ghc-hetmet.git
19 years ago[project @ 2004-08-25 16:40:48 by sof]
sof [Wed, 25 Aug 2004 16:40:48 +0000 (16:40 +0000)]
[project @ 2004-08-25 16:40:48 by sof]
FP_READDIR_EOF_ERRNO: better use AC_DEFINE_UNQUOTED here

19 years ago[project @ 2004-08-25 10:37:06 by simonmar]
simonmar [Wed, 25 Aug 2004 10:37:06 +0000 (10:37 +0000)]
[project @ 2004-08-25 10:37:06 by simonmar]
fix braino in previous commit: we should cast the function result to
the right type, not just assume (W_).

19 years ago[project @ 2004-08-25 10:33:52 by simonpj]
simonpj [Wed, 25 Aug 2004 10:33:52 +0000 (10:33 +0000)]
[project @ 2004-08-25 10:33:52 by simonpj]
Fix typo that caused non -O to see unfoldings

19 years ago[project @ 2004-08-25 10:33:26 by simonpj]
simonpj [Wed, 25 Aug 2004 10:33:26 +0000 (10:33 +0000)]
[project @ 2004-08-25 10:33:26 by simonpj]
Spelling mistake in comment

19 years ago[project @ 2004-08-24 19:26:45 by sof]
sof [Tue, 24 Aug 2004 19:26:47 +0000 (19:26 +0000)]
[project @ 2004-08-24 19:26:45 by sof]
Pair of new macros for checking how dirent.h is implemented. Needed
to deal with versioning issues under mingw only, but won't do
any harm on other plats. The macros are:

- FP_READDIR_EOF_ERRNO
     checks what readdir() sets errno to upon reaching end of a directory.
     (that value is available as READDIR_ERRNO_EOF in mk/config.h). Up
     until recently, mingw's readdir() did set errno rather than just
     leaving it alone.
- FP_DIRENT_FLAT_LAYOUT
     checks how 'struct dirent' is defined; defines
     config.h:STRUCT_DIRENT_FLAT_LAYOUT to 1 if 'struct dirent' is terminated
     with a d_name array. Up until recently, mingw didn't define it as such.

To be hooked up to libraries/base/{include/HsBase.h,cbits/dirUtils.c} and
quite possibly merged to STABLE.

19 years ago[project @ 2004-08-23 10:11:23 by simonmar]
simonmar [Mon, 23 Aug 2004 10:11:24 +0000 (10:11 +0000)]
[project @ 2004-08-23 10:11:23 by simonmar]
Fix an infinite loop in the cmm-optimiser in the native codegen, and
refactor: move isAssociativeMachOp into MachOp.

19 years ago[project @ 2004-08-22 16:40:38 by panne]
panne [Sun, 22 Aug 2004 16:40:38 +0000 (16:40 +0000)]
[project @ 2004-08-22 16:40:38 by panne]
Terminate program if execPage fails, this is more honest and
simplifies things a bit.

19 years ago[project @ 2004-08-22 16:34:54 by panne]
panne [Sun, 22 Aug 2004 16:34:54 +0000 (16:34 +0000)]
[project @ 2004-08-22 16:34:54 by panne]
Ooops, fixed previous commit...

19 years ago[project @ 2004-08-22 16:27:50 by panne]
panne [Sun, 22 Aug 2004 16:27:50 +0000 (16:27 +0000)]
[project @ 2004-08-22 16:27:50 by panne]
stgMallocBytes never returns NULL

19 years ago[project @ 2004-08-22 16:20:42 by panne]
panne [Sun, 22 Aug 2004 16:20:42 +0000 (16:20 +0000)]
[project @ 2004-08-22 16:20:42 by panne]
* Renamed __obscure_ccall_ret_code to obscure_ccall_ret_code, leading
  underscores are bad for user code.

* Let the code for obscure_ccall_ret_code live in dynamically
  allocated memory, just like the adjustors. This should fix problems
  on OpenBSD, which doesn't like executable code in the data section.

19 years ago[project @ 2004-08-22 15:50:39 by panne]
panne [Sun, 22 Aug 2004 15:50:42 +0000 (15:50 +0000)]
[project @ 2004-08-22 15:50:39 by panne]
Removed a few "use of cast expressions as lvalues is deprecated"
warnings. Note that gcc is really pedantic nowadays, so we have to go
via a temporary to avoid "cast does not match function type" warnings.

19 years ago[project @ 2004-08-21 13:48:58 by panne]
panne [Sat, 21 Aug 2004 13:48:59 +0000 (13:48 +0000)]
[project @ 2004-08-21 13:48:58 by panne]
If "hhc" is in the PATH, the "htmlhelp" target builds a *.chm file now. Note:
This is untested and "hhc" should really be detected via autoconf.

