ghc-hetmet.git
23 years ago[project @ 2000-07-26 13:29:03 by simonmar]
simonmar [Wed, 26 Jul 2000 13:29:03 +0000 (13:29 +0000)]
[project @ 2000-07-26 13:29:03 by simonmar]
- add wORD_SIZE (size of an StgWord in bytes)
- add bLOCK_SIZE_W (size of a storage manager block in words)

23 years ago[project @ 2000-07-26 13:27:54 by simonmar]
simonmar [Wed, 26 Jul 2000 13:27:54 +0000 (13:27 +0000)]
[project @ 2000-07-26 13:27:54 by simonmar]
Add a constant definition for WORD_SIZE, the size of an StgWord in bytes.

23 years ago[project @ 2000-07-26 13:13:49 by simonmar]
simonmar [Wed, 26 Jul 2000 13:13:49 +0000 (13:13 +0000)]
[project @ 2000-07-26 13:13:49 by simonmar]
Revert part of the previous change, which broke the build by
generating a PrelStable.hi with some bogus kind information in it.

23 years ago[project @ 2000-07-25 18:16:53 by rrt]
rrt [Tue, 25 Jul 2000 18:16:53 +0000 (18:16 +0000)]
[project @ 2000-07-25 18:16:53 by rrt]
Advised Windows users to read the Windows installation notes for GHC before
trying to build anything.

23 years ago[project @ 2000-07-25 18:10:38 by rrt]
rrt [Tue, 25 Jul 2000 18:10:38 +0000 (18:10 +0000)]
[project @ 2000-07-25 18:10:38 by rrt]
Pointed out that Windows users should read the GHC installation guide before
trying to build anything.

23 years ago[project @ 2000-07-25 15:54:38 by simonmar]
simonmar [Tue, 25 Jul 2000 15:54:38 +0000 (15:54 +0000)]
[project @ 2000-07-25 15:54:38 by simonmar]
increase the repeat count; this bug is pretty fragile

23 years ago[project @ 2000-07-25 15:48:16 by simonmar]
simonmar [Tue, 25 Jul 2000 15:48:16 +0000 (15:48 +0000)]
[project @ 2000-07-25 15:48:16 by simonmar]
Add test for handle finalization bug, fixed in
  1.60      +1 -2      fptools/ghc/lib/std/PrelHandle.lhs
  1.15      +4 -10     fptools/ghc/lib/std/PrelIO.lhs

23 years ago[project @ 2000-07-25 15:20:10 by simonmar]
simonmar [Tue, 25 Jul 2000 15:20:10 +0000 (15:20 +0000)]
[project @ 2000-07-25 15:20:10 by simonmar]
Fix bug reported by Hannah Schroeter: reading a file lazily using
hGetContents and then closing it using hClose can cause the program to
fall over with a deadlock.

The reason is that when closing the file in lazyRead{Block,Line,Char},
we set the foreign object in the handle to nullFile__, which causes
the finalizer to run (at some point in the future).  The finalizer
takes the MVar in the handle, frees the contents, but never puts the
MVar back.  hClose then tries to take the MVar, and deadlocks.

The solution is not to set the foreign object to nullFile__ in the
first place; I'm not sure why it was done this way, and in fact it
leads to a memory leak.  hClose itself has a similar problem, leading
to a leak of the fileObject.

23 years ago[project @ 2000-07-25 12:24:14 by michaelw]
michaelw [Tue, 25 Jul 2000 12:24:14 +0000 (12:24 +0000)]
[project @ 2000-07-25 12:24:14 by michaelw]
giving the version number when asked for is NOT an error, therefore
output now goes to stdout (cf. Coding Standards)

23 years ago[project @ 2000-07-24 15:31:47 by simonmar]
simonmar [Mon, 24 Jul 2000 15:31:47 +0000 (15:31 +0000)]
[project @ 2000-07-24 15:31:47 by simonmar]
update documentation to reflect changes in the format of a package spec.

23 years ago[project @ 2000-07-24 15:16:44 by simonmar]
simonmar [Mon, 24 Jul 2000 15:16:44 +0000 (15:16 +0000)]
[project @ 2000-07-24 15:16:44 by simonmar]
Sigh, change the package definition again.  We weren't making a
distinction between libraries which need a suffix tag
(eg. libHSstd_p), and those that don't (eg. libHSstd_cbits).

