ghc-hetmet.git
16 years agounbreak "recompile utils with stage 1"
Simon Marlow [Fri, 29 Jun 2007 09:07:05 +0000 (09:07 +0000)]
unbreak "recompile utils with stage 1"

16 years agoChecking that type indexes contain no synonym family applications
Manuel M T Chakravarty [Fri, 29 Jun 2007 06:36:56 +0000 (06:36 +0000)]
Checking that type indexes contain no synonym family applications

16 years agoRHS of a type instance must be a tau type
Manuel M T Chakravarty [Fri, 29 Jun 2007 04:59:31 +0000 (04:59 +0000)]
RHS of a type instance must be a tau type

16 years agoOverlap check for type families
Manuel M T Chakravarty [Fri, 29 Jun 2007 04:20:07 +0000 (04:20 +0000)]
Overlap check for type families
- If two "type instance"s overlap, they right-hand sides must be syntactically
  equal under the overlap substitution.  (Ie, we admit limited overlap, but
  require the system to still be confluent.)

16 years agoFurther compileToCore improvements
Tim Chevalier [Fri, 29 Jun 2007 01:48:31 +0000 (01:48 +0000)]
Further compileToCore improvements

Per suggestions from Simon M:

* Changed GHC.checkModule so that it doesn't call depanal.
* Changed GHC.checkModule to optionally return Core bindings
as a component of the CheckedModule that it returns (and
resulting changes to HscMain.hscFileCheck).
* As a result, simplified GHC.compileToCore and changed it
to load the given file so that the caller doesn't have to.

16 years agoFixed a documentation error
Michael D. Adams [Fri, 8 Jun 2007 09:11:57 +0000 (09:11 +0000)]
Fixed a documentation error

16 years agodisable .type directives on Windows; they confuse mingw's assembler
Simon Marlow [Thu, 28 Jun 2007 10:45:16 +0000 (10:45 +0000)]
disable .type directives on Windows; they confuse mingw's assembler

16 years agoSet .type @object for all global symbols in NCG
Clemens Fruhwirth [Wed, 27 Jun 2007 16:51:50 +0000 (16:51 +0000)]
Set .type @object for all global symbols in NCG

When linking against a dynamic library, the linker will emit a warning
if no type information is present within the library. We generate the
most trivial type for all externally visible labels, namely @object.

16 years agoAlso make proper references when not using fPIC but linking against dynamic libs
Clemens Fruhwirth [Tue, 19 Jun 2007 14:29:31 +0000 (14:29 +0000)]
Also make proper references when not using fPIC but linking against dynamic libs

16 years agoGenerate RelocatableReadOnlyData as .data
Clemens Fruhwirth [Wed, 27 Jun 2007 16:06:40 +0000 (16:06 +0000)]
Generate RelocatableReadOnlyData as .data

The GNU linker expects read-only sections to be relocation
free. Presumably because the dynamic linker maps all read-only
sections as read-only mmaps and hence can't do relocations.

If we want text-relocation-free shared libraries we have to put all
relocations into writable sections, hence .data.  See
http://www.mail-archive.com/cvs-all@haskell.org/msg15119.html

16 years agoGeneralize linkDynLib for ELF platforms
Clemens Fruhwirth [Wed, 27 Jun 2007 09:08:19 +0000 (09:08 +0000)]
Generalize linkDynLib for ELF platforms

To create a dynamic shared object (DSO) on ELF platforms, we invoke
"gcc -shared". This in turn invokes ld. We supply -Bsymbolic to the
linker, as "ld -Bsymbolic -shared -o <lib> <objs>" resolves all
references from <objs> to <objs> at library creation time.

See http://hackage.haskell.org/trac/ghc/wiki/Commentary/PositionIndependentCode

16 years agoRenames functions and constructors to fit their new ability for dynamic linking
Clemens Fruhwirth [Wed, 27 Jun 2007 09:01:00 +0000 (09:01 +0000)]
Renames functions and constructors to fit their new ability for dynamic linking
function: staticLink -> linkBinary (will link non static binaries too)
function: doMkDLL -> linkDynLib (as we can link on ELF too where DLL is an inappropriate term)
constructor: MkDLL -> LinkDynLib

16 years agoPprMach.hs more accurate hack for x86-64, CmmLabelOff is not a relative reference.
Clemens Fruhwirth [Tue, 26 Jun 2007 13:46:36 +0000 (13:46 +0000)]
PprMach.hs more accurate hack for x86-64, CmmLabelOff is not a relative reference.

