ghci and other test failures on Windows.

[Mike Thomas]

Hi Simon.

| > The other thing I was planning on looking into in the near future is
| > the consistent failure of the GHCi versions of the tests on Windows.
|
| That would be good - we're in bug-fixing mode for 6.2.2 at the moment,
| planning to put out a release candidate soon.

As time may be running short I thought I had better keep you up to date on
where this went - applies to CVS HEAD from last week, MinGW build hosted on
Cygwin.

I finally got round to it at 5.00 am on Sunday morning while waiting for
another build at work and I am too tired to chase it all up!


============================================================================
=
1. The ghci tests fail under the nightly build system.  For example:

  -rw-rw-rw-   1 miketh   root            0 08-06 15:09 arith001.comp.stderr
  -rw-rw-rw-   1 miketh   root          104 08-06 15:09 arith001.comp.stdout
  -rwxrwxrwx   1 miketh   root      2422282 08-06 15:09 arith001.exe
  -rw-rw-rw-   1 miketh   root          213 08-06 15:09 arith001.genscript
  -rw-rw-rw-   1 miketh   root          351 08-06 15:09 arith001.hi
  -rw-rw-rw-   1 miketh   root          657 2001-06-26  arith001.hs
  -rw-rw-rw-   1 miketh   root           35 08-06 15:09
arith001.interp.stderr
  -rw-rw-rw-   1 miketh   root          104 08-06 15:09
arith001.interp.stdout
  -rw-rw-rw-   1 miketh   root        18295 08-06 15:09 arith001.o
  -rw-rw-rw-   1 miketh   root            0 08-06 15:09 arith001.run.stderr
  -rw-rw-rw-   1 miketh   root           70 08-06 15:09 arith001.run.stdout
  -rw-rw-rw-   1 miketh   root           65 07-06 16:04 arith001.stdout

give the following:

=====> arith001(ghci)
cd ./numeric/should_run &&
'/cygdrive/c/cvs/head/i386-unknown-mingw32/ghc/compiler/stage3/ghc-inplace' 
-no-recomp -dcore-lint -Di386_unknown_mingw32 arith001.hs --interactive -v0
<arith001.genscript 1>arith001.interp.stdout 2>arith001.interp.stderr
Actual stdout output differs from expected:
*** ./numeric/should_run/arith001.stdout	Tue Jul  6 16:04:44 2004
--- ./numeric/should_run/arith001.run.stdout	Fri Aug  6 15:09:34 2004
***************
*** 1,5 ****
- 42%1
- 42, 1
- 42.0, 1.0
- 5910974510923776, -47
- 5910974510923776, -47
--- 0 ----
*** unexpected failure for arith001(ghci)

Both files are identical other than for the line endings:

42%1
42, 1
42.0, 1.0
5910974510923776, -47
5910974510923776, -47

42%1
42, 1
42.0, 1.0
5910974510923776, -47
5910974510923776, -47

At first glance, the Python test code (check_stdout_ok) accounts for line
ending differences so perhaps the two files are getting mixed up or the
error message is erroneous in the sense that it might be related to the
stderr file - I'm too tired to tell.


============================================================================
=
2. On a related note, ghci after make install looks in the wrong place for
(I think) system.IO, which is imported from InteractiveUI.hs: import
System.IO as IO, line 68)

-----------------------------------------------------------------------
*** Parser:

<interactive>:1:22:
    Failed to load interface for `IO':
        Could not find interface file for `IO'
        locations searched: c:/lang/ghc/imports/IO.hi
*** Deleting temp files
Deleting:
ghc.exe: panic! (the `impossible' happened, GHC version 6.3):
        interactiveUI:setBuffering

Please report it as a compiler bug to glasgow-haskell-bugs at haskell.org,
or http://sourceforge.net/projects/ghc/.
-----------------------------------------------------------------------

System/IO.hi seems to be correctly installed as shown below, so my guess is
that the "import as" alias is going wrong by not remembering the actual
hierarchical path to the true import file.

-----------------------------------------------------------------------
$ find /c/lang/ghc -iname IO.hi
/c/lang/ghc/imports/Data/Array/IO.hi
/c/lang/ghc/imports/GHC/IO.hi
/c/lang/ghc/imports/System/IO.hi
-----------------------------------------------------------------------


=======================================================================
3.  The next step after fixing the test script issue for point 1 for the
Windows regression tests will need to be the Windows compiler crashes and
network hangs reported a while ago - no promises of course - the problems I
expect are long standing.

Cheers

Mike Thomas.