The package spec now has two components for libraries, hs_libraries
(tagged) and extra_libraries (untagged).  The ordering of these
components is important: we specify that hs_libraries are linked
before extra_libraries.  This sounds problematic if you want to link
some plain C libraries before some Haskell libs, but in these cases it
should be possible to create separate packages for the two libs and
specify the dependencies explicitly.

23 years ago[project @ 2000-07-24 14:29:55 by simonmar]
simonmar [Mon, 24 Jul 2000 14:29:55 +0000 (14:29 +0000)]
[project @ 2000-07-24 14:29:55 by simonmar]
Some changes to the way FFI decls are handled:

  - a foreign export dynamic which returns a newtype of
    an Addr now works correctly.  Similarly for foreign label.

  - unlifted types are not allowed in the arguments of a foreign
    export.  Previously we generated incorrect code for these cases.

Newtypes in FFI declarations now work everywhere they should, as far
as I can see.

These changes will be backported into 4.08.1.

23 years ago[project @ 2000-07-23 10:53:11 by panne]
panne [Sun, 23 Jul 2000 10:53:12 +0000 (10:53 +0000)]
[project @ 2000-07-23 10:53:11 by panne]
Strictfp-like behaviour is the default now, which can be switched off
via -fexcess-precision. (Has anybody a better name for this option?)

23 years ago[project @ 2000-07-21 09:48:47 by rrt]
rrt [Fri, 21 Jul 2000 09:48:47 +0000 (09:48 +0000)]
[project @ 2000-07-21 09:48:47 by rrt]
Fix EXTINFO_RTS.

23 years ago[project @ 2000-07-21 09:31:46 by rrt]
rrt [Fri, 21 Jul 2000 09:31:46 +0000 (09:31 +0000)]
[project @ 2000-07-21 09:31:46 by rrt]
Add EXTINFO_RTS and use it (same as EXTFUN_RTS for info tables).

23 years ago[project @ 2000-07-21 09:11:19 by rrt]
rrt [Fri, 21 Jul 2000 09:11:19 +0000 (09:11 +0000)]
[project @ 2000-07-21 09:11:19 by rrt]
Changed ERTSF_ to EXTFUN_RTS.

23 years ago[project @ 2000-07-21 09:01:48 by rrt]
rrt [Fri, 21 Jul 2000 09:01:48 +0000 (09:01 +0000)]
[project @ 2000-07-21 09:01:48 by rrt]
Removed ERTSF_ and used EXTFUN_RTS instead, which does the same thing.

23 years ago[project @ 2000-07-21 08:45:05 by rrt]
rrt [Fri, 21 Jul 2000 08:45:05 +0000 (08:45 +0000)]
[project @ 2000-07-21 08:45:05 by rrt]
Make sure we *don't* try to make hsc_dll (as it will be identical to hsc).

23 years ago[project @ 2000-07-21 08:40:04 by rrt]
rrt [Fri, 21 Jul 2000 08:40:04 +0000 (08:40 +0000)]
[project @ 2000-07-21 08:40:04 by rrt]
Add dll as a fully-fledged way.

23 years ago[project @ 2000-07-21 08:37:30 by rrt]
rrt [Fri, 21 Jul 2000 08:37:30 +0000 (08:37 +0000)]
[project @ 2000-07-21 08:37:30 by rrt]
Make gmp.dll in DLL_PEN.

23 years ago[project @ 2000-07-21 08:35:22 by rrt]
rrt [Fri, 21 Jul 2000 08:35:22 +0000 (08:35 +0000)]
[project @ 2000-07-21 08:35:22 by rrt]
Make sure DLL_NAME only has DLL_PEN prepended if DLL_NAME has already been
set, i.e. a DLL is really to be built.

23 years ago[project @ 2000-07-20 17:06:41 by rrt]
rrt [Thu, 20 Jul 2000 17:06:41 +0000 (17:06 +0000)]
[project @ 2000-07-20 17:06:41 by rrt]
What's the story, Rory?

   My name's not Rory. Owing to circumstances not fully understood, the RTS
   used to work DLLized, even though it was largely broken. Unfortunately,
   this was recently fixed.

What used to happen?

   The EF_ macro from StgMacros.h was used to refer to primops &c.

