Hi Simon.
I've made the following change to get nofib tests running for Cygwin hosted
MinGW32 builds, at least on my computer:
Index: runstdtest.prl
===================================================================
RCS file: /home/cvs/root/fptools/glafp-utils/runstdtest/runstdtest.prl,v
retrieving revision 1.31
diff -r1.31 runstdtest.prl
56a57,63
> # If this is Cygwin, ignore eol and CR characters.
> # Perhaps required for MSYS too, although the cygpath
> # bit is hopefully unnecessary.
> if ( `uname | grep CYGWIN` ) {
> $CONTEXT_DIFF=$CONTEXT_DIFF . " --strip-trailing-cr" ;
> $TmpPrefix = `cygpath -m $TmpPrefix | tr -d \\\\n`;
> }
=======================================================================
I can't test this on a Unix platform so please excuse any overnight test
failures which might result.
The next exciting bug in nofib on Windows is now revealed:
------------------------------------------------------------------------
==fptools== make all - --unix - --no-print-directory -r;
in /c/cvs/head/i386-unknown-mingw32/nofib/spectral/fft2
------------------------------------------------------------------------
HC = /c/cvs/head/i386-unknown-mingw32/ghc/compiler/stage3/ghc-inplace
HC_OPTS = -H16m -O -O -Rghc-timing -H32m -hisuf hi
RUNTEST_OPTS = -ghc-timing +RTS -H10m -K10m -RTS
==nofib== fft2: size of fft2 follows...
text data bss dec hex filename
1354692 70320 57072 1482084 169d64 fft2.exe
==nofib== fft2: time to run fft2 follows...
/usr/bin/time ../../../glafp-utils/runstdtest/runstdtest ./fft2 \
\
-o1 fft2.stdout -o1 fft2.stdout-mingw -o1 fft2.stdout1 -o1
fft2.stdout2 -o1 ff
t2.stdout3 -o1 fft2.stdout4 -o1 fft2.stdout -o1 fft2.stdout-mingw -o1
fft2.stdou
t1 -o1 fft2.stdout2 -o1 fft2.stdout3 -o1 fft2.stdout4 \
\
-ghc-timing +RTS -H10m -K10m -RTS
/bin/sh -c ././fft2 +RTS -Sc:/DOCUME~1/miketh/LOCALS~1/Temp/stats8152 -RTS
+RTS
-H10m -K10m -RTS < /dev/null 1>
c:/DOCUME~1/miketh/LOCALS~1/Temp/runtest8152.1
2> c:/DOCUME~1/miketh/LOCALS~1/Temp/runtest8152.2 3>
c:/DOCUME~1/miketh/LOCALS~1
/Temp/runtest8152.3
././fft2 +RTS -H10m -K10m -RTS < /dev/null
expected stdout not matched by reality
*** fft2.stdout Wed Nov 3 02:12:56 1999
--- c:/DOCUME~1/miketh/LOCALS~1/Temp/runtest8152.1 Mon Sep 20 17:28:51
2004
***************
*** 2,3 ****
result2 = 2.65193921505278e-12
! result3 = 3.4766401313390816e-8
--- 2,3 ----
result2 = 2.65193921505278e-12
! result3 = 4.829354338653502e-8
Command exited with non-zero status 1
1.09user 0.56system 0:01.64elapsed 100%CPU (0avgtext+0avgdata
447872maxresident)
k
0inputs+0outputs (28528major+0minor)pagefaults 0swaps
make[2]: *** [runtests] Error 1
make[1]: *** [all] Error 1
make: *** [all] Error 1
========================================================================
It may be that this is merely the result of differences in numerical
precision across platforms, and given the comment at the bottom of Main.lhs,
it might be best just to have a test, with some kind of tolerance say 1e-6,
for closeness to zero.
As a matter of interest, should this test failure cause the entire nofib
suite to fall over?
Cheers
Mike Thomas
_______________________________________________
Glasgow-haskell-bugs mailing list
[EMAIL PROTECTED]
http://www.haskell.org/mailman/listinfo/glasgow-haskell-bugs