19 years ago[project @ 2004-08-21 13:29:26 by panne]
panne [Sat, 21 Aug 2004 13:29:28 +0000 (13:29 +0000)]
[project @ 2004-08-21 13:29:26 by panne]
* Factored out labeling options (used by FO now, too)
* Cleaned up stylesheet handling

19 years ago[project @ 2004-08-21 12:49:14 by panne]
panne [Sat, 21 Aug 2004 12:49:14 +0000 (12:49 +0000)]
[project @ 2004-08-21 12:49:14 by panne]
* Header cleanup
* Improved type of execPage
* Use prog_belch instead of fprintf(stderr, ...)

19 years ago[project @ 2004-08-21 12:48:00 by panne]
panne [Sat, 21 Aug 2004 12:48:00 +0000 (12:48 +0000)]
[project @ 2004-08-21 12:48:00 by panne]
Perform initialisation of adjustor thunk layer at startup.

19 years ago[project @ 2004-08-21 12:47:17 by panne]
panne [Sat, 21 Aug 2004 12:47:17 +0000 (12:47 +0000)]
[project @ 2004-08-21 12:47:17 by panne]
Moved createAdjustor and freeHaskellFunctionPtr to a header visible in
*.hc code.  The whole header layout is a little bit baroque, IMHO...

19 years ago[project @ 2004-08-20 15:02:40 by simonmar]
simonmar [Fri, 20 Aug 2004 15:02:40 +0000 (15:02 +0000)]
[project @ 2004-08-20 15:02:40 by simonmar]
Hack to work around bug on Windows reported by Levent Erkok
<levent.erkok@intel.com>:

Prelude> let x=x in x
*** Exception: <<loop>>
Prelude> 3
Fail: thread blocked indefinitely

Details in the comment.

MERGE TO STABLE

19 years ago[project @ 2004-08-20 13:05:50 by simonmar]
simonmar [Fri, 20 Aug 2004 13:05:50 +0000 (13:05 +0000)]
[project @ 2004-08-20 13:05:50 by simonmar]
Fixes

19 years ago[project @ 2004-08-20 12:21:03 by simonmar]
simonmar [Fri, 20 Aug 2004 12:21:05 +0000 (12:21 +0000)]
[project @ 2004-08-20 12:21:03 by simonmar]
Simplify the "impossible branch" handling, and fix a bug in the
process.  CmmSwitch encodes the possibility of having impossible
branches (the destinations are Maybe BlockId rather than just BlockId)
so we don't need to encode impossible branches as dummy blocks
containing a jump to an impossible location (currently 0).

However, PprC and PprCmm weren't set up to cope with Nothings in a
CmmSwitch, so this commit fixes that too.

19 years ago[project @ 2004-08-20 11:20:16 by simonmar]
simonmar [Fri, 20 Aug 2004 11:20:17 +0000 (11:20 +0000)]
[project @ 2004-08-20 11:20:16 by simonmar]
Give literal string labels a _str suffix, to make it less likely that
they'll clash with a symbol in scope in a C file.

19 years ago[project @ 2004-08-19 15:43:39 by simonmar]
simonmar [Thu, 19 Aug 2004 15:43:39 +0000 (15:43 +0000)]
[project @ 2004-08-19 15:43:39 by simonmar]
Build fix

19 years ago[project @ 2004-08-19 14:00:48 by simonmar]
simonmar [Thu, 19 Aug 2004 14:00:50 +0000 (14:00 +0000)]
[project @ 2004-08-19 14:00:48 by simonmar]
Windows fixes.

19 years ago[project @ 2004-08-19 11:27:45 by simonmar]
simonmar [Thu, 19 Aug 2004 11:27:45 +0000 (11:27 +0000)]
[project @ 2004-08-19 11:27:45 by simonmar]
bugfix: set tso->link = END_TSO_QUEUE before APPEND_TO_RUN_QUEUE().

19 years ago[project @ 2004-08-18 19:12:42 by sof]
sof [Wed, 18 Aug 2004 19:12:42 +0000 (19:12 +0000)]
[project @ 2004-08-18 19:12:42 by sof]
- set mingw_top, old_ghc_top, gcc_version defaults if not
  defined in the environment block.
- bundle ar.exe (as bin/ar.exe)
- more comments

19 years ago[project @ 2004-08-18 16:42:51 by panne]
panne [Wed, 18 Aug 2004 16:42:56 +0000 (16:42 +0000)]
[project @ 2004-08-18 16:42:51 by panne]
Re-added indices + minor fixes

19 years ago[project @ 2004-08-18 16:23:14 by panne]
panne [Wed, 18 Aug 2004 16:23:14 +0000 (16:23 +0000)]
[project @ 2004-08-18 16:23:14 by panne]
Improved section labeling for HTML formats. Have to check other formats...

19 years ago[project @ 2004-08-18 16:16:54 by panne]
panne [Wed, 18 Aug 2004 16:16:57 +0000 (16:16 +0000)]
[project @ 2004-08-18 16:16:54 by panne]
* Preliminary support for HTML Help (target "htmlhelp". Note: No
  support for *.chm yet, I have to download hhc first for testing.