Why was this a Bad Thing?

   Because you need to use EDF_ when building with DLLs.

What was the result?

   primops such as divExactInteger were compiled as _divExactInteger, rather
   than __imp__div_Exact_Integer, so they didn't refer to the DLL routines.

How on earth did it work?

   I'm not sure, but somehow the code for the relevant routines got linked
   into the import libraries (e.g. libHSrts_imp.a), and were thus linked
   statically into the final binary.

So that explains why the import libraries (supposedly just containing stubs)
were larger than the DLLs they were stubbing, or the static libraries, for
that matter?

   Perhaps.

Golly!

   Indeed.

What caused this code leakage?

   That's what I don't know. Perhaps a bug in the Cygwin DLL-building tools?
   I've upgraded from B20.1 to 1.1 recently.

So, what's the hack, Mac?

   My name's not Mac. I added a new macro, ERTSF_, which uses DLL_IMPORT_RTS
   to make sure that the Right Thing is always done (because you mustn't
   make DLL references to RTS routines when compiling the RTS).

Thanks for the chat, Matt!

   <punch>

23 years ago[project @ 2000-07-20 12:33:09 by rrt]
rrt [Thu, 20 Jul 2000 12:33:09 +0000 (12:33 +0000)]
[project @ 2000-07-20 12:33:09 by rrt]
4.07->4.08

23 years ago[project @ 2000-07-20 12:32:50 by rrt]
rrt [Thu, 20 Jul 2000 12:32:50 +0000 (12:32 +0000)]
[project @ 2000-07-20 12:32:50 by rrt]
Fixed a version-o.

23 years ago[project @ 2000-07-20 10:50:47 by rrt]
rrt [Thu, 20 Jul 2000 10:50:47 +0000 (10:50 +0000)]
[project @ 2000-07-20 10:50:47 by rrt]
&amp;gt; => &gt;

23 years ago[project @ 2000-07-20 10:34:42 by rrt]
rrt [Thu, 20 Jul 2000 10:34:42 +0000 (10:34 +0000)]
[project @ 2000-07-20 10:34:42 by rrt]
Updated instructions for Windows install to reflect reality as of v4.08.

23 years ago[project @ 2000-07-19 14:46:40 by rrt]
rrt [Wed, 19 Jul 2000 14:46:40 +0000 (14:46 +0000)]
[project @ 2000-07-19 14:46:40 by rrt]
Added -dll option for dll way, now a proper way.

23 years ago[project @ 2000-07-19 14:45:53 by rrt]
rrt [Wed, 19 Jul 2000 14:45:53 +0000 (14:45 +0000)]
[project @ 2000-07-19 14:45:53 by rrt]
Target dlls directly into DLL_PEN so that they don't get rebuilt all the time.

23 years ago[project @ 2000-07-19 10:07:29 by rrt]
rrt [Wed, 19 Jul 2000 10:07:29 +0000 (10:07 +0000)]
[project @ 2000-07-19 10:07:29 by rrt]
Gave driver access to bash for running system calls under Windows by writing
command out to a temp file, then sending that as the argument to "sh -". Sigh.
This is #ifdefed on mingw32_TARGET_OS; saner OSs just use system as normal.

23 years ago[project @ 2000-07-19 08:46:10 by rrt]
rrt [Wed, 19 Jul 2000 08:46:10 +0000 (08:46 +0000)]
[project @ 2000-07-19 08:46:10 by rrt]
Stop doing all system calls via "sh -c", as this messed up quoting.

23 years ago[project @ 2000-07-18 17:34:52 by rrt]
rrt [Tue, 18 Jul 2000 17:34:52 +0000 (17:34 +0000)]
[project @ 2000-07-18 17:34:52 by rrt]
oops, ghc-inplace ain't an EXE.

23 years ago[project @ 2000-07-18 17:31:02 by rrt]
rrt [Tue, 18 Jul 2000 17:31:02 +0000 (17:31 +0000)]
[project @ 2000-07-18 17:31:02 by rrt]
Added a few EXE_SUFFIXes to get the driver going on Windows.

23 years ago[project @ 2000-07-18 17:29:53 by rrt]
rrt [Tue, 18 Jul 2000 17:29:53 +0000 (17:29 +0000)]
[project @ 2000-07-18 17:29:53 by rrt]
Changed ExeSuffix to exeext (which is what configure.in actually calls it).

