ghc-hetmet.git
19 years ago[project @ 2005-01-04 13:41:53 by simonmar]
simonmar [Tue, 4 Jan 2005 13:41:54 +0000 (13:41 +0000)]
[project @ 2005-01-04 13:41:53 by simonmar]
ghc/mk/config.mk was never being included in ordinary Makefiles.  It
was only included in the top-level fptools/Makefile for the purposes
of obtaining binary distribution settings.

This fixes that problem, and now as a side-effect $(GhcHasReadline)
will start working again.

19 years ago[project @ 2005-01-04 10:26:34 by simonpj]
simonpj [Tue, 4 Jan 2005 10:26:35 +0000 (10:26 +0000)]
[project @ 2005-01-04 10:26:34 by simonpj]
Import trimming

19 years ago[project @ 2005-01-04 09:19:50 by simonpj]
simonpj [Tue, 4 Jan 2005 09:19:50 +0000 (09:19 +0000)]
[project @ 2005-01-04 09:19:50 by simonpj]
Remove redundant Subst hi-boot files

19 years ago[project @ 2004-12-30 22:14:59 by simonpj]
simonpj [Thu, 30 Dec 2004 22:15:19 +0000 (22:15 +0000)]
[project @ 2004-12-30 22:14:59 by simonpj]
Fix to the pre-Xmas simplifier changes, which should make
everything work again.  I'd forgotten to attend to this
corner.  Still not properly tested I fear.

Also remove dead code from SimplEnv, and simplify the remainder (hooray).

19 years ago[project @ 2004-12-27 18:28:21 by simonpj]
simonpj [Mon, 27 Dec 2004 18:28:21 +0000 (18:28 +0000)]
[project @ 2004-12-27 18:28:21 by simonpj]
Wibble

19 years ago[project @ 2004-12-27 11:27:56 by simonpj]
simonpj [Mon, 27 Dec 2004 11:27:56 +0000 (11:27 +0000)]
[project @ 2004-12-27 11:27:56 by simonpj]
import wibble

19 years ago[project @ 2004-12-24 16:14:36 by simonpj]
simonpj [Fri, 24 Dec 2004 16:15:15 +0000 (16:15 +0000)]
[project @ 2004-12-24 16:14:36 by simonpj]
---------------------------
          Refactor the simplifier
   ---------------------------

Driven by a GADT bug, I have refactored the simpifier, and the way GHC
treats substitutions.  I hope I have gotten it right.  Be cautious about updating.

* coreSyn/Subst.lhs has gone

* coreSyn/CoreSubst replaces it, except that it's quite a bit simpler

* simplCore/SimplEnv is added, and contains the simplifier-specific substitution
  stuff

Previously Subst was trying to be all things to all men, and that was making
it Too Complicated.

There may be a little more code now, but it's much easier to understand.

19 years ago[project @ 2004-12-24 11:38:09 by simonpj]
simonpj [Fri, 24 Dec 2004 11:38:20 +0000 (11:38 +0000)]
[project @ 2004-12-24 11:38:09 by simonpj]
Reset the export flag for the new bindings in LiberateCase

19 years ago[project @ 2004-12-24 11:02:39 by simonpj]
simonpj [Fri, 24 Dec 2004 11:03:06 +0000 (11:03 +0000)]
[project @ 2004-12-24 11:02:39 by simonpj]
Further wibbles to the scoped-tyvar story.

This commit tidies up the ATyVar in TcTyThing, making it
ATyVar Name Type
instead of the previous misleading
ATyVar TyVar Type

But the main thing is that we must take care with definitions
like this:

type T a = forall b. b -> (a,b)

f :: forall c. T c
f = ...

Here, we want only 'c' to scope over the RHS of f.  The renamer ensures
that... but we must also take care that we freshly instantiate the
expanded type signature (forall c b. b -> (c,b)) before checking f's RHS,
so that we don't get false sharing between uses of T.

19 years ago[project @ 2004-12-23 14:59:46 by simonpj]
simonpj [Thu, 23 Dec 2004 14:59:50 +0000 (14:59 +0000)]
[project @ 2004-12-23 14:59:46 by simonpj]
Simplifications, dead code elimination

19 years ago[project @ 2004-12-23 13:44:06 by simonpj]
simonpj [Thu, 23 Dec 2004 13:44:10 +0000 (13:44 +0000)]
[project @ 2004-12-23 13:44:06 by simonpj]
minor nomenclature wibble

19 years ago[project @ 2004-12-23 11:50:55 by simonpj]
simonpj [Thu, 23 Dec 2004 11:50:57 +0000 (11:50 +0000)]
[project @ 2004-12-23 11:50:55 by simonpj]
Enable scoped type variables only where there is an explicit forall

19 years ago[project @ 2004-12-23 09:07:30 by simonpj]
simonpj [Thu, 23 Dec 2004 09:07:39 +0000 (09:07 +0000)]
[project @ 2004-12-23 09:07:30 by simonpj]
---------------------------------
          Template Haskell: names again
   ---------------------------------

On 2 Dec 04 I made this commit (1.58 in Convert.lhs)

    Fix a Template Haskell bug that meant that top-level names created
    with newName were not made properly unique.