* Made directory/file layout more consistent:
     * All-in-one HTML for foo.xml goes into foo.html
     * Chunked HTML for foo.xml goes into foo-html directory with root index.html
     * HTML Help for foo.xml goes into foo-htmlhelp directory
  For the rationale behind this, read "Cool URIs don't change" at
  http://www.w3.org/Provider/Style/URI.

* Renamed "no-chunks-html" make target to "html-no-chunks"

19 years ago[project @ 2004-08-18 15:47:55 by panne]
panne [Wed, 18 Aug 2004 15:47:55 +0000 (15:47 +0000)]
[project @ 2004-08-18 15:47:55 by panne]
Use *nix instead of Windoze line ends

19 years ago[project @ 2004-08-18 13:16:28 by simonmar]
simonmar [Wed, 18 Aug 2004 13:16:28 +0000 (13:16 +0000)]
[project @ 2004-08-18 13:16:28 by simonmar]
Fix parse error in the generated package.conf

19 years ago[project @ 2004-08-18 13:14:21 by simonmar]
simonmar [Wed, 18 Aug 2004 13:14:21 +0000 (13:14 +0000)]
[project @ 2004-08-18 13:14:21 by simonmar]
Add cmm directory

19 years ago[project @ 2004-08-18 10:07:29 by simonmar]
simonmar [Wed, 18 Aug 2004 10:07:29 +0000 (10:07 +0000)]
[project @ 2004-08-18 10:07:29 by simonmar]
Remove extra semicolon; exposed by other recent changes.

19 years ago[project @ 2004-08-18 09:33:03 by simonpj]
simonpj [Wed, 18 Aug 2004 09:33:11 +0000 (09:33 +0000)]
[project @ 2004-08-18 09:33:03 by simonpj]
-------------------------------
Fix a fundep bug
-------------------------------

MERGE TO STABLE

Big thank-you to Martin Sulzmann for finding this functional dependency bug.

The new defn of FunDeps.Equation is:
type Equation = (TyVarSet, [(Type, Type)])
Before it was (TyVarSet, Type, Type), so each pair of types was separately
quantified (wrong).

It's important that we have a *list* of pairs of types.  Consider
  class C a b c | a -> b c where ...
instance C Int x x where ...
Then, given the constraint (C Int Bool v) we should improve v to Bool,
via the equation ({x}, [(Bool,x), (v,x)])
This would not happen if the class had looked like
class C a b c | a -> b, a -> c

Test in typecheck/should_compile/tc180

19 years ago[project @ 2004-08-17 16:00:58 by simonmar]
simonmar [Tue, 17 Aug 2004 16:00:59 +0000 (16:00 +0000)]
[project @ 2004-08-17 16:00:58 by simonmar]
- phases.xml: add documentation on <os>_TARGET_OS and <arch>_TARGET_ARCH
  symbols.

- some fixes for editing in emacs

19 years ago[project @ 2004-08-17 15:23:47 by simonpj]
simonpj [Tue, 17 Aug 2004 15:24:13 +0000 (15:24 +0000)]
[project @ 2004-08-17 15:23:47 by simonpj]
-------------------------------
Use merge-sort not quicksort
Nuke quicksort altogether
-------------------------------

Quicksort has O(n**2) behaviour worst case, and this occasionally bites.
In particular, when compiling large files consisting only of static data,
we get loads of top-level delarations -- and that led to more than half the
total compile time being spent in the strongly connected component analysis
for the occurrence analyser.  Switching to merge sort completely solved the
problem.

I've nuked quicksort altogether to make sure this does not happen again.

19 years ago[project @ 2004-08-17 08:18:41 by simonpj]
simonpj [Tue, 17 Aug 2004 08:18:41 +0000 (08:18 +0000)]
[project @ 2004-08-17 08:18:41 by simonpj]
Typo in windows-specific code

19 years ago[project @ 2004-08-16 19:59:28 by panne]
panne [Mon, 16 Aug 2004 19:59:53 +0000 (19:59 +0000)]
[project @ 2004-08-16 19:59:28 by panne]
XMLification

19 years ago[project @ 2004-08-16 16:26:51 by simonpj]
simonpj [Mon, 16 Aug 2004 16:26:51 +0000 (16:26 +0000)]
[project @ 2004-08-16 16:26:51 by simonpj]
-------------------------------
Add instance information to :i
  Get rid of the DeclPool
-------------------------------

1.  Add instance information to :info command.  GHCi now prints out
    which instances a type or class belongs to, when you use :i

2.  Tidy up printing of unqualified names in user output.
    Previously Outputable.PrintUnqualified was
type PrintUnqualified = Name -> Bool
    but it's now
type PrintUnqualified = ModuleName -> OccName -> Bool
    This turns out to be tidier even for Names, and it's now also usable
    when printing IfaceSyn stuff in GHCi, eliminating a grevious hack.