23 years ago[project @ 2000-07-18 17:09:02 by simonmar]
simonmar [Tue, 18 Jul 2000 17:09:02 +0000 (17:09 +0000)]
[project @ 2000-07-18 17:09:02 by simonmar]
remove hscpp

23 years ago[project @ 2000-07-18 15:57:16 by rrt]
rrt [Tue, 18 Jul 2000 15:57:16 +0000 (15:57 +0000)]
[project @ 2000-07-18 15:57:16 by rrt]
Add $(EXE_SUFFIX) to hp2ps and unlit

23 years ago[project @ 2000-07-18 15:55:19 by rrt]
rrt [Tue, 18 Jul 2000 15:55:19 +0000 (15:55 +0000)]
[project @ 2000-07-18 15:55:19 by rrt]
Move dlls to $(DLL_PEN) after generation.

23 years ago[project @ 2000-07-18 15:39:59 by rrt]
rrt [Tue, 18 Jul 2000 15:39:59 +0000 (15:39 +0000)]
[project @ 2000-07-18 15:39:59 by rrt]
Move DLLs to $(DLL_PEN) after generation.

23 years ago[project @ 2000-07-18 14:50:32 by simonpj]
simonpj [Tue, 18 Jul 2000 14:50:32 +0000 (14:50 +0000)]
[project @ 2000-07-18 14:50:32 by simonpj]
Arrange that type signatures that have no "parent" signatures
work right.  For example

   module A
      f :: Int -> Int

The panic in RnEnv.lookupBndrRn becomes an error message.  Also
RnBinds.rnTopMonoBinds and rnMonoBinds were wrongly ignoring all
signatures altogether if there were no bindings!

23 years ago[project @ 2000-07-18 14:50:07 by simonmar]
simonmar [Tue, 18 Jul 2000 14:50:07 +0000 (14:50 +0000)]
[project @ 2000-07-18 14:50:07 by simonmar]
fix finding of the usage message (again).

23 years ago[project @ 2000-07-18 14:47:52 by simonpj]
simonpj [Tue, 18 Jul 2000 14:47:52 +0000 (14:47 +0000)]
[project @ 2000-07-18 14:47:52 by simonpj]
Add comment

23 years ago[project @ 2000-07-18 14:46:48 by simonpj]
simonpj [Tue, 18 Jul 2000 14:46:48 +0000 (14:46 +0000)]
[project @ 2000-07-18 14:46:48 by simonpj]
Add type-sig test

23 years ago[project @ 2000-07-18 13:31:34 by rrt]
rrt [Tue, 18 Jul 2000 13:31:34 +0000 (13:31 +0000)]
[project @ 2000-07-18 13:31:34 by rrt]
Holding pen for dlls to stop them getting in the way of pre-compiled
dynamic programs.

23 years ago[project @ 2000-07-18 08:08:35 by simonpj]
simonpj [Tue, 18 Jul 2000 08:08:35 +0000 (08:08 +0000)]
[project @ 2000-07-18 08:08:35 by simonpj]
Remove dead code: checkConstraints and univErr

23 years ago[project @ 2000-07-17 18:44:16 by keithw]
keithw [Mon, 17 Jul 2000 18:44:16 +0000 (18:44 +0000)]
[project @ 2000-07-17 18:44:16 by keithw]
Insert code to handle instruction counts from Heimdall, as well as
CacheProf.

23 years ago[project @ 2000-07-17 16:02:02 by rrt]
rrt [Mon, 17 Jul 2000 16:02:02 +0000 (16:02 +0000)]
[project @ 2000-07-17 16:02:02 by rrt]
Fixed a bracket-o.

23 years ago[project @ 2000-07-17 15:27:15 by rrt]
rrt [Mon, 17 Jul 2000 15:27:15 +0000 (15:27 +0000)]
[project @ 2000-07-17 15:27:15 by rrt]
Windows fixes and tidying up.

23 years ago[project @ 2000-07-17 15:25:05 by rrt]
rrt [Mon, 17 Jul 2000 15:25:05 +0000 (15:25 +0000)]
[project @ 2000-07-17 15:25:05 by rrt]
Windows fixes.