But that just introduced a new bug!  THe trouble is that names created by
newName are NameUs; but I was *also* using NameU for names of free varaibles,
such as the 'x' in the quoted code here
f x = $( g [| \y -> (x,y) |])

But when converting to HsSyn, the x and y must be treated diffferently.
The 'x' must convert to an Exact RdrName, so that it binds to the 'x' that's
in the type environment; but the 'y' must generate a nice unique RdrName.

So this commit adds NameL for the lexically-scoped bindings like 'x'.

19 years ago[project @ 2004-12-22 16:58:34 by simonpj]
simonpj [Wed, 22 Dec 2004 16:59:22 +0000 (16:59 +0000)]
[project @ 2004-12-22 16:58:34 by simonpj]
----------------------------------------
     Add more scoped type variables
----------------------------------------

Now the top-level forall'd variables of a type signature scope
over the right hand side of that function.

f :: a -> a
f x = ....

The type variable 'a' is in scope in the RHS, and in f's patterns.

It's implied by -fglasgow-exts, but can also be switched off independently
using -fscoped-type-variables (and the -fno variant)

19 years ago[project @ 2004-12-22 12:06:13 by simonpj]
simonpj [Wed, 22 Dec 2004 12:07:41 +0000 (12:07 +0000)]
[project @ 2004-12-22 12:06:13 by simonpj]
----------------------------------------
     New Core invariant: keep case alternatives in sorted order
----------------------------------------

We now keep the alternatives of a Case in the Core language in sorted
order.  Sorted, that is,
by constructor tag for DataAlt
by literal for LitAlt

The main reason is that it makes matching and equality testing more robust.
But in fact some lines of code vanished from SimplUtils.mkAlts.

WARNING: no change to interface file formats, but you'll need to recompile
your libraries so that they generate interface files that respect the
invariant.

19 years ago[project @ 2004-12-22 12:04:14 by simonpj]
simonpj [Wed, 22 Dec 2004 12:04:24 +0000 (12:04 +0000)]
[project @ 2004-12-22 12:04:14 by simonpj]
--------------------------
Add -fwarn-orphans flag
--------------------------

This gives a decent report for modules that contain 'orphan' instance and
rule declarations.  These are to be avoided, because GHC has to proactively
read the interface file every single time, just in case the instance/rule is
needed.

The flag just gives a convenient way of identifying the culprits.

19 years ago[project @ 2004-12-21 17:08:59 by simonpj]
simonpj [Tue, 21 Dec 2004 17:09:02 +0000 (17:09 +0000)]
[project @ 2004-12-21 17:08:59 by simonpj]
---------------------------------
     Template Haskell: dynamically scoped qualified names
---------------------------------

This commit adds a constructor to TH.Name, so that

nameBase (mkName "Foo.baz")    == "baz"
nameModule (MkName "Foo.baz") == "Foo"

We always did parse the module name off the front, but it used to
be done in hsSyn/Convert, but now it's done in TH.Syntax, which is
a better place.

19 years ago[project @ 2004-12-21 16:23:23 by simonpj]
simonpj [Tue, 21 Dec 2004 16:23:23 +0000 (16:23 +0000)]
[project @ 2004-12-21 16:23:23 by simonpj]
Fix buglet in CSE

19 years ago[project @ 2004-12-21 12:35:10 by simonpj]
simonpj [Tue, 21 Dec 2004 12:35:10 +0000 (12:35 +0000)]
[project @ 2004-12-21 12:35:10 by simonpj]
An old commit about paths on windows

19 years ago[project @ 2004-12-21 12:22:22 by simonpj]
simonpj [Tue, 21 Dec 2004 12:23:03 +0000 (12:23 +0000)]
[project @ 2004-12-21 12:22:22 by simonpj]
---------------------------------
     Improve handling of lexically scoped type variables
---------------------------------

If we have

f :: T a -> a
f (x :: T b) = ...

then the lexically scoped variable 'b' should refer to the rigid
type variable 'a', without any intervening wobbliness.  Previously
the in-scope type variables were always mutable TyVars, which were
instantatiated to point to the type they were bound to; but since
the advent of GADTs the intervening mutable type variable is a bad
thing.

Hence
  * In the type environment, ATyVar now carries a type
  * The call to refineTyVars in tc_pat on SigPatIn
    finds the types by matching
  * Then tcExtendTyVarEnv3 extends the type envt appropriately

Rater a lot of huff and puff, but it's quite natural for ATyVar
to contain a type.

Various other small nomenclature changes along the way.

19 years ago[project @ 2004-12-21 12:21:16 by simonpj]
simonpj [Tue, 21 Dec 2004 12:21:16 +0000 (12:21 +0000)]
[project @ 2004-12-21 12:21:16 by simonpj]
Comments

19 years ago[project @ 2004-12-21 12:14:31 by simonpj]
simonpj [Tue, 21 Dec 2004 12:14:31 +0000 (12:14 +0000)]
[project @ 2004-12-21 12:14:31 by simonpj]
Remove debug output

19 years ago[project @ 2004-12-21 12:11:37 by simonpj]
simonpj [Tue, 21 Dec 2004 12:11:37 +0000 (12:11 +0000)]
[project @ 2004-12-21 12:11:37 by simonpj]
Comments