3.  On the way to doing this, Simon M had the great idea that we could
    get rid of the DeclPool holding pen, which held declarations read from
    interface files but not yet type-checked.   We do this by eagerly
    populating the TypeEnv with thunks what, when poked, do the type
    checking.   This is just a logical continuation of lazy import
    mechanism we've now had for some while.

The InstPool and RulePool still exist, but I plan to get rid of them in
the same way.  The new scheme does mean that more rules get sucked in than
before, because previously the TypeEnv was used to mean "this thing was needed"
and hence to control which rules were sucked in.  But now the TypeEnv is
populated more eagerly => more rules get sucked in.  However this problem
will go away when I get rid of the Inst and Rule pools.

I should have kept these changes separate, but I didn't.  Change (1)
affects mainly
TcRnDriver, HscMain, CompMan, InteractiveUI
whereas change (3) is more wide ranging.

19 years ago[project @ 2004-08-16 16:14:59 by simonpj]
simonpj [Mon, 16 Aug 2004 16:14:59 +0000 (16:14 +0000)]
[project @ 2004-08-16 16:14:59 by simonpj]
stg_interp_constr_entry and friends don't have info tables, so declare
them using RTS_FUN() rather than RTS_ENTRY().

19 years ago[project @ 2004-08-16 11:57:19 by simonmar]
simonmar [Mon, 16 Aug 2004 11:57:19 +0000 (11:57 +0000)]
[project @ 2004-08-16 11:57:19 by simonmar]
Avoid casting lvalues; gcc 3.4 issues warnings for it.

19 years ago[project @ 2004-08-16 11:19:24 by simonmar]
simonmar [Mon, 16 Aug 2004 11:19:27 +0000 (11:19 +0000)]
[project @ 2004-08-16 11:19:24 by simonmar]
Define <arch>_TARGET_ARCH and <os>_TARGET_OS by default when CPP'ing.
This avoids the main reason for needing to #include config.h into
Haskell source, so most files that previously just #include "config.h" can now

  #if __GLASGOW_HASKELL__ < 603
  #include "config.h"
  #endif

19 years ago[project @ 2004-08-16 11:17:20 by simonmar]
simonmar [Mon, 16 Aug 2004 11:17:20 +0000 (11:17 +0000)]
[project @ 2004-08-16 11:17:20 by simonmar]
Add a config.h to ease the transition.  Now if you include config.h
you get a warning instead of a failure:

In file included from System/Posix/Internals.hs:24:
/playpen/ghc/nightly/HEAD-cam-02-unx/i386-unknown-linux/ghc/includes/config.h:4: warning: #warning config.h is deprecated; please use ghcconfig.h instead

19 years ago[project @ 2004-08-16 10:59:37 by simonmar]
simonmar [Mon, 16 Aug 2004 10:59:37 +0000 (10:59 +0000)]
[project @ 2004-08-16 10:59:37 by simonmar]
config.h ==> ghcconfig.h

19 years ago[project @ 2004-08-16 09:53:47 by simonpj]
simonpj [Mon, 16 Aug 2004 09:54:51 +0000 (09:54 +0000)]
[project @ 2004-08-16 09:53:47 by simonpj]
-------------------------------
Add instance information to :i
  Get rid of the DeclPool
-------------------------------

1.  Add instance information to :info command.  GHCi now prints out
    which instances a type or class belongs to, when you use :i

2.  Tidy up printing of unqualified names in user output.
    Previously Outputable.PrintUnqualified was
type PrintUnqualified = Name -> Bool
    but it's now
type PrintUnqualified = ModuleName -> OccName -> Bool
    This turns out to be tidier even for Names, and it's now also usable
    when printing IfaceSyn stuff in GHCi, eliminating a grevious hack.

3.  On the way to doing this, Simon M had the great idea that we could
    get rid of the DeclPool holding pen, which held declarations read from
    interface files but not yet type-checked.   We do this by eagerly
    populating the TypeEnv with thunks what, when poked, do the type
    checking.   This is just a logical continuation of lazy import
    mechanism we've now had for some while.

The InstPool and RulePool still exist, but I plan to get rid of them in
the same way.  The new scheme does mean that more rules get sucked in than
before, because previously the TypeEnv was used to mean "this thing was needed"
and hence to control which rules were sucked in.  But now the TypeEnv is
populated more eagerly => more rules get sucked in.  However this problem
will go away when I get rid of the Inst and Rule pools.

I should have kept these changes separate, but I didn't.  Change (1)
affects mainly
TcRnDriver, HscMain, CompMan, InteractiveUI
whereas change (3) is more wide ranging.

19 years ago[project @ 2004-08-16 09:51:20 by simonpj]
simonpj [Mon, 16 Aug 2004 09:51:20 +0000 (09:51 +0000)]
[project @ 2004-08-16 09:51:20 by simonpj]
Commments only

19 years ago[project @ 2004-08-16 09:41:19 by simonpj]
simonpj [Mon, 16 Aug 2004 09:41:19 +0000 (09:41 +0000)]
[project @ 2004-08-16 09:41:19 by simonpj]
Fix problem with :i for (:)

19 years ago[project @ 2004-08-16 09:31:50 by simonmar]
simonmar [Mon, 16 Aug 2004 09:31:50 +0000 (09:31 +0000)]
[project @ 2004-08-16 09:31:50 by simonmar]
#ifdefery to work around change in name of config.h.