23 years ago[project @ 2000-07-17 15:15:40 by rrt]
rrt [Mon, 17 Jul 2000 15:15:40 +0000 (15:15 +0000)]
[project @ 2000-07-17 15:15:40 by rrt]
Changed __MINGW32__ to mingw32_TARGET_OS to cope with latest mingwin.

23 years ago[project @ 2000-07-17 15:11:00 by rrt]
rrt [Mon, 17 Jul 2000 15:11:00 +0000 (15:11 +0000)]
[project @ 2000-07-17 15:11:00 by rrt]
Rename files in line with RTS.

23 years ago[project @ 2000-07-17 15:10:14 by rrt]
rrt [Mon, 17 Jul 2000 15:10:14 +0000 (15:10 +0000)]
[project @ 2000-07-17 15:10:14 by rrt]
Use GMP dll rather than static library under mingwin.

23 years ago[project @ 2000-07-17 15:09:35 by rrt]
rrt [Mon, 17 Jul 2000 15:09:35 +0000 (15:09 +0000)]
[project @ 2000-07-17 15:09:35 by rrt]
Removed definitions of symbols which are now defined in the mingwin header
files.

23 years ago[project @ 2000-07-17 14:33:49 by keithw]
keithw [Mon, 17 Jul 2000 14:33:49 +0000 (14:33 +0000)]
[project @ 2000-07-17 14:33:49 by keithw]
Fix bug in SetLevels that meant an occurrence of a non-cloned binder,
where the non-cloned binder shadowed a cloned binder, resulted in
the occurrence incorrectly being substituted with the clone of the
*outer* binder.  Curious that this never got tickled before!

23 years ago[project @ 2000-07-17 11:28:32 by simonpj]
simonpj [Mon, 17 Jul 2000 11:28:32 +0000 (11:28 +0000)]
[project @ 2000-07-17 11:28:32 by simonpj]
Eliminate need for built-in Rational type

23 years ago[project @ 2000-07-17 11:28:00 by simonpj]
simonpj [Mon, 17 Jul 2000 11:28:00 +0000 (11:28 +0000)]
[project @ 2000-07-17 11:28:00 by simonpj]
Print operator names in HsExpr better

23 years ago[project @ 2000-07-17 11:15:07 by simonmar]
simonmar [Mon, 17 Jul 2000 11:15:08 +0000 (11:15 +0000)]
[project @ 2000-07-17 11:15:07 by simonmar]
Pass flags -agc to $(HAPPY) if the $(HAPPY_VERSION) is >= 1.8
(previous versions weren't reliable enough with these settings).

23 years ago[project @ 2000-07-17 10:05:59 by rrt]
rrt [Mon, 17 Jul 2000 10:05:59 +0000 (10:05 +0000)]
[project @ 2000-07-17 10:05:59 by rrt]
Merged Julian's changes to make it more comprehensible to non FP programmers.

23 years ago[project @ 2000-07-17 09:02:18 by simonmar]
simonmar [Mon, 17 Jul 2000 09:02:18 +0000 (09:02 +0000)]
[project @ 2000-07-17 09:02:18 by simonmar]
accept '-fvia-c' as an alternative capitalisation for '-fvia-C' (the
old driver used to do this, and it's documented in the Users' Guide)

23 years ago[project @ 2000-07-17 07:47:04 by simonpj]
simonpj [Mon, 17 Jul 2000 07:47:04 +0000 (07:47 +0000)]
[project @ 2000-07-17 07:47:04 by simonpj]
Add comments

23 years ago[project @ 2000-07-16 21:31:36 by panne]
panne [Sun, 16 Jul 2000 21:31:36 +0000 (21:31 +0000)]
[project @ 2000-07-16 21:31:36 by panne]
Document new -fstrictfp option

23 years ago[project @ 2000-07-16 21:10:48 by panne]
panne [Sun, 16 Jul 2000 21:10:48 +0000 (21:10 +0000)]
[project @ 2000-07-16 21:10:48 by panne]
This commit tries to fix the discrepancies between the results of
floating point calculations during runtime and compile time, see
e.g. fptools/ghc/tests/numeric/should_run/arith008.hs. Part of the
story was the fact that floating point values are represented as
Rationals in GHC and therefore never lost precision, at least for the
operations for which constant folding is done. To compensate for this,
before and after floating point operations the operands are
temporarily converted to/from Float/Double. This is wrong, because
host architecture and target architecture are confused this way, but
in a non-cross-compiling scenario it works.