19 years ago[project @ 2004-12-21 12:09:55 by simonpj]
simonpj [Tue, 21 Dec 2004 12:09:55 +0000 (12:09 +0000)]
[project @ 2004-12-21 12:09:55 by simonpj]
Add missing NoteTy cases to unify_

19 years ago[project @ 2004-12-21 12:09:14 by simonpj]
simonpj [Tue, 21 Dec 2004 12:09:14 +0000 (12:09 +0000)]
[project @ 2004-12-21 12:09:14 by simonpj]
Comments only

19 years ago[project @ 2004-12-21 09:08:08 by simonpj]
simonpj [Tue, 21 Dec 2004 09:08:08 +0000 (09:08 +0000)]
[project @ 2004-12-21 09:08:08 by simonpj]
Fix bogon in type comparison

19 years ago[project @ 2004-12-21 08:52:54 by simonpj]
simonpj [Tue, 21 Dec 2004 08:52:54 +0000 (08:52 +0000)]
[project @ 2004-12-21 08:52:54 by simonpj]
Comment only

19 years ago[project @ 2004-12-21 08:50:27 by simonpj]
simonpj [Tue, 21 Dec 2004 08:50:27 +0000 (08:50 +0000)]
[project @ 2004-12-21 08:50:27 by simonpj]
Another wibble to the Panic change; this time on Windows

19 years ago[project @ 2004-12-20 17:16:24 by simonpj]
simonpj [Mon, 20 Dec 2004 17:17:10 +0000 (17:17 +0000)]
[project @ 2004-12-20 17:16:24 by simonpj]
--------------------------------
Deal properly with dual-renaming
--------------------------------

When comparing types and terms, and during matching, we are faced
with
\x.e1 ~   \y.e2

There are many pitfalls here, and GHC has never done the job properly.
Now, at last it does, using a new abstraction VarEnv.RnEnv2.  See
comments there for how it works.

There are lots of consequential changes to use the new stuff, especially
in
types/Type (type comparison),
types/Unify (matching on types)
coreSyn/CoreUtils (equality on expressions),
specialise/Rules (matching).

I'm not 100% certain of that I've covered all the bases, so let me
know if something unexpected happens after you update.  Maybe wait until
a nightly build has worked ok first!

19 years ago[project @ 2004-12-18 15:45:56 by panne]
panne [Sat, 18 Dec 2004 15:45:56 +0000 (15:45 +0000)]
[project @ 2004-12-18 15:45:56 by panne]
The documentation breakage continues... >:-( Fixed DocBook. Again: "make html"
(or the faster "make validate" in the case of DocBook XML) is your friend! >:-(

19 years ago[project @ 2004-12-17 17:43:24 by simonpj]
simonpj [Fri, 17 Dec 2004 17:43:24 +0000 (17:43 +0000)]
[project @ 2004-12-17 17:43:24 by simonpj]
Restore a test; fixes rnfail042

19 years ago[project @ 2004-12-17 16:13:27 by simonpj]
simonpj [Fri, 17 Dec 2004 16:13:27 +0000 (16:13 +0000)]
[project @ 2004-12-17 16:13:27 by simonpj]
Fix breakage in previous commit re signal handling

19 years ago[project @ 2004-12-17 15:26:07 by simonmar]
simonmar [Fri, 17 Dec 2004 15:26:07 +0000 (15:26 +0000)]
[project @ 2004-12-17 15:26:07 by simonmar]
Add entry about Ctrl-C on Windows

19 years ago[project @ 2004-12-17 15:19:53 by simonmar]
simonmar [Fri, 17 Dec 2004 15:19:53 +0000 (15:19 +0000)]
[project @ 2004-12-17 15:19:53 by simonmar]
Add comment about Ctrl-C/Ctrl-Break issue

19 years ago[project @ 2004-12-17 15:14:47 by simonmar]
simonmar [Fri, 17 Dec 2004 15:14:47 +0000 (15:14 +0000)]
[project @ 2004-12-17 15:14:47 by simonmar]
Support for Ctrl-C on Windows.

This works fine in a cmd.exe shell, but in a Cygwin bash shell you
have to press Ctrl-Break to get an interrupt.  There appears to be a
long thread on the cygwin mailing list back in 2002 about this, but
life is probably too short to investigate further.

  http://www.cygwin.com/ml/cygwin/2002-01/msg00163.html

19 years ago[project @ 2004-12-16 10:36:57 by simonmar]
simonmar [Thu, 16 Dec 2004 10:36:59 +0000 (10:36 +0000)]
[project @ 2004-12-16 10:36:57 by simonmar]
Sanity check when configuring for mingw32: make sure that $CC is a
mingw gcc before proceeding.

19 years ago[project @ 2004-12-16 10:23:44 by simonmar]
simonmar [Thu, 16 Dec 2004 10:23:44 +0000 (10:23 +0000)]
[project @ 2004-12-16 10:23:44 by simonmar]
--with-gcc: export the CC environment variable, so the setting gets
picked up by sub-configures.  Perhaps we should be advising people to use

  CC=c:/mingw/bin/gcc ./configure

because that works for all configure scripts, not just the top level one.

Background:

 - We want --with-gcc to do the right thing, because we don't want
   to require Cygwin users to put c:/mingw/bin first on their
   path: that would break their Cygwin environment.

 - The build system should work with *no* gcc on your PATH, as long
   as you use --with-gcc (equivalently, CC=... ./configure).