CmmLabelDiffOff are generated in .text and if printed as .quad, causes
the assembler to emit a PC64 relocation. binutils prior to 2.17 don't
understand PC64 relocation properly.

pprDataItem is also used for printing SRT description tables. They are
part of the .data section and there no PC relative relocations are
emited. Hence, if we print a .long here, we get a absolute 32-bit
relocation. 32-bit relocations are problematic in dynamic libraries,
because dynamic library load addresses are loaded beyond the 32 bit
boundary, causing the dynamic linker to warn at dynamic linking
(loading the executable) that there are overflows in the
relocation. The executable still seems to work because of the small
memory model, but this is obviously wrong. Hence, remove CmmLabelOff
from the classification, causing these references to be printed as
.quad, causing correct 64-bit relocation as in the rest of the .data
section.

So, this hack now prints PC32 relocations in .text (mostly in the info
tables), and absolute 64-bit relocations in .data.

16 years agoFix typechecking bug with implicit parameters (Trac #1445)
simonmar@microsoft.com [Thu, 28 Jun 2007 08:11:05 +0000 (08:11 +0000)]
Fix typechecking bug with implicit parameters (Trac #1445)

tcSplitFunTy_maybe wasn't dealing with types like
(?x::Bool) => Int
Here, tcSplitFunTy_maybe should fail (as it would if there
was a for-all), because the type is a quantified type.

See Trac #1445, and test tc230.

16 years agoWhite space only
simonmar@microsoft.com [Thu, 28 Jun 2007 08:02:03 +0000 (08:02 +0000)]
White space only

16 years agoImprove error message when there is instance overlap
simonpj@microsoft.com [Thu, 28 Jun 2007 07:36:41 +0000 (07:36 +0000)]
Improve error message when there is instance overlap

16 years agoModify compileToCore to take just a filename
Tim Chevalier [Wed, 27 Jun 2007 22:48:09 +0000 (22:48 +0000)]
Modify compileToCore to take just a filename

Modified compileToCore to take just a session and a
filename, rather than a module name as well, since the module
name can be computed from the filename.

16 years agomaking hpc live
andy@galois.com [Wed, 27 Jun 2007 18:15:13 +0000 (18:15 +0000)]
making hpc live

16 years ago+RTS -xbXXXXX sets the "heap base" to 0xXXXXXX
Simon Marlow [Wed, 27 Jun 2007 09:36:46 +0000 (09:36 +0000)]
+RTS -xbXXXXX sets the "heap base" to 0xXXXXXX
When debugging the GC and storage manager we often want repeated runs
of the program to allocate memory at the same addresses, so that we
can set watch points.  Unfortunately the OS doesn't always give us
memory at predictable addresses.  This flag gives the OS a hint as to
where we would like our memory allocated.  Previously I did this by
changing the HEAP_BASE setting in MBlock.h and recompiling, this patch
just adds a flag so I don't have to recompile.

16 years ago+RTS -V0 disables the interval timer completely (for repeatable debugging)
Simon Marlow [Wed, 27 Jun 2007 09:29:41 +0000 (09:29 +0000)]
+RTS -V0 disables the interval timer completely (for repeatable debugging)

16 years agoFIX BUILD (on Windows): follow changes to make threaded RTS compile with -fasm
simonmar@microsoft.com [Wed, 27 Jun 2007 10:16:04 +0000 (10:16 +0000)]
FIX BUILD (on Windows): follow changes to make threaded RTS compile with -fasm

16 years agoFix names of coercions in newtype instances
Manuel M T Chakravarty [Wed, 27 Jun 2007 09:42:00 +0000 (09:42 +0000)]
Fix names of coercions in newtype instances
- Thanks to Roman for spotting the problem.

16 years agoCleaning up Hpc.c; adding support for reflection into Hpc.
andy@galois.com [Wed, 27 Jun 2007 06:36:31 +0000 (06:36 +0000)]
Cleaning up Hpc.c; adding support for reflection into Hpc.