23 years ago[project @ 2000-07-16 20:58:48 by panne]
panne [Sun, 16 Jul 2000 20:58:48 +0000 (20:58 +0000)]
[project @ 2000-07-16 20:58:48 by panne]
Added new -fstrictfp flag

23 years ago[project @ 2000-07-16 20:54:45 by panne]
panne [Sun, 16 Jul 2000 20:54:45 +0000 (20:54 +0000)]
[project @ 2000-07-16 20:54:45 by panne]
* Fixed handling of filename arguments without extension
* Added new -fstrictfp flag
* Some cosmetic changes (unless/when instead of if)

23 years ago[project @ 2000-07-15 15:09:45 by panne]
panne [Sat, 15 Jul 2000 15:09:45 +0000 (15:09 +0000)]
[project @ 2000-07-15 15:09:45 by panne]
Merged Marcin's patches for unboxed thingies

23 years ago[project @ 2000-07-14 23:54:06 by lewie]
lewie [Fri, 14 Jul 2000 23:54:06 +0000 (23:54 +0000)]
[project @ 2000-07-14 23:54:06 by lewie]
Functional Dependencies were not getting simplified away when the dictionary
that generated them was simplified by instance resolution.  Fixed.

23 years ago[project @ 2000-07-14 16:43:20 by simonpj]
simonpj [Fri, 14 Jul 2000 16:43:20 +0000 (16:43 +0000)]
[project @ 2000-07-14 16:43:20 by simonpj]
Add mutually recursive kind inference test

23 years ago[project @ 2000-07-14 13:38:55 by simonpj]
simonpj [Fri, 14 Jul 2000 13:38:55 +0000 (13:38 +0000)]
[project @ 2000-07-14 13:38:55 by simonpj]
Fix typeKind

23 years ago[project @ 2000-07-14 13:38:39 by simonpj]
simonpj [Fri, 14 Jul 2000 13:38:39 +0000 (13:38 +0000)]
[project @ 2000-07-14 13:38:39 by simonpj]
Arrange that type signatures work right.  Consider:

   module A
import M( f )
f :: Int -> Int
f x = x

Here, the 'f' in the signature isn't ambiguous; it
refers to the locally defined f.  (This isn't clear in
the Haskell 98 report, but it will be.)

23 years ago[project @ 2000-07-14 13:37:53 by simonpj]
simonpj [Fri, 14 Jul 2000 13:37:53 +0000 (13:37 +0000)]
[project @ 2000-07-14 13:37:53 by simonpj]
Wibbles in the new kind-checking stuff

23 years ago[project @ 2000-07-14 13:37:07 by simonpj]
simonpj [Fri, 14 Jul 2000 13:37:07 +0000 (13:37 +0000)]
[project @ 2000-07-14 13:37:07 by simonpj]
Add a type signature

23 years ago[project @ 2000-07-14 13:36:33 by simonpj]
simonpj [Fri, 14 Jul 2000 13:36:33 +0000 (13:36 +0000)]
[project @ 2000-07-14 13:36:33 by simonpj]
Tidy up newMutTyVar/newSigTyVar

23 years ago[project @ 2000-07-14 13:28:35 by simonmar]
simonmar [Fri, 14 Jul 2000 13:28:35 +0000 (13:28 +0000)]
[project @ 2000-07-14 13:28:35 by simonmar]
remove unused #include "gmp.h"

23 years ago[project @ 2000-07-14 09:34:02 by simonmar]
simonmar [Fri, 14 Jul 2000 09:34:02 +0000 (09:34 +0000)]
[project @ 2000-07-14 09:34:02 by simonmar]
Add a section describing GHC's version numbering policy.

23 years ago[project @ 2000-07-14 09:26:23 by simonmar]
simonmar [Fri, 14 Jul 2000 09:26:23 +0000 (09:26 +0000)]
[project @ 2000-07-14 09:26:23 by simonmar]
wibble

23 years ago[project @ 2000-07-14 09:03:57 by simonmar]
simonmar [Fri, 14 Jul 2000 09:03:57 +0000 (09:03 +0000)]
[project @ 2000-07-14 09:03:57 by simonmar]
- add a new section on "keeping intermediate files"
- document the new -keep-tmp-files and -keep-raw-s-file options.