19 years ago[project @ 2004-08-16 07:24:25 by panne]
panne [Mon, 16 Aug 2004 07:24:29 +0000 (07:24 +0000)]
[project @ 2004-08-16 07:24:25 by panne]
Correct DocBook XML now

19 years ago[project @ 2004-08-16 06:44:47 by panne]
panne [Mon, 16 Aug 2004 06:44:47 +0000 (06:44 +0000)]
[project @ 2004-08-16 06:44:47 by panne]
Fixed default font-family, color, and background. It contained a typo
and didn't work for chunked documentation.

19 years ago[project @ 2004-08-15 20:37:22 by panne]
panne [Sun, 15 Aug 2004 20:38:45 +0000 (20:38 +0000)]
[project @ 2004-08-15 20:37:22 by panne]
Started to convert the users guide to DocBook XML. Not yet finished, there are
still *tons* of misplaced indexterms, but some sensible documentation can
already be generated.

19 years ago[project @ 2004-08-15 20:32:47 by panne]
panne [Sun, 15 Aug 2004 20:33:05 +0000 (20:33 +0000)]
[project @ 2004-08-15 20:32:47 by panne]
Converted the building guide to DocBook XML

19 years ago[project @ 2004-08-15 20:28:02 by panne]
panne [Sun, 15 Aug 2004 20:28:08 +0000 (20:28 +0000)]
[project @ 2004-08-15 20:28:02 by panne]
<fanfare>Finally: Support for DocBook XML!</fanfare>

After endless frustrating hours, I came to the conclusion that using (Open)Jade
for DocBook XML is virtually impossible, there are *tons* of problems with it:
Wildly differing paths, incorrect stylesheets, broken catalogs at funny places,
broken wrapper scripts, missing support on Cygwin etc.  >:-P * * *

So we follow the ubiquitous XSL hype and use xsltproc + DocBook XSL stylesheets
to transform DocBook XML to HTML and FO. From FO there are two routes to PDF and
PostScript: Either via FOP (preferred) or via PassiveTeX. Validation can be done
via xmllint using the new make target "validate". When PassiveTeX is available,
DVI can be generated, too.  A new make target "no-chunks-html" is available for
generating an all-in-one HTML document. Currently there is no way to generate
plain text or RTF, but at least the former could easily be added.

Generating HTML works out-of-the-box on Cygwin now, but you will have to install
FOP for other formats, because there seems to be no standard package for it.

The HTML appearance can be controlled via CSS, but the current location of the
master stylesheet below fptools/mk is a bit debatable. Better suggestions are
welcome.

Currently there is still support for SGML documents, but it will be dropped when
all documents are converted to DocBook XML. The build system is complex enough
already with the support for a *single* kind of DocBook...

19 years ago[project @ 2004-08-13 14:22:16 by simonmar]
simonmar [Fri, 13 Aug 2004 14:22:16 +0000 (14:22 +0000)]
[project @ 2004-08-13 14:22:16 by simonmar]
Mention that this doc is out of date.  It's pretty huge, so I'm not
going to update it right now.

19 years ago[project @ 2004-08-13 13:59:12 by simonmar]
simonmar [Fri, 13 Aug 2004 13:59:12 +0000 (13:59 +0000)]
[project @ 2004-08-13 13:59:12 by simonmar]
Fixes to previous

19 years ago[project @ 2004-08-13 13:57:04 by simonmar]
simonmar [Fri, 13 Aug 2004 13:57:08 +0000 (13:57 +0000)]
[project @ 2004-08-13 13:57:04 by simonmar]
Various fixes and removal of dead code.

19 years ago[project @ 2004-08-13 13:39:29 by simonmar]
simonmar [Fri, 13 Aug 2004 13:39:29 +0000 (13:39 +0000)]
[project @ 2004-08-13 13:39:29 by simonmar]
forgot to add this file

19 years ago[project @ 2004-08-13 13:30:44 by simonmar]
simonmar [Fri, 13 Aug 2004 13:30:44 +0000 (13:30 +0000)]
[project @ 2004-08-13 13:30:44 by simonmar]
config.h ==> ghcconfig.h

19 years ago[project @ 2004-08-13 13:26:37 by simonmar]
simonmar [Fri, 13 Aug 2004 13:26:37 +0000 (13:26 +0000)]
[project @ 2004-08-13 13:26:37 by simonmar]
sIZEOF_StgFunInfoExtra ==> sIZEOF_StgFunInfoExtraRev

19 years ago[project @ 2004-08-13 13:25:45 by simonmar]
simonmar [Fri, 13 Aug 2004 13:25:45 +0000 (13:25 +0000)]
[project @ 2004-08-13 13:25:45 by simonmar]
Fix minor merge-o

19 years ago[project @ 2004-08-13 13:04:50 by simonmar]
simonmar [Fri, 13 Aug 2004 13:11:23 +0000 (13:11 +0000)]
[project @ 2004-08-13 13:04:50 by simonmar]
Merge backend-hacking-branch onto HEAD.  Yay!