16 years agoFixing Makefile hpc to use compat
andy@galois.com [Wed, 27 Jun 2007 06:07:21 +0000 (06:07 +0000)]
Fixing Makefile hpc to use compat

16 years agoAdding Tix to compat library, so that hpc in utils can use it
andy@galois.com [Wed, 27 Jun 2007 06:04:30 +0000 (06:04 +0000)]
Adding Tix to compat library, so that hpc in utils can use it

16 years agofixing creation of directory for html output; fixing html markup for 0% bars.
andy@galois.com [Wed, 27 Jun 2007 05:48:46 +0000 (05:48 +0000)]
fixing creation of directory for html output; fixing html markup for 0% bars.

16 years agoFixing -fhpc flag to allow -fno-hpc
andy@galois.com [Wed, 27 Jun 2007 05:38:08 +0000 (05:38 +0000)]
Fixing -fhpc flag to allow -fno-hpc

16 years agoFixed deriving of associated data types
Manuel M T Chakravarty [Wed, 27 Jun 2007 05:48:34 +0000 (05:48 +0000)]
Fixed deriving of associated data types
- We forgot to pull the data declarations nested in class instances out of
  the instances when collecting all the predicates that we need derive.
  Thanks to Roman for spotting this.

16 years agoadd missing case for compiling 64-bit operations on x86
Simon Marlow [Tue, 26 Jun 2007 21:11:53 +0000 (21:11 +0000)]
add missing case for compiling 64-bit operations on x86

16 years agoMake the threaded RTS compilable using -fasm
Simon Marlow [Tue, 26 Jun 2007 21:10:58 +0000 (21:10 +0000)]
Make the threaded RTS compilable using -fasm
We needed to turn some inline C functions and C macros into either
real C functions or C-- macros.

16 years agox86_64: fix a few bugs in the >8 floating point args case
Simon Marlow [Tue, 26 Jun 2007 10:30:55 +0000 (10:30 +0000)]
x86_64: fix a few bugs in the >8 floating point args case

16 years agomake inplace scripts less sensitive to /bin/sh quoting by avoiding \
Simon Marlow [Sat, 23 Jun 2007 20:00:06 +0000 (20:00 +0000)]
make inplace scripts less sensitive to /bin/sh quoting by avoiding \

16 years agoexcluding Trace.Hpc.* when using the compat library
andy@galois.com [Tue, 26 Jun 2007 07:22:19 +0000 (07:22 +0000)]
excluding Trace.Hpc.* when using the compat library

16 years agoMaking -fhpc work with a stage1 build, via the compat 'package'.
andy@galois.com [Tue, 26 Jun 2007 05:34:29 +0000 (05:34 +0000)]
Making -fhpc work with a stage1 build, via the compat 'package'.

16 years agoAdding hpc lib as part of the compat 'package'
andy@galois.com [Tue, 26 Jun 2007 05:11:36 +0000 (05:11 +0000)]
Adding hpc lib as part of the compat 'package'

16 years agoCleanup Hpc sub-system, remove hpc-tracer implementation.
andy@galois.com [Tue, 26 Jun 2007 04:19:58 +0000 (04:19 +0000)]
Cleanup Hpc sub-system, remove hpc-tracer implementation.

16 years agoAdd a compileToCore function to the GHC API
Tim Chevalier [Mon, 25 Jun 2007 22:06:08 +0000 (22:06 +0000)]
Add a compileToCore function to the GHC API

Added a compileToCore function to the GHC API that takes a
  session, module, and filename, and returns a list of Core
  bindings if successful. This is just a first try and could
  probably be improved (for example, there's probably a way to
  get the filename from the module so that it doesn't have to
  be passed in, I just don't see it offhand.)

16 years agoDefine SUBDIRS in Makefile (needed for clean; fixes trac #1440)
Ian Lynagh [Mon, 25 Jun 2007 17:49:52 +0000 (17:49 +0000)]
Define SUBDIRS in Makefile (needed for clean; fixes trac #1440)

16 years agoTweak the configuration and installation slightly
Ian Lynagh [Mon, 25 Jun 2007 17:40:11 +0000 (17:40 +0000)]
Tweak the configuration and installation slightly
Make it so that the documentation ends up where Cabal expects it to be
on Windows (prep-bin-dist-mingw used to move it).