23 years ago[project @ 2000-07-14 08:19:02 by simonmar]
simonmar [Fri, 14 Jul 2000 08:19:02 +0000 (08:19 +0000)]
[project @ 2000-07-14 08:19:02 by simonmar]
Add -keep-tmp-files flag, to tell the driver not to delete tmp files.
Useful mainly for debugging the driver, and obsoletes the GHC
folklore of using ^Z for this purpose.

23 years ago[project @ 2000-07-14 08:17:36 by simonpj]
simonpj [Fri, 14 Jul 2000 08:17:41 +0000 (08:17 +0000)]
[project @ 2000-07-14 08:17:36 by simonpj]
This commit completely re-does the kind-inference mechanism.
Previously it was inter-wound with type inference, but that was
always hard to understand, and it finally broke when we started
checking for ambiguity when type-checking a type signature (details
irrelevant).

So now kind inference is more clearly separated, so that it never
takes place at the same time as type inference.  The biggest change
is in TcTyClsDecls, which does the kind inference for a group of
type and class declarations.  It now contains comments to explain
how it all works.

There are also comments in TypeRep which describes the slightly
tricky way in which we deal with the fact that kind 'type' (written
'*') actually has 'boxed type' and 'unboxed type' as sub-kinds.
The whole thing is a bit of a hack, because we don't really have
sub-kinding, but it's less of a hack than before.

A lot of general tidying up happened at the same time.
In particular, I removed some dead code here and there

23 years ago[project @ 2000-07-14 08:14:53 by simonpj]
simonpj [Fri, 14 Jul 2000 08:16:55 +0000 (08:16 +0000)]
[project @ 2000-07-14 08:14:53 by simonpj]
Remove dead code

23 years ago[project @ 2000-07-13 15:45:55 by simonmar]
simonmar [Thu, 13 Jul 2000 15:45:55 +0000 (15:45 +0000)]
[project @ 2000-07-13 15:45:55 by simonmar]
Take the README for a binary dist from the project directory, not fptools.

23 years ago[project @ 2000-07-13 15:44:40 by simonmar]
simonmar [Thu, 13 Jul 2000 15:44:40 +0000 (15:44 +0000)]
[project @ 2000-07-13 15:44:40 by simonmar]
update, and make less GHC-specific

23 years ago[project @ 2000-07-13 14:45:13 by simonmar]
simonmar [Thu, 13 Jul 2000 14:45:13 +0000 (14:45 +0000)]
[project @ 2000-07-13 14:45:13 by simonmar]
update to match reality; try to be a bit more helpful

23 years ago[project @ 2000-07-13 09:22:26 by sewardj]
sewardj [Thu, 13 Jul 2000 09:22:26 +0000 (09:22 +0000)]
[project @ 2000-07-13 09:22:26 by sewardj]
Zap yet another bogon in x86 floating args to ccalls.

23 years ago[project @ 2000-07-13 09:17:57 by michaelw]
michaelw [Thu, 13 Jul 2000 09:17:57 +0000 (09:17 +0000)]
[project @ 2000-07-13 09:17:57 by michaelw]
more changes to FPTOOLS_GHC_VERSION:
* fixed bug in too-many-args case
* reworked regexp for versions

23 years ago[project @ 2000-07-13 08:56:42 by sewardj]
sewardj [Thu, 13 Jul 2000 08:56:42 +0000 (08:56 +0000)]
[project @ 2000-07-13 08:56:42 by sewardj]
sparc NCG is now enabled by default for non-opt compilation.

23 years ago[project @ 2000-07-13 08:49:24 by sewardj]
sewardj [Thu, 13 Jul 2000 08:49:24 +0000 (08:49 +0000)]
[project @ 2000-07-13 08:49:24 by sewardj]
Reg needs to be imported non-abstractly for sparc.

23 years ago[project @ 2000-07-12 13:04:31 by sewardj]
sewardj [Wed, 12 Jul 2000 13:04:31 +0000 (13:04 +0000)]
[project @ 2000-07-12 13:04:31 by sewardj]
Make the x86 NCG work again following recent sparc hackage.

Also, fix the x86 bits pertaining to the floats-promoted-to-doubles-
in-ccalls problem.  So this problem should no longer exist on x86
or sparc via NCG.