19 years ago[project @ 2004-08-13 10:45:16 by simonmar]
simonmar [Fri, 13 Aug 2004 10:45:16 +0000 (10:45 +0000)]
[project @ 2004-08-13 10:45:16 by simonmar]
Work around stricter CPP in gcc 3.4.1

19 years ago[project @ 2004-08-13 10:42:34 by simonmar]
simonmar [Fri, 13 Aug 2004 10:42:34 +0000 (10:42 +0000)]
[project @ 2004-08-13 10:42:34 by simonmar]
Fix to previous commit

19 years ago[project @ 2004-08-12 13:11:35 by simonmar]
simonmar [Thu, 12 Aug 2004 13:11:35 +0000 (13:11 +0000)]
[project @ 2004-08-12 13:11:35 by simonmar]
wibble

19 years ago[project @ 2004-08-12 13:10:35 by simonmar]
simonmar [Thu, 12 Aug 2004 13:10:42 +0000 (13:10 +0000)]
[project @ 2004-08-12 13:10:35 by simonmar]
avoid string gaps

19 years ago[project @ 2004-08-12 13:06:51 by simonmar]
simonmar [Thu, 12 Aug 2004 13:06:51 +0000 (13:06 +0000)]
[project @ 2004-08-12 13:06:51 by simonmar]
Avoid string gaps

19 years ago[project @ 2004-08-12 12:16:53 by simonmar]
simonmar [Thu, 12 Aug 2004 12:16:53 +0000 (12:16 +0000)]
[project @ 2004-08-12 12:16:53 by simonmar]
Avoid string gaps.

19 years ago[project @ 2004-08-12 12:12:54 by simonmar]
simonmar [Thu, 12 Aug 2004 12:12:55 +0000 (12:12 +0000)]
[project @ 2004-08-12 12:12:54 by simonmar]
Avoid string gaps in CPP'd source.

19 years ago[project @ 2004-08-12 11:19:39 by simonmar]
simonmar [Thu, 12 Aug 2004 11:19:44 +0000 (11:19 +0000)]
[project @ 2004-08-12 11:19:39 by simonmar]
Take the timestamp of an interpreted linkable from the timestamp of
the source file.  This works better when the local clock is out of
sync with the filesystem, and it's just as accurate.

19 years ago[project @ 2004-08-12 10:45:51 by simonmar]
simonmar [Thu, 12 Aug 2004 10:45:51 +0000 (10:45 +0000)]
[project @ 2004-08-12 10:45:51 by simonmar]
Finally fix this problem:

make: *** No rule to make target `Text/Regex/Posix_hsc_make.c', needed by `depend'.  Stop.

caused by temporary droppings from hsc2hs.

19 years ago[project @ 2004-08-10 14:36:25 by simonmar]
simonmar [Tue, 10 Aug 2004 14:36:25 +0000 (14:36 +0000)]
[project @ 2004-08-10 14:36:25 by simonmar]
Windows fixes.  There were some strange things going on with Windows
installation directories - I'm sure I've mucked up whatever was
happening with 'make install' for GHC on Windows, but there were
defintitely some bugs in here.  eg. we shouldn't have been setting
datadir to $(prefix)/imports unconditionally on Windows.

Make install now works for Happy, Alex and Haddock.  GHC probably
needs a bit of tweaking to get it working again.

19 years ago[project @ 2004-08-10 13:51:25 by simonmar]
simonmar [Tue, 10 Aug 2004 13:51:25 +0000 (13:51 +0000)]
[project @ 2004-08-10 13:51:25 by simonmar]
Add $(Windows), which is "YES" if the host platform is a Windows
platform (mingw32 or cygwin32), and "NO" otherwise.

19 years ago[project @ 2004-08-10 09:02:36 by simonmar]
simonmar [Tue, 10 Aug 2004 09:02:41 +0000 (09:02 +0000)]
[project @ 2004-08-10 09:02:36 by simonmar]
Fix problem with inline foreign-call changes yesterday.  Foreign call
args sometimes have to be modified using shimFCallArg - nowadays this
is done at code generation time, whereas it used to be done at
pretty-printing time.

19 years ago[project @ 2004-08-09 14:27:53 by simonmar]
simonmar [Mon, 9 Aug 2004 14:27:53 +0000 (14:27 +0000)]
[project @ 2004-08-09 14:27:53 by simonmar]
Two changes:

  - newly created or unblocked threads go to the back of the run
    queue.  This is fairer because it avoids possible starvation.
    Fixes problem with conc004 test going much slower with optimisation.