16 years agowithExtendedLinkerState: don't revert the whole state
Simon Marlow [Mon, 25 Jun 2007 15:14:55 +0000 (15:14 +0000)]
withExtendedLinkerState: don't revert the whole state
Fixes test failures print017 and print024

16 years agoPrint infix type constructors in an infix way
simonpj@microsoft.com [Mon, 25 Jun 2007 15:28:58 +0000 (15:28 +0000)]
Print infix type constructors in an infix way

Fixes Trac #1425.  The printer for types doesn't know about fixities.
(It could be educated to know, but it doesn't at the moment.)  So it
treats all infix tycons as of precedence less than application and function
arrrow.

I took a slight shortcut and reused function-arrow prededence, so I think
you may get
T -> T :% T
meaning
T -> (T :% T)

If that becomes a problem we can fix it.

16 years agoClamp -O flag to beween 0 and 2
simonpj@microsoft.com [Mon, 25 Jun 2007 15:26:16 +0000 (15:26 +0000)]
Clamp -O flag to beween 0 and 2

Fixes Trac #1272

16 years agoAdding hpc tools, as a single program.
andy@galois.com [Mon, 25 Jun 2007 07:09:43 +0000 (07:09 +0000)]
Adding hpc tools, as a single program.

16 years agoUpdate version numbering policy in the users guide
Ian Lynagh [Sat, 23 Jun 2007 22:44:40 +0000 (22:44 +0000)]
Update version numbering policy in the users guide

16 years agoFix an error message
Ian Lynagh [Sat, 23 Jun 2007 19:06:53 +0000 (19:06 +0000)]
Fix an error message
`y' in the error message
    `x' is not a (visible) method of class `y'
had gone missing.

16 years agoturning off -fhpc in stage1 built ghc
andy@galois.com [Fri, 22 Jun 2007 22:52:26 +0000 (22:52 +0000)]
turning off -fhpc in stage1 built ghc

16 years agoremoving -fhpc-tracer from ghc, is subsumed by the GHC debugger
andy@galois.com [Fri, 22 Jun 2007 22:42:26 +0000 (22:42 +0000)]
removing -fhpc-tracer from ghc, is subsumed by the GHC debugger

16 years agoStage2 now used the package hpc to get the hpc datastructures
andy@galois.com [Fri, 22 Jun 2007 22:40:38 +0000 (22:40 +0000)]
Stage2 now used the package hpc to get the hpc datastructures
Stage1 no longer supports hpc (-fhpc is ignored)

16 years agoFix typo in Makefile
Ian Lynagh [Fri, 22 Jun 2007 22:21:33 +0000 (22:21 +0000)]
Fix typo in Makefile

16 years agoAdding hpc package to ghc core libraries
andy@galois.com [Fri, 22 Jun 2007 17:18:51 +0000 (17:18 +0000)]
Adding hpc package to ghc core libraries

16 years agoChange how the libraries Makefile adds --configure-option= flags; fixes #1431
Ian Lynagh [Fri, 22 Jun 2007 16:09:51 +0000 (16:09 +0000)]
Change how the libraries Makefile adds --configure-option= flags; fixes #1431
We now assume that each configure option is quoted with '', and thus
replace " '" with " --configure-option='".

16 years agoupdate with new libraries
Simon Marlow [Fri, 22 Jun 2007 13:12:11 +0000 (13:12 +0000)]
update with new libraries

16 years agoFIX BUILD on Windows: horrible hack to work around make(?) bug
Simon Marlow [Thu, 21 Jun 2007 14:45:19 +0000 (14:45 +0000)]
FIX BUILD on Windows: horrible hack to work around make(?) bug

16 years agoFIX BUILD (on Windows): Cabal must invoke compiler/stage1/ghc-inplace
Simon Marlow [Fri, 22 Jun 2007 11:16:34 +0000 (11:16 +0000)]
FIX BUILD (on Windows): Cabal must invoke compiler/stage1/ghc-inplace
not compiler/ghc-inplace, which is a symlink.

