Test-suite titivation
		      ---------------------

This commit achieves several glorious goals and adds many frivolous
features:

    * Makes the test driver work on Windows (use bash for system calls)
    * Adds -Di386_unknown_mingw32 to CPP_OPTS so that Win-specific code is
      compiled properly
    * Adds the ability to have platform-specific test results (by setting
      the new $platform variable to config.mk's TARGETPLATFORM)
    * Fixes several tests to work on Windows, mostly by adding platform-
      specific result files.
    * Pipes all stderr files through normalise_errmsg, which itself is
      improved to handle .exe at the end of filenames, and backslashes.
    * Allows stdout output to be piped through normalise_errmsg; useful in
      the rare cases where output includes filenames.
    * Comprehensively breaks the testsuite on all other platforms (with any
      luck)
    * Splundig vur thrig, earthlets!

Update expected result.

MERGE TO STABLE

Fix some signatures after Ord was removed as a superclass of Ix.

Fix some signatures after the change to the Ix class.

Fix expected output.

Don't buffer stdout when running tests, to keep log files tidy
when we redirect both stderr and stdout into the same log file
from runtest.

THIS CHANGE AFFECTS ALL OBJECT FILES COMPILED FROM HASKELL.
Please say "make -C ghc/lib/std clean; make -C hslibs clean".

This commit eliminates spurious warning messages when compiling on
the Alpha.  There are two kinds of spurious warning messages:

(1) gcc: -noprefix_recognition: linker input file unused since linking not done

    This warning is because we pass the flag "-Xlinker -noprefix_recognition"
    to gcc.  We remove this warning by no longer passing the flag to gcc,
    and by removing the reason we were passing the flag in the first place:
    __init_* is now renamed to __stginit_*.

(2) .../includes/Regs.h: warning: call-clobbered register used for global
    register variable

    This warning and all other warnings except (1), we eliminate by
    passing the -w flag to gcc.

MERGE TO STABLE BRANCH

Expunge more bash-isms.

MERGE TO STABLE

Changes to make it work with Solaris tools (/bin/sh, diff).

Julian pointed out that as 0.0 is a bit string of 0s (at least for IEEE),
it's an easy result to get bogusly. So use cos instead of sin, expecting 1.0
back. Has all the other nice properties, plus better error detection.

There is no file getPermissions001 on Windows, so comment out this test.
There's no other obvious source of executable files which are named the same
on Windows and Unix either. If you really want to, you could always set the
permissions (if that's possible from Haskell) and then test them, but for
now content ourselves with non-executable files and directories.

PLEASE MERGE WITH STABLE BRANCH. (Also my last commit to the test suite,
which I forgot to label.)

Change this test to make it mathematically trivial, but keep the ccall
intact. This has two salutary effects:

1. We can expect to get the same result (sin 0.0 = 0.0) on all systems.

2. sin exists on Windows. erf doesn't. Hence the test now works on Windows.

Update expected output; MERGE TO STABLE BRANCH

update expected output

Update test after the behaviour of scan{r,l}1 changed

Make it work better on windows MERGE TO STABLE BRANCH

Make it work on Windows

note about expected failure rnfail034 (nothing major, just an error
message wibble)

Move Numeric tests into lib/Numeric

Expand this test to show a range of floats and doubles using
showEFLoat, showFFloat and showGFloat with a range of precisions, and
also attempt to "read" back the results.

Correct spurious instructions.

Update this test for the changes to the Ix class.  In particular, we
can no longer depend on Eq being a superclass of Ix, so when comparing
indices we have to use Ix.index and comapre the Int results.

bugfix in check_stderr_ok()

Add irrefutable pattern test from the Hugs bugs list.

rn030 was a duplicate of rn026

add '!!!' to the comment in here

Test that a qualified name can't be bound in a local binding group.

Currently an expected failure because the error message is slightly
wrong (has a duplicated "in").

note that rn038 is a new expected failure.

Test that hiding a non-exported identifier is legal.

(Currently an expected failure, as GHC doesn't implement this change
yet.  I was surprised to discover this proposed change in Haskell 98,
and I think I'd prefer if this case were an error.)

Test that using hiding on an unqualified import also hides the qualified name.

Test that hiding works on a qualified import

Add a test to make sure that qualified imports can be restricted to
the specified entities.

Add a test for unqualified imports bringing into scope qualified names.

add test for conflicting exports

comment this test

Test for exporting of an unimported module.

Add some tests for hGetBuf & hPutBuf

Modify this test to do something reasonable: it now tests that hiding
a constructor leaves the corresponding type constructor visible (this
is a bug reported by George Russell a while back).  Enable rn037 in all.T.

Add functional dependency/missing instance test.