windows head build problems: Setup.exe: errorreading dist\setup-config;

[Claus Reinke]

> Incidentally, I've just realised that if it did work then it would also
> tell you that you have added all the files in all the libraries,
> testsuite, etc.

indeed, and one has to be careful when filtering those out, because
mixed in between are real left-over files that need removing.. that's
one reason why a proper darcsclean target would be useful.

>> the libraries/Makefile diff is my ghc-inplace.exe change, but most of
>> the other files should really have disappeared after distclean, right?
>
> Yup; I have ambitions to go through making sure everything gets cleaned
> at some point before the release.

something to look forward to, then!-) i assume i'm not the only
one who only rebuilds his ghc head occasionally, and gets bitten
by leftovers often, especially after high-activity phases.

having cleaned as much as i could, the build seems to go through,
but with some missing file messages, of two kinds:

    ../utils/mkdependC/mkdependC -f 
.depend-BASE -D__GLASGOW_HASKELL__=607 -I../includes    -- -mno-cygwin -O -Iparser -I. -O    --  
parser/cutils.c parser/hschooks.c
    In file included from ../includes/Stg.h:150,
                     from ../includes/Rts.h:19,
                     from parser/cutils.c:6:
    ../includes/Regs.h:30:45: gmp.h: No such file or directory
    In file included from ../includes/Stg.h:150,
                     from ../includes/Rts.h:19,
                     from parser/cutils.c:6:
    ../includes/Regs.h:30:45: gmp.h: No such file or directory

the only gmp.h is in ../gmp/gmpbuild/

    GNUmakefile:61: .depend: No such file or directory

this appears to be an include that follows the build rule for .depend,
but the message appears *before* that build rule is used.

also, there is this error message, in gmp/Makefile:

    set -o igncr; export SHELLOPTS; \
        export PATH=`pwd`:$PATH; \
        cd gmpbuild && \
        CC=c:/MinGW/bin/gcc /bin/sh configure \
              --enable-shared=no --host=i386-unknown-mingw32 --build=i386-unknown-mingw32
    /bin/sh: line 0: set: igncr: invalid option name

if i understand correctly, exactly those versions of sh that
understand this option will need it, due to a change in default
behaviour, so the message is harmless?

claus