GitLab

update aLL_NON_PTRS for the new RET_DYN stack frame layout.

Unbreak 2nd stage

Make a leading '~' expand to $HOME in commands that load source files
(:load, :add) in addition to :cd which already had this behaviour.

Order declarations in reifications in order of source line number.
The bugs still there but it bites less often now...

Also remove the type parameterisation and do some type renaming as
discussed on the template-haskell list.

Fix the context for derived Typeable instances

Stage-2 wibbles to the Expected type changes

----------------------------------
	Implement Typeable properly
	----------------------------------

1.  Add 'deriving' for Typeable class. So you can say

	data T a b = .... deriving( Typeable )

    At the moment you only get this if you ask for it. If you say
    nothing you get nothing.

2.  Implement Typeable better, with proper O(1) comparison of
    type representations

3.  Add the 'cast' operation described in 'Scrap your boilerplate'
    and use it.


4.  Consequence: need to move the definition of IOArray from
    Data.Array.IO.Internals to GHC.IOBase, where it joins IORef.
    This is necssary so that HashTable can be low down in the compilation
    hierarchy, and hence so can Dynamic.



	WARNING: I'm not certain the imports in HashTable and Dynamic
		 will all be right for Hugs and NHC. I hope you can
	  	 fix them up.

Comments

Comments

Update hi-boot files

----------------------------------
 Use the Infer/Check idea for typechecking higher-rank types
	----------------------------------

The main idea is that

	data Expected ty = Infer (TcRef ty) | Check ty

	tcMonoExpr :: Expr -> Expected TcRhoType -> TcM Expra


This "Expected" type tells tcMonoExpr whether it's doing inference or
checking.  It replaces the "HoleTv" flavour of type variable.

This actually leads to slightly more lines of code, but it's much
clearer, and the new type distinctions showed up several subtle bugs
in the previous implementation.  It all arose out of writing the
prototype implementation for the paper.

Error messages wibble around a little bit.  I'm not quite certain why!  But the
changes look like improvements to me.

Typo :-}

Unbreak 2nd stage: GHC.Posix => System.Posix.Internals

The Id for a foreign import should be a LocalId, not a GlobalId.
Making it a GlobalId violates the invariant that all Ids defined in
the current module should be LocalIds until CoreTidy.

Add an ASSERT to confirm an invariant: all Ids defined in this module
should be LocalIds until the CoreTidy phase.

More simplifier wibbles to do with the arity transfer stuff

Wibble to arity fix

A subtle type checker bug, which showed up in rebindable1 test

Handle Int literals in TH

----------------------------------
       Fix a long-standing eta-reduction bug
	----------------------------------

Consider the stupid definition

	f = \x -> f x

We were erroneously eta-reducing this to

	f = f

(unsound because they'd be distinguishable by `seq`)

The reason was that simplLazyBind was exposing the arity of
a recursive function to its own RHS, when all it was really
trying to do was expose the *rules* for the function.

Easily fixed.   This fixes some

	"Bad eta expand"

warnings.  Good all round.  In particular, fixes rn006.

comments

Comments

-------------------------------------
      Fix a functional-dependency-related bug in
		tcSimpifyRestricted
	-------------------------------------

	MERGE TO STABLE if it goes over conveniently
		(but I rather think it may not)

tcSimplifyRestricted works by (a) simplifying brutall to find out
what the constrained type variables are, and (b) simplifying more
gently, knowing the constrained type varaibles.  The bug is that
in step (b) we were not doing the check-for-improvement-and-loop
part, thinking that step (a) had alrady done all the improvement.
But not so, as an example in the code now shows.

Simple to fix.  I rather think we could tidy up these various loops.

Comments

----------------------------------
   Try swapping order of inference for applications
	----------------------------------

For a long time we've inferred the argument types of a function
call before using information from the expected type passed in.
This commit swaps the order.  See comments with tcApp for an
example where this makes a real difference.

I don't think any error messages get signficantly worse, but yell
if you find any.

Better error messages

ToDo comments

Improve error messages on newtype deriving

Fix unbox-strict-fields deriving bug; MERGE TO STABLE

Typecheck type reification brackets [t| ... |]

Have Literal.Literal support the representation of NULL pointers only,
and not arbitrary pointer values.

(MachAddr <some-pointer-value-as-an-Integer>) wasn't being used,
except to handle nullAddr#. It (MachAddr) is a potential source of
problems should the compiler start doing constant folding or other
interesting operations over MachAddrs (think: interface files +
cross-compilation), so we might as well scale back the representation
of raw pointer values.

Don't constant-fold (negateFloat# 0.0#), because the compiler's
internal representation of floating-point literals (Rational) can't
represent -0.0.  This means that

   main = print (-0.0)

now gives the same results with -O as it does without.

Fixes test arith005.

Comments only

Submitted by: Urban Boquist <boquist@crt.se>

Off-by-one tidyup.

ALLOC_AP, ALLOC_PAP and MKAP were all being constructed
with size arguments equal to (1+number of args/FVs) in
ByteCodeGen.schemeE, only for Interpreter.c to subtract 1
when fishing out the payloads. This commit drops the
up-and-downery.

Simplification spotted by Andy Moran

NCG support for f.e.d. stdcall -- Literal.MachLabels now optionally carry
the size (in bytes) of the stack frame it expects, if known. That just
so happens to match what stdcall labels need to be annotated with when
emitting them in the NCG..

-------------------------------------
      Fix a wibble in default-type setting for TH
	-------------------------------------

When a bunch of declarations contains no 'default' declaration,
we were setting 'defaultDefaultTys' as appropriate default types.
This isn't right for Template Haskell, which may have more than
one bunch of top-level decls.  (The bunches are separated by
top-level declaration splices.)

Instead, if there is no 'default' declaration we should do nothing.

-------------------------------------
      Respect SPECIALISE pragmas in instance decls
	-------------------------------------

For some time now we have simply been discarding SPECIALISE pragmas
in instance declarations.  I think this was my fault, at some point
when I was re-plumbing TcClassDcl.lhs, but it's been this way for
some time.

The only uses of this facility in the Prelude are in GHC/Float.lhs
and GHC/Real.lhs, which affected the efficiency of the Float and
Double instance of properFraction, floor, ceiling, etc.

Ah well, it's fixed now.

Reduce tctrace verbosity

Import wibbles

Update the out-of-memory error message: the request size isn't known
any more.