32, sizeof(long) = 64
and that these are both sized at 32 for 32-bit platforms.

Replace CLong with CInt32 to fix 32-bit assumption. MERGE TO STABLE BRANCH

THIS CHANGE AFFECTS ALL OBJECT FILES COMPILED FROM HASKELL.
Please say "make -C ghc/lib/std clean; make -C hslibs clean".

This commit eliminates spurious warning messages when compiling on
the Alpha.  There are two kinds of spurious warning messages:

(1) gcc: -noprefix_recognition: linker input file unused since linking not done

    This warning is because we pass the flag "-Xlinker -noprefix_recognition"
    to gcc.  We remove this warning by no longer passing the flag to gcc,
    and by removing the reason we were passing the flag in the first place:
    __init_* is now renamed to __stginit_*.

(2) .../includes/Regs.h: warning: call-clobbered register used for global
    register variable

    This warning and all other warnings except (1), we eliminate by
    passing the -w flag to gcc.

MERGE TO STABLE BRANCH

On the Alpha we can only handle <= 4 integer arguments with foreign export
dynamic.  Following the example of the corresponding Sparc hack, we detect
when we're being asked to do something we can't and refuse.

MERGE TO STABLE BRANCH

Fix ELF compilation for new section table scheme.

Build system hacks to split HSwin32.o into two parts, so that it
can be loaded into GHCi.  Uses the same gruesome hacks as HSstd.o.

Further linker fixes.  Record .bss space allocated, for the benefit of
LOOKS_LIKE_GHC_INFO etc tests done by the GC.  In this commit, for
Win32 only.

Also some minor tweaks so that -package win32 can be loaded into GHCi.

Just as we do on sparc:
On alpha, create code to shuffle a0 .. a3 into a2 .. a5 before the
old code.  This is a GHASTLY HACK which at least makes f-x-dynamic on
alpha work for callees with up to 32 bytes (4 words) of args.

Clean core, not *core (because of ext-core dirs).

add ext-core example programs -- MERGE TO STABLE (surely harmless?)

add documentation for ext-core -- MERGE TO STABLE (surely harmless?)

fix documentation typos

Fix worker-wrapper generation.  See commments in WwLib.mk_ww_str

Fix Linux compile b0rkage.  Darn.

MERGE TO STABLE BRANCH

Pass -w to gcc for .hc compilation on sparc-solaris so as to suppress
warnings about call-clobbered regs being used for global vars.

Reapplied my "FCode as a monad" patch, now that 5.02 has forked into
a separate branch.

I'm fairly sure that this doesn't change the behaviour of anything.

.bss section fixes for ELF.

MERGE TO STABLE BRANCH (and test better!)

Properly handle the linker debugging (-D4096) before.

MERGE TO STABLE BRANCH

Remove a spurious space. Oh, yes.

Advance HEAD version to 5.03.

Don't automatically enable compaction when generations == 1.

Automatically disable compaction when there's only one generation
(-G1) and print a warning to that effect.

Back out recent changes to the code generator as too destabilising.
Revert files as follows:

revert to 1.35   CgBindery.lhs
revert to 1.26   CgMonad.lhs
revert to 1.15   CgStackery.lhs
revert to 1.10   CgUsages.lhs