19 years ago[project @ 2004-12-16 08:56:15 by simonpj]
simonpj [Thu, 16 Dec 2004 08:56:15 +0000 (08:56 +0000)]
[project @ 2004-12-16 08:56:15 by simonpj]
Darn!  Forgot this commit, which killed the nightly build

19 years ago[project @ 2004-12-15 15:28:39 by simonpj]
simonpj [Wed, 15 Dec 2004 15:28:39 +0000 (15:28 +0000)]
[project @ 2004-12-15 15:28:39 by simonpj]
Comments only

19 years ago[project @ 2004-12-15 12:51:15 by simonpj]
simonpj [Wed, 15 Dec 2004 12:51:16 +0000 (12:51 +0000)]
[project @ 2004-12-15 12:51:15 by simonpj]
Make ghc/lib/compat/Compat/Directory.hs use the C function
__compat_long_path_size, rather than
__hscore_long_path_size, as the libraries/ version does
And make ghc/lib/compat/cbits/directory.c define it.

In this way we avoid spurious duplicate-symbol errors when we
compile GHC with ghc6.2.1 etc.

19 years ago[project @ 2004-12-15 12:15:16 by simonpj]
simonpj [Wed, 15 Dec 2004 12:15:16 +0000 (12:15 +0000)]
[project @ 2004-12-15 12:15:16 by simonpj]
Tidy the type of an interactively-bound Id before adding
it to the environment.  Generally speaking, top level
environments are assumed to be in tidied form, so we
don't tidy the types before printing.

19 years ago[project @ 2004-12-15 12:14:06 by simonpj]
simonpj [Wed, 15 Dec 2004 12:14:06 +0000 (12:14 +0000)]
[project @ 2004-12-15 12:14:06 by simonpj]
comment

19 years ago[project @ 2004-12-13 12:09:25 by simonpj]
simonpj [Mon, 13 Dec 2004 12:09:25 +0000 (12:09 +0000)]
[project @ 2004-12-13 12:09:25 by simonpj]
Comments only

19 years ago[project @ 2004-12-10 13:34:38 by simonmar]
simonmar [Fri, 10 Dec 2004 13:34:38 +0000 (13:34 +0000)]
[project @ 2004-12-10 13:34:38 by simonmar]
stg_sel_##offset##_upd_entry: the ENTER() here is unnecessary, because
we know the closure in R1 will evaluate to a constructor, so we can
save some time by just entering it directly.

19 years ago[project @ 2004-12-08 15:09:14 by simonpj]
simonpj [Wed, 8 Dec 2004 15:09:14 +0000 (15:09 +0000)]
[project @ 2004-12-08 15:09:14 by simonpj]
Undo bogus 1.4 commit

19 years ago[project @ 2004-12-08 15:03:50 by simonpj]
simonpj [Wed, 8 Dec 2004 15:03:50 +0000 (15:03 +0000)]
[project @ 2004-12-08 15:03:50 by simonpj]
First commit for new Cmm code generation (branch)

19 years ago[project @ 2004-12-08 14:32:28 by simonpj]
simonpj [Wed, 8 Dec 2004 14:32:34 +0000 (14:32 +0000)]
[project @ 2004-12-08 14:32:28 by simonpj]
Comments and asserts only

19 years ago[project @ 2004-12-08 12:05:51 by simonmar]
simonmar [Wed, 8 Dec 2004 12:05:51 +0000 (12:05 +0000)]
[project @ 2004-12-08 12:05:51 by simonmar]
Fix bug #1073501: checkProddableBlock: invalid fixup in runtime linker

The bug manifested when trying to load an object with debugging info
(compiled with gcc -g) into GHCi.

The problem was that the object loader was ignoring the sections
containing debugging info, but then it was later trying to do
relocations for those sections, and its own sanity checking code
correctly detected that the relocations were in unknown parts of the
object file.

The fix is to ignore relocations whose target section isn't one of the
sections that we're interested in, using the same test in both cases
(the code to test section kind has been extracted).  The code could
probably benefit from more refactoring: it looks like the list of
sections we build up in the first phase isn't even used in the second
phase, instead we traverse the section table in the image again.  This
looks like cruft leftover from when the GC used to check whether an
address was in text or data space.

19 years ago[project @ 2004-12-08 10:29:01 by simonmar]
simonmar [Wed, 8 Dec 2004 10:29:01 +0000 (10:29 +0000)]
[project @ 2004-12-08 10:29:01 by simonmar]
deleteThread: don't attempt to delete threads blocked in foreign calls.

19 years ago[project @ 2004-12-08 09:28:57 by simonmar]
simonmar [Wed, 8 Dec 2004 09:28:57 +0000 (09:28 +0000)]
[project @ 2004-12-08 09:28:57 by simonmar]
Prevent any intermediate file deletion using

  .SECONDARY:

and remove now-unnecessary .PRECIOUS and .SECONDARY targets.
Suggestion from Ashley Yakeley.

19 years ago[project @ 2004-12-06 10:58:06 by simonpj]
simonpj [Mon, 6 Dec 2004 10:58:06 +0000 (10:58 +0000)]
[project @ 2004-12-06 10:58:06 by simonpj]
---------------------
Bug in specialisation
---------------------

Laszlo managed to get a function like this:

foo :: Enum a => (# a, Int #)

The specialiser specialised it, resulting in an unboxed tuple
binding, which Lint objected to.

