GitLab

Two SMP-related changes:

  - New storage manager interface:

    bdescr *allocateLocal(StgRegTable *reg, nat words)

    which allocates from the current thread's nursery (being careful
    not to clash with the heap pointer).  It can do this without
    taking any locks; the lock only has to be taken if a block needs
    to be allocated.  allocateLocal() is now used instead of allocate()
    in a few PrimOps.

    This removes locks from most Integer operations, cutting down
    the overhead for SMP a bit more.

    To make this work, we have to be able to grab the current thread's
    Capability out of thin air (i.e. when called from GMP), so the
    Capability subsystem needs to keep a hash from thread IDs to
    Capabilities.

  - Small MVar optimisation: instead of taking the global
    storage-manager lock, do our own locking of MVars with a bit of
    inline assembly (x86 only for now).

Mangler support for GNU/Hurd.

From: Thomas Schwinge

Add -I options to SRC_CC_OPTS for bootstrapping

From: Thomas Schwinge

Bootstrap fix from GNU/Hurd port

From: Thomas Schwinge

GNU/Hurd platform (from Thomas Schwinge
<schwinge-list-ghc-users@nic-nac-project.de>)

-ignore-package lang if GHC >= 6.3, fixes build breakage if wxHaskell
 is installed.

More details about type checking.

printAllThreads: handle ThreadRelocated

deleteAllThreads: handle ThreadRelocated

ProjectPatchLevel hack: the STABLE branch versions now look like
6.4.1.20050508 which confuses the compiler which thinks
ProjectPatchLevel should be an Int.

GCC 4.0 Fix: Do not emit lvalue casts for foreign calls.

Note: This fix might break cmm code that directly assigns the result
      of a foreign call to a "strange type" register (but we don't do that).

MERGE TO STABLE

Allow unicode in string and character literals when pretty printing
external core.

Rule overlap message should be only in ifdef DEBUG

[mingw only]
Work around bug in win32 Console API which showed up in the GHCi UI:
if the user typed in characters prior to the appearance of the prompt,
the first of these characters always came out as a 'g'. The GHCi UI does
for good reasons one-character reads from 'stdin', which causes the
underlying APIs to become confused. A simple repro case is the following
piece of C code:

/*----------------------*/
#include <stdio.h>
#include <windows.h>
int main()
{
    char ch1,ch2;
    HANDLE hStdIn = GetStdHandle(STD_INPUT_HANDLE);
    DWORD dw;

    /* Type in some characters before the prompt appears and be amused.. */
    sleep(1000); printf("? ");
    ReadConsoleA(hStdIn,&ch1,1,&dw,NULL);
    ReadConsoleA(hStdIn,&ch2,1,&dw,NULL);
/*  or, if you want to use libc:
    read(0,&ch1,1); read(0,&ch2,1); */

    printf("%c%c\n", ch1,ch2);
    return 0;
}
/*----------------------*/

This happens across win32 OSes, and I can't see anything untoward as far
as API usage goes (the GHC IO implementation uses read(), but that
reduces to ReadConsoleA() calls.) People inside the Behemoth might want
to have a closer look at this..

Not much we can do about this except work around the problem by flushing
the input buffer prior to reading from stdin. Not ideal, as type-ahead
is a useful feature. Flushing is handled by GHC.ConsoleHandler.flushConsole

Merge to STABLE.

ocResolve_PEi386():
    when fixing up REL32 relocations, _add_ displacement to value at the
    given offset. The existing value has so far been assumed to be zero
    (which we've asserted for), but curiously wxhaskell-0.9.4's wx.o
    contains lots of interesting non-zero values. Information / specifications
    are awfully thin on the ground as to how to precisely handle these
    relocations, but adding rather than overwriting seems to have a
    generally healthy effect; unable to crash wxhaskell-0.9.4 with a 6.4 build.

Merge to STABLE.

Some fixes to the blackhole garbage collection strategy. conc049 is a
great test case.

Report instances correctly in GHCi

Only compare hi-boot iface with mother module if there *is* an hi-boot iface

Be a bit more parsimonious about type-error contexts

Treat the blackhole_queue as a weak list: it shouldn't keep its
elements alive, because otherwise we can't detect deadlock using the
GC.  Fortunately the fix is quite simple.

AP_STACK fix for SMP

Fix the offsets and macros for AP_STACK closures (was wrong for SMP only)

Sanity fix for SMP

Stop the renamer from barfing on arrow notation outside of proc, since
this will be caught by the typechecker (tcfail120).

Make GHC.modInfoPrintUnqualified work for package modules too.  Also
refactor a bit: move mkExportEnv from TcRnDriver up to GHC which is
the only use of it.

Make it so that you can deprecate a data constructor.
Previously {-# DEPRECATED T "no" #-} referred only to the type
or class T.  Now it refers to the data constructor T as well,
just like in fixity declarations.  

There's no way to deprecate the data constructor T without also
deprecating the type T, alas.  Same problem in fixity decls.
Main problem is coming up with a suitable concrete syntax to do
so.

We could consider merging this to the STABLE branch.

NB: Sven, the manual fixes are not XML-valideated!  I'm at home.

Two small things

a) report duplicate declarations in canonical order
b) report deprecations for all uses (a longstanding bug)

	both MERGE TO STABLE

[mingw only]
Provide (unsatisfying) workaround for pesky Ctrl-C issue, whereby a
cygwin-based bash will terminate the 'ghci.exe' wrapper without the
wrapper process getting a chance to respond (and shut down its child,
if nothing else.) The result instead being that the child (ghc.exe) ends up
lingering in the background.

To steer users away from such badness, the wrapper checks to see whether
it is running in a *nix-like shell, and if it is, issue a warning that
it would be better to invoke GHCi using the 'ghcii.sh' shell script instead.

Merge to STABLE.

getModuleInfo now does something reasonable for package modules.

Fix instance-matching issue (affects nofib/real/prolog).

Add lookupGlobalName

Updated and extended the section about the renamer.

Make GHC accept external core files with higher rank types.

Fix porting instructions

Further wibbles to unused-import reporting; merge to stable

Another go at the handling of -< in arrow notation, continuing and
fixing the changes in

	http://www.haskell.org/pipermail/cvs-all/2005-April/040391.html

Now do the same thing in the renamer as we do in the type checker,
i.e. return to the environment of the proc when considering the left
argument of -<.

This is much simpler than the old proc_level stuff, and matches the
type rules more clearly.  But there is a change in error messages.
For the input

	f :: Int -> Int
	f = proc x -> (+x) -< 1

GHC 6.4 says

	test.hs:6:
	    Command-bound variable `x' is not in scope here
		Reason: it is used in the left argument of (-<)
	    In the second argument of `(+)', namely `x'
	    In the command: (+ x) -< 1
	    In the definition of `f': f = proc x -> (+ x) -< 1

but now we just get the blunt

	test.hs:6:16: Not in scope: `x'

The beauty is all on the inside.

Similarly leakage of existential type variables (arrow1) is detected,
but the error message isn't very helpful.

Check for illegal declarations in hs-boot files

Second stab at the duplicate-import warnings

Improve generation of 'duplicate import' warnings.
This involved changing (actually simplifying) the
definition of RdrName.ImportSpec.

I'm not sure whether this one merits merging or not.
Perhaps worth a try.

Fix the test for duplicate local bindings, so that it works with
Template Haskell.  Pre-TH, all the local bindings came into scope
at once, but with TH they come into scope in groups, and we must
check for conflict with existing local bindings.

	MERGE TO STABLE