16 years agoremove unnecessary cruft
Simon Marlow [Fri, 22 Jun 2007 07:56:02 +0000 (07:56 +0000)]
remove unnecessary cruft

16 years agoFIX BUILD: all builds need --template, not just Windows
Simon Marlow [Fri, 22 Jun 2007 07:54:44 +0000 (07:54 +0000)]
FIX BUILD: all builds need --template, not just Windows

16 years agoignore all but the last --template option
Simon Marlow [Fri, 22 Jun 2007 07:54:17 +0000 (07:54 +0000)]
ignore all but the last --template option

16 years agoUnbreak the stage-2 compiler (record-type changes)
simonpj@microsoft.com [Thu, 21 Jun 2007 17:01:31 +0000 (17:01 +0000)]
Unbreak the stage-2 compiler (record-type changes)

16 years ago-fglasgow-exts implies -X=GADTs
simonpj@microsoft.com [Thu, 21 Jun 2007 16:21:52 +0000 (16:21 +0000)]
-fglasgow-exts implies -X=GADTs

16 years agoFIX read040: patterns with type sig on LHS of do-binding
simonpj@microsoft.com [Thu, 21 Jun 2007 14:37:21 +0000 (14:37 +0000)]
FIX read040: patterns with type sig on LHS of do-binding

f () = do { x :: Bool <- return True; ... }

For some reason the production for 'pat' required 'infixexp' on the
LHS of a do-notation binding.  This patch makes it an 'exp', which
thereby allows an expression with a type sig.

Happily, there are no new shift-reduce errors, so I don't think this
will break anything else.

16 years agoAnother wibble to the head-exploded error message (suggested by David Roundy)
simonpj@microsoft.com [Thu, 21 Jun 2007 14:11:44 +0000 (14:11 +0000)]
Another wibble to the head-exploded error message (suggested by David Roundy)

16 years agoUse the correct flag for controlling scoped type variables in an instance decl
simonpj@microsoft.com [Thu, 21 Jun 2007 14:11:14 +0000 (14:11 +0000)]
Use the correct flag for controlling scoped type variables in an instance decl

16 years agoImprove 'my head exploded' error message
simonpj@microsoft.com [Thu, 21 Jun 2007 13:16:50 +0000 (13:16 +0000)]
Improve 'my head exploded' error message

16 years agoFIX BUILD: add missing prime!
simonpj@microsoft.com [Thu, 21 Jun 2007 13:16:25 +0000 (13:16 +0000)]
FIX BUILD: add missing prime!

16 years agoFIX BUILD on Windows: horrible hack to work around make(?) bug
Simon Marlow [Thu, 21 Jun 2007 14:45:19 +0000 (14:45 +0000)]
FIX BUILD on Windows: horrible hack to work around make(?) bug

16 years agoyet more fixes: Cygwin broke this time
Simon Marlow [Thu, 21 Jun 2007 14:06:53 +0000 (14:06 +0000)]
yet more fixes: Cygwin broke this time

16 years agofix bugs with hsc2hs-inplace
Lemmih [Thu, 21 Jun 2007 13:28:25 +0000 (13:28 +0000)]
fix bugs with hsc2hs-inplace

16 years agomk/build.mk is optional
Lemmih [Thu, 21 Jun 2007 13:07:27 +0000 (13:07 +0000)]
mk/build.mk is optional

16 years agofurther fixes to the inplace scripts
Lemmih [Thu, 21 Jun 2007 13:06:57 +0000 (13:06 +0000)]
further fixes to the inplace scripts

16 years agouse a binary for hsc2hs-inplace too
Simon Marlow [Thu, 21 Jun 2007 12:14:26 +0000 (12:14 +0000)]
use a binary for hsc2hs-inplace too

16 years agoFix problems with new inplace stuff on Cygwin
Simon Marlow [Thu, 21 Jun 2007 11:41:47 +0000 (11:41 +0000)]
Fix problems with new inplace stuff on Cygwin

16 years agoFIX BUILD: can't build the makefiles during make boot
Simon Marlow [Thu, 21 Jun 2007 09:51:54 +0000 (09:51 +0000)]
FIX BUILD: can't build the makefiles during make boot

16 years agoclean up lib/{GNUmakefile,Makefile.local}
Simon Marlow [Wed, 20 Jun 2007 13:28:09 +0000 (13:28 +0000)]
clean up lib/{GNUmakefile,Makefile.local}

16 years agodefault_target should be "all", not "build"
Simon Marlow [Wed, 20 Jun 2007 13:20:01 +0000 (13:20 +0000)]
default_target should be "all", not "build"

16 years agoadd comment
Simon Marlow [Wed, 20 Jun 2007 13:19:49 +0000 (13:19 +0000)]
add comment

16 years agoFix a problem with package.mk being included too early
Simon Marlow [Wed, 20 Jun 2007 12:22:24 +0000 (12:22 +0000)]
Fix a problem with package.mk being included too early