This commit adds a dummy argument to the specialised function,
very like the case for strictness analysis.  For example, at
type Char we'd get

foo_char :: State# RealWorld -> (# Char, Int #)
  foo_char = \_ -> ...

We use a State# type because it generates no argument-passing code
at runtime.  (We should really have some other void type for this
purpose, because State# is misleading, but this way avoids extra
types.)

19 years ago[project @ 2004-12-06 10:51:36 by simonpj]
simonpj [Mon, 6 Dec 2004 10:51:36 +0000 (10:51 +0000)]
[project @ 2004-12-06 10:51:36 by simonpj]
------------------------------------
Bug in loop detection in TcSimplify
------------------------------------

The type-class context simplifier has been able to
build recursive dictionaries for some time: co-induction.
That is, you can build a proof for constraint C by assuming
that C holds when proving the preconditions of C.

You need to be in -fallow-undecidable-instances land to
make use of this: see comments with [RECURSIVE DICTIONARIES]
in TcSimplify.lhs.

Anyway, this is all fine, but I'd implemented it wrong!  You need
to be very careful with superclasses, or you can make a bogus
loop by mistake.  This commit fixes it; tests LoopOfTheDay{1,2,3}
will test it (thanks Ralf Laemmel).

19 years ago[project @ 2004-12-06 10:47:22 by simonmar]
simonmar [Mon, 6 Dec 2004 10:47:22 +0000 (10:47 +0000)]
[project @ 2004-12-06 10:47:22 by simonmar]
Error message fixup

19 years ago[project @ 2004-12-04 14:07:25 by panne]
panne [Sat, 4 Dec 2004 14:07:27 +0000 (14:07 +0000)]
[project @ 2004-12-04 14:07:25 by panne]
Moved GTK_CONFIG detection to ghc subdirectory (untested).

19 years ago[project @ 2004-12-04 13:48:05 by panne]
panne [Sat, 4 Dec 2004 13:48:06 +0000 (13:48 +0000)]
[project @ 2004-12-04 13:48:05 by panne]
Nuked some outdated stuff

19 years ago[project @ 2004-12-04 12:57:57 by panne]
panne [Sat, 4 Dec 2004 12:57:57 +0000 (12:57 +0000)]
[project @ 2004-12-04 12:57:57 by panne]
No need for --enable-hopengl anymore

19 years ago[project @ 2004-12-03 16:59:43 by simonpj]
simonpj [Fri, 3 Dec 2004 16:59:43 +0000 (16:59 +0000)]
[project @ 2004-12-03 16:59:43 by simonpj]
Make Core Lint check for 1-tuples

19 years ago[project @ 2004-12-03 16:25:58 by simonmar]
simonmar [Fri, 3 Dec 2004 16:25:58 +0000 (16:25 +0000)]
[project @ 2004-12-03 16:25:58 by simonmar]
Get rid of some `...' style quotes

19 years ago[project @ 2004-12-03 16:02:44 by simonpj]
simonpj [Fri, 3 Dec 2004 16:02:44 +0000 (16:02 +0000)]
[project @ 2004-12-03 16:02:44 by simonpj]
Update TH import

19 years ago[project @ 2004-12-03 15:35:56 by simonmar]
simonmar [Fri, 3 Dec 2004 15:35:59 +0000 (15:35 +0000)]
[project @ 2004-12-03 15:35:56 by simonmar]
distclean: clean config.mk only *after* doing the recursive cleaning.
Otherwise we don't clean properly in the subdirectories or sub-ways.

19 years ago[project @ 2004-12-03 13:57:19 by simonmar]
simonmar [Fri, 3 Dec 2004 13:57:19 +0000 (13:57 +0000)]
[project @ 2004-12-03 13:57:19 by simonmar]
- Implement expose/hide
- fix parsing of package identifiers (forgot to commit this the other day)

19 years ago[project @ 2004-12-03 13:49:00 by simonpj]
simonpj [Fri, 3 Dec 2004 13:49:05 +0000 (13:49 +0000)]
[project @ 2004-12-03 13:49:00 by simonpj]
A fix to kind signatures for GADT data type decls

19 years ago[project @ 2004-12-03 13:47:22 by simonpj]
simonpj [Fri, 3 Dec 2004 13:47:22 +0000 (13:47 +0000)]
[project @ 2004-12-03 13:47:22 by simonpj]
TH refication for primitive TyCons

19 years ago[project @ 2004-12-03 13:01:28 by simonmar]
simonmar [Fri, 3 Dec 2004 13:01:28 +0000 (13:01 +0000)]
[project @ 2004-12-03 13:01:28 by simonmar]
Support the --auto-ghci-libs option in conjuction with 'register'.

19 years ago[project @ 2004-12-02 23:39:13 by wolfgang]
wolfgang [Thu, 2 Dec 2004 23:39:13 +0000 (23:39 +0000)]
[project @ 2004-12-02 23:39:13 by wolfgang]
Mac OS X/Darwin/Mach-O:

Improve handling of object files without dynamic symbol tables.