  - The context switch flag is only reset after a yield.  Previously,
    it was being reset before we ran a thread, which meant that a tick
    that arrived during GC would be lost, and a thread doing lots of
    GC could hog the CPU.  Now, GC is effectively counted as part of
    the last running thread's timeslice.

MERGE TO STABLE.

19 years ago[project @ 2004-08-09 13:19:29 by simonmar]
simonmar [Mon, 9 Aug 2004 13:19:29 +0000 (13:19 +0000)]
[project @ 2004-08-09 13:19:29 by simonmar]
Allow case-of-unsafe-ccall to compile to straight-line code, like it
used to.  This has already been fixed on the backend-hacking-branch,
but I'm doing it here so that it can be merged into the STABLE branch,
where it will help to work around a bug.

The bug is in CgExpr.lhs:primRetUnboxedTuple, which picks temporaries
to assign the result of a ccall to.  The Cg monad doesn't have a uniq
supply (in the HEAD), so we always pick the same temporaries.  This
leads to clashes in complex function with multiple ccalls.  Again,
this is already fixed in the backend-hacking-branch.  I don't see an
easy fix for this bug.

The compilation of case-of-unsafe-ccall doesn't suffer from this
problem, and it will help work around some cases of the bug, so I'm
going to merge this onto the STABLE branch after some testing.

19 years ago[project @ 2004-08-09 12:35:02 by simonmar]
simonmar [Mon, 9 Aug 2004 12:35:02 +0000 (12:35 +0000)]
[project @ 2004-08-09 12:35:02 by simonmar]
Fix scoped type variable example.

19 years ago[project @ 2004-08-09 12:15:09 by simonmar]
simonmar [Mon, 9 Aug 2004 12:15:09 +0000 (12:15 +0000)]
[project @ 2004-08-09 12:15:09 by simonmar]
Put a "-- " before the location info, since the rest of the output
looks like Haskell source.

19 years ago[project @ 2004-08-08 20:46:44 by panne]
panne [Sun, 8 Aug 2004 20:46:52 +0000 (20:46 +0000)]
[project @ 2004-08-08 20:46:44 by panne]
DocBook XML is case-sensitive. Note that this is still not correct DocBook, tons
of indexterm elements within varlistentries are still misplaced.

19 years ago[project @ 2004-08-08 17:26:26 by krasimir]
krasimir [Sun, 8 Aug 2004 17:26:47 +0000 (17:26 +0000)]
[project @ 2004-08-08 17:26:26 by krasimir]
xmlize sgml docbooks

19 years ago[project @ 2004-08-06 11:33:06 by simonmar]
simonmar [Fri, 6 Aug 2004 11:33:06 +0000 (11:33 +0000)]
[project @ 2004-08-06 11:33:06 by simonmar]
Update from Robert Ennals: allow either the ctags or etags file to be
generated, so as to be more friendly to case-insensitive filesystems.

19 years ago[project @ 2004-08-05 18:29:47 by krasimir]
krasimir [Thu, 5 Aug 2004 18:29:47 +0000 (18:29 +0000)]
[project @ 2004-08-05 18:29:47 by krasimir]
Bugfix. --auto-ghci-libs doesn't work if the package was previously installed.

19 years ago[project @ 2004-08-02 18:23:25 by panne]
panne [Mon, 2 Aug 2004 18:23:25 +0000 (18:23 +0000)]
[project @ 2004-08-02 18:23:25 by panne]
Fixed dvips commandline

MERGE TO STABLE

19 years ago[project @ 2004-07-30 08:40:11 by simonpj]
simonpj [Fri, 30 Jul 2004 08:40:11 +0000 (08:40 +0000)]
[project @ 2004-07-30 08:40:11 by simonpj]
Add case for foreign types

19 years ago[project @ 2004-07-30 08:39:51 by simonpj]
simonpj [Fri, 30 Jul 2004 08:39:51 +0000 (08:39 +0000)]
[project @ 2004-07-30 08:39:51 by simonpj]
Fix foreign-type predicate

19 years ago[project @ 2004-07-28 12:59:53 by simonpj]
simonpj [Wed, 28 Jul 2004 12:59:53 +0000 (12:59 +0000)]
[project @ 2004-07-28 12:59:53 by simonpj]
Comments only

19 years ago[project @ 2004-07-28 12:59:14 by simonpj]
simonpj [Wed, 28 Jul 2004 12:59:14 +0000 (12:59 +0000)]
[project @ 2004-07-28 12:59:14 by simonpj]
Fix GHCi buglet when browsing module

19 years ago[project @ 2004-07-23 15:31:58 by ross]
ross [Fri, 23 Jul 2004 15:32:00 +0000 (15:32 +0000)]
[project @ 2004-07-23 15:31:58 by ross]
To make Text.Regex.Posix work with Hugs, move cbits/regex/regex.h under
include, where Hugs can find it.