16 years agoUse setup makefile + make by default to build libraries
Simon Marlow [Wed, 20 Jun 2007 12:20:09 +0000 (12:20 +0000)]
Use setup makefile + make by default to build libraries

The advantages of this are
  (a) it's a step closer to getting -j working again (make -j works in
      an individual library, but not in libraries/ yet).
  (b) it's easier to hack on libraries: make dist/build/Foo.o
  (c) it's a step closer to getting HC bootstrapping again

The build system creates <lib>/GNUmakefile as part of 'make boot'.
This was chosen so as not to interfere with existing Makefiles, but
it's a bit of a hack.  (previously I used CabalMakefile, but that
means adding -f CabalMakefile each time you run make, and that's a
pain).

16 years agoUse a real binary instead of scripts for ghc-inplace
Simon Marlow [Thu, 21 Jun 2007 10:13:24 +0000 (10:13 +0000)]
Use a real binary instead of scripts for ghc-inplace
Fixes various problems with getting the scripts right on Windows.
Binaries are universally executable by /bin/sh, cmd.exe and rawSystem,
so this allows us to remove some platform-specific hacks.

16 years agoAdd several new record features
Lemmih [Thu, 21 Jun 2007 09:15:52 +0000 (09:15 +0000)]
Add several new record features

1. Record disambiguation (-fdisambiguate-record-fields)

In record construction and pattern matching (although not
in record updates) it is clear which field name is intended
even if there are several in scope.  This extension uses
the constructor to disambiguate.  Thus
C { x=3 }
uses the 'x' field from constructor C (assuming there is one)
even if there are many x's in scope.

2. Record punning (-frecord-puns)

In a record construction or pattern match or update you can
omit the "=" part, thus
C { x, y }
This is just syntactic sugar for
C { x=x, y=y }

3.  Dot-dot notation for records (-frecord-dot-dot)

In record construction or pattern match (but not update)
you can use ".." to mean "all the remaining fields".  So

C { x=v, .. }

means to fill in the remaining fields to give

C { x=v, y=y }

(assuming C has fields x and y).  This might reasonably
considered very dodgy stuff.  For pattern-matching it brings
into scope a bunch of things that are not explictly mentioned;
and in record construction it just picks whatver 'y' is in
scope for the 'y' field.   Still, Lennart Augustsson really
wants it, and it's a feature that is extremely easy to explain.