19 years ago[project @ 2004-12-02 17:18:15 by simonpj]
simonpj [Thu, 2 Dec 2004 17:18:32 +0000 (17:18 +0000)]
[project @ 2004-12-02 17:18:15 by simonpj]
Sorry for the fact that there are overlapping three commits in here...

1.  Make -fno-monomorphism-restriction
    and -fno-implicit-prelude reversible, like other flags

2.  Fix a wibble in the new ImportAvails story, in RnNames.mkExportAvails

3.  Fix a Template Haskell bug that meant that top-level names created
    with newName were not made properly unique.

19 years ago[project @ 2004-12-02 17:17:28 by simonpj]
simonpj [Thu, 2 Dec 2004 17:17:28 +0000 (17:17 +0000)]
[project @ 2004-12-02 17:17:28 by simonpj]
Trim imports

19 years ago[project @ 2004-12-02 16:55:02 by simonpj]
simonpj [Thu, 2 Dec 2004 16:55:02 +0000 (16:55 +0000)]
[project @ 2004-12-02 16:55:02 by simonpj]
Missing include

19 years ago[project @ 2004-12-02 16:33:16 by simonpj]
simonpj [Thu, 2 Dec 2004 16:33:16 +0000 (16:33 +0000)]
[project @ 2004-12-02 16:33:16 by simonpj]
Missing import on Windows

19 years ago[project @ 2004-12-01 09:51:50 by simonmar]
simonmar [Wed, 1 Dec 2004 09:52:14 +0000 (09:52 +0000)]
[project @ 2004-12-01 09:51:50 by simonmar]
Update the excluded modules for libghccompat

19 years ago[project @ 2004-11-30 16:07:01 by simonmar]
simonmar [Tue, 30 Nov 2004 16:07:01 +0000 (16:07 +0000)]
[project @ 2004-11-30 16:07:01 by simonmar]
Data.Version --> Distribution.Version

19 years ago[project @ 2004-11-30 16:05:15 by simonmar]
simonmar [Tue, 30 Nov 2004 16:05:15 +0000 (16:05 +0000)]
[project @ 2004-11-30 16:05:15 by simonmar]
Data.Version -> Distribution.Version

19 years ago[project @ 2004-11-30 15:51:35 by simonmar]
simonmar [Tue, 30 Nov 2004 15:51:35 +0000 (15:51 +0000)]
[project @ 2004-11-30 15:51:35 by simonmar]
Data.Version isn't necessary any more, the Version type is exported by
Distribution.Version.

19 years ago[project @ 2004-11-30 15:31:56 by simonmar]
simonmar [Tue, 30 Nov 2004 15:31:56 +0000 (15:31 +0000)]
[project @ 2004-11-30 15:31:56 by simonmar]
Allow -ignore-package P when P doesn't exist (thanks to George Russell
for the report).

19 years ago[project @ 2004-11-30 14:55:10 by simonpj]
simonpj [Tue, 30 Nov 2004 14:55:10 +0000 (14:55 +0000)]
[project @ 2004-11-30 14:55:10 by simonpj]
Docmunent -fmonomorphism-restriction and -fimplicit-prelude

19 years ago[project @ 2004-11-30 14:28:02 by simonpj]
simonpj [Tue, 30 Nov 2004 14:28:19 +0000 (14:28 +0000)]
[project @ 2004-11-30 14:28:02 by simonpj]
Import trimming

19 years ago[project @ 2004-11-30 13:59:25 by simonpj]
simonpj [Tue, 30 Nov 2004 13:59:25 +0000 (13:59 +0000)]
[project @ 2004-11-30 13:59:25 by simonpj]
Wibbles for new ImportAvails story

19 years ago[project @ 2004-11-30 10:46:58 by simonmar]
simonmar [Tue, 30 Nov 2004 10:46:58 +0000 (10:46 +0000)]
[project @ 2004-11-30 10:46:58 by simonmar]
Add space to error message

19 years ago[project @ 2004-11-29 23:28:58 by wolfgang]
wolfgang [Mon, 29 Nov 2004 23:28:58 +0000 (23:28 +0000)]
[project @ 2004-11-29 23:28:58 by wolfgang]
Revert part of the previous commit.
Something unrelated slipped in :-(.

19 years ago[project @ 2004-11-29 23:21:43 by wolfgang]
wolfgang [Mon, 29 Nov 2004 23:21:43 +0000 (23:21 +0000)]
[project @ 2004-11-29 23:21:43 by wolfgang]
Fix -dynamic compilation - don't use nameModule on names that might be
local.

19 years ago[project @ 2004-11-29 16:25:03 by simonpj]
simonpj [Mon, 29 Nov 2004 16:25:12 +0000 (16:25 +0000)]
[project @ 2004-11-29 16:25:03 by simonpj]
---------------------
Simplify ImportAvails
---------------------

Every Name has, for some while, contained its "parent";
the type or class inside which it is defined.  But the rest
of the renamer wasn't using this information as much as it
could do.  In particular, the ImportAvails type was more elaborate
than necessary.

This commit combines these two fields of ImportAvails:
imp_env :: AvailEnv
imp_qual :: ModuleEnv AvailEnv
into one
imp_env :: ModuleEnv NameSet

This is quite a bit simpler.  Less redundancy and, I think, less
code.