19 years ago[project @ 2004-07-21 10:47:28 by simonmar]
simonmar [Wed, 21 Jul 2004 10:47:29 +0000 (10:47 +0000)]
[project @ 2004-07-21 10:47:28 by simonmar]
Make total_allocated be an ullong, to accommodate programs that do a lot
of allocation.

MERGE TO STABLE

19 years ago[project @ 2004-07-21 10:44:49 by simonpj]
simonpj [Wed, 21 Jul 2004 10:44:49 +0000 (10:44 +0000)]
[project @ 2004-07-21 10:44:49 by simonpj]
Wibble to :i for (,); make algTyConRhs behave right

19 years ago[project @ 2004-07-21 10:07:29 by simonpj]
simonpj [Wed, 21 Jul 2004 10:07:39 +0000 (10:07 +0000)]
[project @ 2004-07-21 10:07:29 by simonpj]
Add location information to :i command

19 years ago[project @ 2004-07-21 09:25:42 by simonpj]
simonpj [Wed, 21 Jul 2004 09:25:45 +0000 (09:25 +0000)]
[project @ 2004-07-21 09:25:42 by simonpj]
-------------------------------
       Sort out the :i command for GHCi
-------------------------------

The :info command has been broken in the HEAD for some time, since the new IfaceSyn
story.  This commit sorts it out, and makes it nicer than before. For example, when
you :i a record selector, you get a cut-down data type declaration, so you can see
the context.

19 years ago[project @ 2004-07-19 11:51:16 by simonmar]
simonmar [Mon, 19 Jul 2004 11:51:16 +0000 (11:51 +0000)]
[project @ 2004-07-19 11:51:16 by simonmar]
do the file writes one after another, so if the underlying file system
happens to be case-insensitive we won't run into a file locking issue.

MERGE TO STABLE

19 years ago[project @ 2004-07-19 11:29:34 by simonpj]
simonpj [Mon, 19 Jul 2004 11:29:39 +0000 (11:29 +0000)]
[project @ 2004-07-19 11:29:34 by simonpj]
Template Haskell improvements

a) Make '() and '[] work.
b) Add tupleTypeName, tupleDataName
b) Try to improve error message for (lack of) existential data constructors in TH

19 years ago[project @ 2004-07-19 11:26:13 by simonpj]
simonpj [Mon, 19 Jul 2004 11:27:36 +0000 (11:27 +0000)]
[project @ 2004-07-19 11:26:13 by simonpj]
Comments only

19 years ago[project @ 2004-07-15 20:50:40 by sof]
sof [Thu, 15 Jul 2004 20:50:40 +0000 (20:50 +0000)]
[project @ 2004-07-15 20:50:40 by sof]
oops, back out rev. 1.31

19 years ago[project @ 2004-07-15 20:42:45 by sof]
sof [Thu, 15 Jul 2004 20:42:45 +0000 (20:42 +0000)]
[project @ 2004-07-15 20:42:45 by sof]
awaitEvent(): for Integrity, use socketpair() instead of pipe() to implement socket()-aborting mechanism

19 years ago[project @ 2004-07-12 08:47:19 by simonmar]
simonmar [Mon, 12 Jul 2004 08:47:19 +0000 (08:47 +0000)]
[project @ 2004-07-12 08:47:19 by simonmar]
Add a (rather vague) FAQ entry about the readline build problem
reported in bug #988585.

19 years ago[project @ 2004-07-09 21:13:35 by sof]
sof [Fri, 9 Jul 2004 21:13:35 +0000 (21:13 +0000)]
[project @ 2004-07-09 21:13:35 by sof]
back out rev 1.61; to hardwire path to gcc, use --with-gcc=<foo>

19 years ago[project @ 2004-07-09 16:47:59 by sof]
sof [Fri, 9 Jul 2004 16:47:59 +0000 (16:47 +0000)]
[project @ 2004-07-09 16:47:59 by sof]
tidy up gcc warnings re: %-related format specifiers, mainly rewriting \%%\% to %%%%

19 years ago[project @ 2004-06-29 17:14:01 by panne]
panne [Tue, 29 Jun 2004 17:14:01 +0000 (17:14 +0000)]
[project @ 2004-06-29 17:14:01 by panne]
Workaround for hbc: It compiles matching against 0-ary constructors in
list comprehensions into wrong code, e.g.

   [ () | True <- [False] ]

results in something like

   Error: No match in I4680Pinteractive

when issued in hbi.

19 years ago[project @ 2004-06-28 16:35:08 by sof]
sof [Mon, 28 Jun 2004 16:35:08 +0000 (16:35 +0000)]
[project @ 2004-06-28 16:35:08 by sof]
sec-makefile-dependencies: -o option is no longer supported.

Merge to STABLE.

19 years ago[project @ 2004-06-27 20:41:07 by panne]
panne [Sun, 27 Jun 2004 20:41:07 +0000 (20:41 +0000)]
[project @ 2004-06-27 20:41:07 by panne]
Small syntactic changes to please hbc's interpretation of the layout rule and records