Implementation
~~~~~~~~~~~~~~
I thought of using the "parent" field in the GlobalRdrEnv, but
that's really used for import/export and just isn't right for this.
For example, for import/export a field is a subordinate of the *type
constructor* whereas here we need to know what fields belong to a
particular *data* constructor.

The main thing is that we need to map a data constructor to its
fields, and we need to do so in the renamer.   For imported modules
it's easy: just look in the imported TypeEnv.  For the module being
compiled, we make a new field tcg_field_env in the TcGblEnv.
The important functions are
RnEnv.lookupRecordBndr
RnEnv.lookupConstructorFields

There is still a significant infelicity in the way the renamer
works on patterns, which I'll tackle next.

I also did quite a bit of refactoring in the representation of
record fields (mainly in HsPat).***END OF DESCRIPTION***

Place the long patch description above the ***END OF DESCRIPTION*** marker.
The first line of this file will be the patch name.

This patch contains the following changes:

M ./compiler/deSugar/Check.lhs -3 +5
M ./compiler/deSugar/Coverage.lhs -6 +7
M ./compiler/deSugar/DsExpr.lhs -6 +13
M ./compiler/deSugar/DsMeta.hs -8 +8
M ./compiler/deSugar/DsUtils.lhs -1 +1
M ./compiler/deSugar/MatchCon.lhs -2 +2
M ./compiler/hsSyn/Convert.lhs -3 +3
M ./compiler/hsSyn/HsDecls.lhs -9 +25
M ./compiler/hsSyn/HsExpr.lhs -13 +3
M ./compiler/hsSyn/HsPat.lhs -25 +63
M ./compiler/hsSyn/HsUtils.lhs -3 +3
M ./compiler/main/DynFlags.hs +6
M ./compiler/parser/Parser.y.pp -13 +17
M ./compiler/parser/RdrHsSyn.lhs -16 +18
M ./compiler/rename/RnBinds.lhs -2 +2
M ./compiler/rename/RnEnv.lhs -22 +82
M ./compiler/rename/RnExpr.lhs -34 +12
M ./compiler/rename/RnHsSyn.lhs -3 +2
M ./compiler/rename/RnSource.lhs -50 +78
M ./compiler/rename/RnTypes.lhs -50 +84
M ./compiler/typecheck/TcExpr.lhs -18 +18
M ./compiler/typecheck/TcHsSyn.lhs -20 +21
M ./compiler/typecheck/TcPat.lhs -8 +6
M ./compiler/typecheck/TcRnMonad.lhs -6 +15
M ./compiler/typecheck/TcRnTypes.lhs -2 +11
M ./compiler/typecheck/TcTyClsDecls.lhs -3 +4
M ./docs/users_guide/flags.xml +7
M ./docs/users_guide/glasgow_exts.xml +42

16 years agoRemove the unused HsExpr constructor DictPat
Lemmih [Mon, 18 Jun 2007 12:46:05 +0000 (12:46 +0000)]
Remove the unused HsExpr constructor DictPat

16 years agoFix a bug in MatchCon, and clarify what dataConInstOrigArgTys does
Lemmih [Thu, 7 Jun 2007 21:38:37 +0000 (21:38 +0000)]
Fix a bug in MatchCon, and clarify what dataConInstOrigArgTys does

There was an outright bug in MatchCon.matchOneCon, in the construction
of arg_tys.  Easily fixed.  It never showed up becuase the arg_tys are
only used in WildPats, and they in turn seldom have their types looked
(except by hsPatType).  So I can't make a test case for htis.

While I was investigating, I added a bit of clarifation and
invariant-checking to dataConInstOrigArgTys and dataConInstArgTys

16 years agoDo not perform a worker/wrapper split for a NOINLINE function
Lemmih [Thu, 7 Jun 2007 21:35:23 +0000 (21:35 +0000)]
Do not perform a worker/wrapper split for a NOINLINE function

This came up in an email exchange with Duncan Coutts in May 2007.
If a function is marked NOINLINE there is really no point in
doing a worker/wrapper split, because the wrapper will never
be inlined.

16 years agoFix Trac #1402: typo in specialiser
Lemmih [Thu, 7 Jun 2007 18:55:34 +0000 (18:55 +0000)]
Fix Trac #1402: typo in specialiser

This patch fixes a plain bug in the specialiser (rhs_bndrs instead
of rhs_ids) which made GHC crash in obscure cases.

It exposed a case in which we might not do all possible specialisation;
see Note [Specialisation shape].  It's not an important case, but I've
added a warning in DEBUG mode.

Trac #1402.  Test is spec003.hs

16 years agoWibble: make -fno-implicit-prelude work
simonpj@microsoft.com [Thu, 21 Jun 2007 07:24:56 +0000 (07:24 +0000)]
Wibble: make -fno-implicit-prelude work

16 years agoRemove an incorrect claim that [t| ... |] isn't implemented yet
Ian Lynagh [Thu, 21 Jun 2007 00:31:03 +0000 (00:31 +0000)]
Remove an incorrect claim that [t| ... |] isn't implemented yet

16 years agoMake building haddock docs opt-in rather than opt-out
Ian Lynagh [Wed, 20 Jun 2007 23:59:09 +0000 (23:59 +0000)]
Make building haddock docs opt-in rather than opt-out

16 years agoDon't assume that the main repo is called "ghc" in darcs-all
Ian Lynagh [Wed, 20 Jun 2007 23:40:07 +0000 (23:40 +0000)]
Don't assume that the main repo is called "ghc" in darcs-all
Fixes working in branches where that isn't true.

16 years agoTrivial fix to clear Trac #1386
simonpj@microsoft.com [Wed, 20 Jun 2007 16:58:36 +0000 (16:58 +0000)]