19 years ago[project @ 2004-11-29 16:24:16 by simonpj]
simonpj [Mon, 29 Nov 2004 16:24:21 +0000 (16:24 +0000)]
[project @ 2004-11-29 16:24:16 by simonpj]
Trim imports

19 years ago[project @ 2004-11-29 16:16:57 by simonpj]
simonpj [Mon, 29 Nov 2004 16:16:57 +0000 (16:16 +0000)]
[project @ 2004-11-29 16:16:57 by simonpj]
Update ambiguity errors

19 years ago[project @ 2004-11-29 11:35:43 by simonmar]
simonmar [Mon, 29 Nov 2004 11:35:43 +0000 (11:35 +0000)]
[project @ 2004-11-29 11:35:43 by simonmar]
Don't quote package names

19 years ago[project @ 2004-11-29 11:32:38 by simonmar]
simonmar [Mon, 29 Nov 2004 11:32:38 +0000 (11:32 +0000)]
[project @ 2004-11-29 11:32:38 by simonmar]
Look for package modules before home modules.  This fixes the build in
ghc/lib when bootstrapping with the HEAD.

19 years ago[project @ 2004-11-26 16:50:56 by simonmar]
simonmar [Fri, 26 Nov 2004 16:50:56 +0000 (16:50 +0000)]
[project @ 2004-11-26 16:50:56 by simonmar]
unregister/describe: allow the package name to be given without the
version, as long as it is unambiguous.

Strange, I was sure I'd implemented expose/hide in here, but they're
stubbed out.  Oh well.

19 years ago[project @ 2004-11-26 16:34:56 by simonmar]
simonmar [Fri, 26 Nov 2004 16:34:56 +0000 (16:34 +0000)]
[project @ 2004-11-26 16:34:56 by simonmar]
Undo

19 years ago[project @ 2004-11-26 16:32:44 by simonmar]
simonmar [Fri, 26 Nov 2004 16:32:44 +0000 (16:32 +0000)]
[project @ 2004-11-26 16:32:44 by simonmar]
Further integration with the new package story.  GHC now supports
pretty much everything in the package proposal.

  - GHC now works in terms of PackageIds (<pkg>-<version>) rather than
    just package names.  You can still specify package names without
    versions on the command line, as long as the name is unambiguous.

  - GHC understands hidden/exposed modules in a package, and will refuse
    to import a hidden module.  Also, the hidden/eposed status of packages
    is taken into account.

  - I had to remove the old package syntax from ghc-pkg, backwards
    compatibility isn't really practical.

  - All the package.conf.in files have been rewritten in the new syntax,
    and contain a complete list of modules in the package.  I've set all
    the versions to 1.0 for now - please check your package(s) and fix the
    version number & other info appropriately.

  - New options:

-hide-package P    sets the expose flag on package P to False
-ignore-package P  unregisters P for this compilation

For comparison, -package P sets the expose flag on package P
        to True, and also causes P to be linked in eagerly.

        -package-name is no longer officially supported.  Unofficially, it's
a synonym for -ignore-package, which has more or less the same effect
as -package-name used to.

Note that a package may be hidden and yet still be linked into
the program, by virtue of being a dependency of some other package.
To completely remove a package from the compiler's internal database,
        use -ignore-package.

The compiler will complain if any two packages in the
        transitive closure of exposed packages contain the same
        module.

You *must* use -ignore-package P when compiling modules for
        package P, if package P (or an older version of P) is already
        registered.  The compiler will helpfully complain if you don't.
The fptools build system does this.

   - Note: the Cabal library won't work yet.  It still thinks GHC uses
     the old package config syntax.

Internal changes/cleanups:

   - The ModuleName type has gone away.  Modules are now just (a
     newtype of) FastStrings, and don't contain any package information.
     All the package-related knowledge is in DynFlags, which is passed
     down to where it is needed.

   - DynFlags manipulation has been cleaned up somewhat: there are no
     global variables holding DynFlags any more, instead the DynFlags
     are passed around properly.

   - There are a few less global variables in GHC.  Lots more are
     scheduled for removal.

   - -i is now a dynamic flag, as are all the package-related flags (but
     using them in {-# OPTIONS #-} is Officially Not Recommended).

   - make -j now appears to work under fptools/libraries/.  Probably
     wouldn't take much to get it working for a whole build.

19 years ago[project @ 2004-11-26 16:30:12 by simonmar]
simonmar [Fri, 26 Nov 2004 16:30:12 +0000 (16:30 +0000)]
[project @ 2004-11-26 16:30:12 by simonmar]
Add missing .hi-boot files

19 years ago[project @ 2004-11-26 16:19:45 by simonmar]
simonmar [Fri, 26 Nov 2004 16:22:13 +0000 (16:22 +0000)]
[project @ 2004-11-26 16:19:45 by simonmar]
Further integration with the new package story.  GHC now supports
pretty much everything in the package proposal.

  - GHC now works in terms of PackageIds (<pkg>-<version>) rather than
    just package names.  You can still specify package names without
    versions on the command line, as long as the name is unambiguous.

  - GHC understands hidden/exposed modules in a package, and will refuse
    to import a hidden module.  Also, the hidden/eposed status of packages
    is taken into account.

  - I had to remove the old package syntax from ghc-pkg, backwards
    compatibility isn't really practical.