23 years ago[project @ 2000-07-12 09:58:49 by simonmar]
simonmar [Wed, 12 Jul 2000 09:58:49 +0000 (09:58 +0000)]
[project @ 2000-07-12 09:58:49 by simonmar]
use Maybe.isJust instead of proprietary verison.

23 years ago[project @ 2000-07-11 23:02:39 by michaelw]
michaelw [Tue, 11 Jul 2000 23:02:39 +0000 (23:02 +0000)]
[project @ 2000-07-11 23:02:39 by michaelw]
added Debian location to DocBook catalog test

23 years ago[project @ 2000-07-11 21:29:46 by michaelw]
michaelw [Tue, 11 Jul 2000 21:29:46 +0000 (21:29 +0000)]
[project @ 2000-07-11 21:29:46 by michaelw]
typo

23 years ago[project @ 2000-07-11 21:25:37 by michaelw]
michaelw [Tue, 11 Jul 2000 21:25:37 +0000 (21:25 +0000)]
[project @ 2000-07-11 21:25:37 by michaelw]
updated comments

23 years ago[project @ 2000-07-11 20:45:29 by panne]
panne [Tue, 11 Jul 2000 20:45:29 +0000 (20:45 +0000)]
[project @ 2000-07-11 20:45:29 by panne]
Hash MachLabel, too

23 years ago[project @ 2000-07-11 19:28:34 by panne]
panne [Tue, 11 Jul 2000 19:28:34 +0000 (19:28 +0000)]
[project @ 2000-07-11 19:28:34 by panne]
Replace RcFloating with RcFloat/RcDouble. I'm not exactly sure what
I'm doing here, so somebody else should better have a look.

23 years ago[project @ 2000-07-11 19:17:20 by panne]
panne [Tue, 11 Jul 2000 19:17:20 +0000 (19:17 +0000)]
[project @ 2000-07-11 19:17:20 by panne]
* linux_TARGET_ARCH => i386_TARGET_ARCH
* move callClobberedRegs into safety in front of the register
  #define-orgy (what the ... is this cpp-ery for? %-]

23 years ago[project @ 2000-07-11 16:41:19 by simonmar]
simonmar [Tue, 11 Jul 2000 16:41:19 +0000 (16:41 +0000)]
[project @ 2000-07-11 16:41:19 by simonmar]
reinstate used import

23 years ago[project @ 2000-07-11 16:24:57 by simonmar]
simonmar [Tue, 11 Jul 2000 16:38:50 +0000 (16:38 +0000)]
[project @ 2000-07-11 16:24:57 by simonmar]
remove unused imports

23 years ago[project @ 2000-07-11 16:12:11 by simonmar]
simonmar [Tue, 11 Jul 2000 16:12:59 +0000 (16:12 +0000)]
[project @ 2000-07-11 16:12:11 by simonmar]
remove unused imports

23 years ago[project @ 2000-07-11 16:04:38 by simonmar]
simonmar [Tue, 11 Jul 2000 16:04:38 +0000 (16:04 +0000)]
[project @ 2000-07-11 16:04:38 by simonmar]
remove unused imports

23 years ago[project @ 2000-07-11 16:03:37 by simonmar]
simonmar [Tue, 11 Jul 2000 16:03:37 +0000 (16:03 +0000)]
[project @ 2000-07-11 16:03:37 by simonmar]
remove unused imports; misc cleanup

23 years ago[project @ 2000-07-11 15:58:26 by simonmar]
simonmar [Tue, 11 Jul 2000 15:58:26 +0000 (15:58 +0000)]
[project @ 2000-07-11 15:58:26 by simonmar]
remove unused imports; misc cleanup

23 years ago[project @ 2000-07-11 15:57:45 by simonmar]
simonmar [Tue, 11 Jul 2000 15:57:45 +0000 (15:57 +0000)]
[project @ 2000-07-11 15:57:45 by simonmar]
remove unused imports

23 years ago[project @ 2000-07-11 15:57:11 by simonmar]
simonmar [Tue, 11 Jul 2000 15:57:11 +0000 (15:57 +0000)]
[project @ 2000-07-11 15:57:11 by simonmar]
use MachLabel rather than MachLitLit for compiling foreign label and
foreign export dynamic.