Trivial fix to clear Trac #1386

16 years agoWibbles in flaggery, concerning backward compatibility with -f flags
simonpj@microsoft.com [Wed, 20 Jun 2007 16:58:01 +0000 (16:58 +0000)]
Wibbles in flaggery, concerning backward compatibility with -f flags

16 years agoImplement -X=GADTs and -X=RelaxedPolyRec
simonpj@microsoft.com [Wed, 20 Jun 2007 16:33:59 +0000 (16:33 +0000)]
Implement -X=GADTs and -X=RelaxedPolyRec

Two new -X flags, one for GADTs and one for relaxed polymorphic recursion

This also fixes a rather confusing error message that the Darcs folk
tripped over.

16 years agoUse -X for language extensions
simonpj@microsoft.com [Wed, 20 Jun 2007 16:26:56 +0000 (16:26 +0000)]
Use -X for language extensions

We've often talked about having a separate flag for language extensions,
and now we have one. You can say

-XImplicitParams
-X=ImplicitParams
-Ximplicit-params

as you like.  These replace the "-f" flags with similar names (though
the -f prefix will serve as a synonym for -X for a while).

There's an optional "=", and the flag is normalised by removing hyphens
and lower-casing, so all the above variants mean the same thing.

The nomenclature is intended to match the LANGUAGE pramgas, which are
defined by Cabal.  So you can also say

{-# LANGUAGE ImplicitParams #-}

But Cabal doesn't have as many language options as GHC does, so the -X
things are a superset of the LANGUAGE things.

The optional "=" applies to all flags that take an argument, so you can,
for example, say

-pgmL=/etc/foo

I hope that's ok.  (It's an unforced change; just fitted in.)

I hope we'll add more -X flags, to replace the portmanteau -fglasgow-exts
which does everything!

I have updated the manual, but doubtless missed something.

16 years agoUnused import
simonpj@microsoft.com [Wed, 20 Jun 2007 16:13:41 +0000 (16:13 +0000)]
Unused import

16 years agoturning back on case liberation when using hpc
andy@galois.com [Wed, 20 Jun 2007 15:02:44 +0000 (15:02 +0000)]
turning back on case liberation when using hpc

16 years agoremove debugging code accidentally left in
Simon Marlow [Wed, 20 Jun 2007 12:57:42 +0000 (12:57 +0000)]
remove debugging code accidentally left in

16 years agoUse .NOTPARALLEL for the libraries Makefile
Ian Lynagh [Wed, 20 Jun 2007 12:07:23 +0000 (12:07 +0000)]
Use .NOTPARALLEL for the libraries Makefile

16 years agoImprove the handling of deriving, in error cases
simonpj@microsoft.com [Wed, 20 Jun 2007 10:28:28 +0000 (10:28 +0000)]
Improve the handling of deriving, in error cases

I'd been too ambitious with error handling for 'deriving', and got it
entirely wrong.  This fixes it.  See extensive
Note [Exotic derived instance contexts]
in TcSimplify.  (Most of the extra lines are comments!)

16 years agoMore refactoring in TcSimplify
simonpj@microsoft.com [Wed, 20 Jun 2007 09:36:46 +0000 (09:36 +0000)]
More refactoring in TcSimplify

This re-jig tides up the top-level simplification, and combines in one
well-commented function, approximateImplications, the rather ad-hoc
way of simplifying implication constraints during type inference.

Error messages get a bit better too.

16 years agoFix egregious sharing bug in LiberateCase
simonpj@microsoft.com [Wed, 20 Jun 2007 07:56:48 +0000 (07:56 +0000)]
Fix egregious sharing bug in LiberateCase

Andy Gill writes: consider the following code

       f = g (case v of
                V a b -> a : t f)

where g is expensive. Liberate case will turn this into

       f = g (case v of
               V a b -> a : t (letrec f = g (case v of
                                             V a b -> a : f t)
                                in f)
             )
Yikes! We evaluate g twice. This leads to a O(2^n) explosion
if g calls back to the same code recursively.

This may be the same as Trac #1366.

16 years agoTurning off case liberation when using the hpc option, for now
andy@galois.com [Wed, 20 Jun 2007 06:49:43 +0000 (06:49 +0000)]
Turning off case liberation when using the hpc option, for now

Consider the following code

     f = g (case v of
              V a b -> a : t f)

where g is expensive. Liberate case will turn this into

     f = g (case v of
             V a b -> a : t (letrec f = g (case v of
                                           V a b -> a : f t)
                              in f)
           )

Yikes! We evaluate g twice. This leads to a O(2^n) explosion
if g calls back to the same code recursively.

This happen sometimes in HPC, because every tick is a liberate-able case,
but is a general problem to case liberation (I think).

16 years agoAdd --core-only flag to push-all
Ian Lynagh [Tue, 19 Jun 2007 20:05:46 +0000 (20:05 +0000)]
Add --core-only flag to push-all