  - All the package.conf.in files have been rewritten in the new syntax,
    and contain a complete list of modules in the package.  I've set all
    the versions to 1.0 for now - please check your package(s) and fix the
    version number & other info appropriately.

  - New options:

-hide-package P    sets the expose flag on package P to False
-ignore-package P  unregisters P for this compilation

For comparison, -package P sets the expose flag on package P
        to True, and also causes P to be linked in eagerly.

        -package-name is no longer officially supported.  Unofficially, it's
a synonym for -ignore-package, which has more or less the same effect
as -package-name used to.

Note that a package may be hidden and yet still be linked into
the program, by virtue of being a dependency of some other package.
To completely remove a package from the compiler's internal database,
        use -ignore-package.

The compiler will complain if any two packages in the
        transitive closure of exposed packages contain the same
        module.

You *must* use -ignore-package P when compiling modules for
        package P, if package P (or an older version of P) is already
        registered.  The compiler will helpfully complain if you don't.
The fptools build system does this.

   - Note: the Cabal library won't work yet.  It still thinks GHC uses
     the old package config syntax.

Internal changes/cleanups:

   - The ModuleName type has gone away.  Modules are now just (a
     newtype of) FastStrings, and don't contain any package information.
     All the package-related knowledge is in DynFlags, which is passed
     down to where it is needed.

   - DynFlags manipulation has been cleaned up somewhat: there are no
     global variables holding DynFlags any more, instead the DynFlags
     are passed around properly.

   - There are a few less global variables in GHC.  Lots more are
     scheduled for removal.

   - -i is now a dynamic flag, as are all the package-related flags (but
     using them in {-# OPTIONS #-} is Officially Not Recommended).

   - make -j now appears to work under fptools/libraries/.  Probably
     wouldn't take much to get it working for a whole build.

19 years ago[project @ 2004-11-26 13:42:37 by simonpj]
simonpj [Fri, 26 Nov 2004 13:42:39 +0000 (13:42 +0000)]
[project @ 2004-11-26 13:42:37 by simonpj]
Document overlapping instances

19 years ago[project @ 2004-11-26 13:42:21 by simonpj]
simonpj [Fri, 26 Nov 2004 13:42:21 +0000 (13:42 +0000)]
[project @ 2004-11-26 13:42:21 by simonpj]
More about hi-boot files

19 years ago[project @ 2004-11-25 11:36:34 by simonpj]
simonpj [Thu, 25 Nov 2004 11:37:19 +0000 (11:37 +0000)]
[project @ 2004-11-25 11:36:34 by simonpj]
------------------------------------------
Keep-alive set and Template Haskell quotes
------------------------------------------

a) Template Haskell quotes should be able to mention top-leve
   things without resorting to lifting.  Example

module Foo( foo ) where
  f x = x
  foo = [| f 4 |]

   Here the reference to 'f' is ok; no need to 'lift' it.
   The relevant changes are in TcExpr.tcId

b) However, we must take care not to discard the binding for f,
   so we add it to the 'keep-alive' set for the module.  I've
   now made this into (another) mutable bucket, tcg_keep,
   in the TcGblEnv

c) That in turn led me to look at the handling of orphan rules;
   as a result I made IdCoreRule into its own data type, which
   has simle but non-local ramifications

19 years ago[project @ 2004-11-24 13:03:26 by simonmar]
simonmar [Wed, 24 Nov 2004 13:03:26 +0000 (13:03 +0000)]
[project @ 2004-11-24 13:03:26 by simonmar]
Update s/r conflict total after Ross added four more.  The total
according to the comment is currently one more than the real number,
so one of them must have gone away.  At some point we should track
down which one.

19 years ago[project @ 2004-11-23 15:24:02 by simonpj]
simonpj [Tue, 23 Nov 2004 15:24:02 +0000 (15:24 +0000)]
[project @ 2004-11-23 15:24:02 by simonpj]
White space only

19 years ago[project @ 2004-11-23 14:24:53 by simonmar]
simonmar [Tue, 23 Nov 2004 14:24:53 +0000 (14:24 +0000)]
[project @ 2004-11-23 14:24:53 by simonmar]
This should fix the build for GHC 6.3+

19 years ago[project @ 2004-11-23 12:35:06 by ross]
ross [Tue, 23 Nov 2004 12:35:19 +0000 (12:35 +0000)]
[project @ 2004-11-23 12:35:06 by ross]
Make the X11 and HGL packages more independent, and pass configuration
parameters to package.conf.in via header files.

Also removed residual traces of OpenGL/GLUT/OpenAL from mk/config.mk.in

19 years ago[project @ 2004-11-22 17:00:54 by simonmar]
simonmar [Mon, 22 Nov 2004 17:00:54 +0000 (17:00 +0000)]
[project @ 2004-11-22 17:00:54 by simonmar]
Fix mkdependHS behaviour

19 years ago[project @ 2004-11-22 16:42:21 by simonmar]
simonmar [Mon, 22 Nov 2004 16:42:21 +0000 (16:42 +0000)]
[project @ 2004-11-22 16:42:21 by simonmar]
\r is whitespace

19 years ago[project @ 2004-11-22 16:39:37 by simonmar]
simonmar [Mon, 22 Nov 2004 16:39:37 +0000 (16:39 +0000)]
[project @ 2004-11-22 16:39:37 by simonmar]
Add readline/GHCi/Windows entry