Sven Panne writes:
> GHC 3.02 compiled successfully on Linux (our HPs ran out of /tmp-space, but
> stay tuned), but there are some buglets:
>
> When configuring GHC, there are a few errors/warnings, but compilation
> continues anyway:
>
> [...]
> /mnt/liesl1/inst/panne/fptools/ghc/utils/unlit:
> configure.in:499: warning: AC_TRY_RUN called without default to allow cross
>compiling
> /soft/bin/autoconf: configure: Permission denied
> configure.in:499: warning: AC_TRY_RUN called without default to allow cross
>compiling
> /soft/bin/autoheader: Symbol `ALIGNMENT_DOUBLE' is not covered by
>/soft/share/autoconf/acconfig.h
> /soft/bin/autoheader: Symbol `ALIGNMENT_FLOAT' is not covered by
>/soft/share/autoconf/acconfig.h
> /soft/bin/autoheader: Symbol `ALIGNMENT_LONG' is not covered by
>/soft/share/autoconf/acconfig.h
> /soft/bin/autoheader: Symbol `ALIGNMENT_UNSIGNED_INT' is not covered by
>/soft/share/autoconf/acconfig.h
> /soft/bin/autoheader: Symbol `LEADING_UNDERSCORE' is not covered by
>/soft/share/autoconf/acconfig.h
>
> The AC_TRY_RUN warnings are nothing serious, but what about the
> "Permission denied" error and the uncovered symbols? I thought we have
> the latest autoconf/autoheader (version 2.12)... :-(
>
The pre-generated configure script included in the source dist doesn't
have any write bits set, hence the complaint. Re: autoheader noise,
acconfig.h is missing from the source dist.
I don't recommend running autoheader, the version of config.h.in that
comes with a distribution is always up-to-date (it has to be), which
you'll potentially overwrite with a partially OK version should
anything go wrong when running autoheader locally.
> The compilation requires an almost obscene amount of temporary space,
> but somehow setting and exporting the TMPDIR environment variable
> doesn't help. Any suggestions?
>
The driver looks at the setting of TMPDIR and tries to use that (as
does gcc), so I don't know what's happening here. What kind of files
are being created in your temp area with TMPDIR set to point
elsewhere?
> When compiling the latest Green Card (290598) with -O, ghc-3.02 fails
> with an internal error:
>
> [...]
> ghc-3.02 -O -recomp -c FillIn.lhs -o FillIn.o
> ClosureInfo.lhs:1128: Non-exhaustive patterns in case
>
> H/Direct manages to provoke the same failure. I fear it's newtype
> handling again.
>
Leave out '-O' when compiling both of these two for now (I noticed
this very problem with 3.02 yesterday.)
> BTW, there are some common glitches in Green Card and H/Direct:
>
> * Both contain a space after a \ in src/Makefile, so make is telling
> me something about a missing separator (if I remember correctly).
>
> * The "higher" Makefiles don't protect their arguments with "" when
> calling their sub-Makefiles, so make HC="ghc -O" fails.
>
Fixed, re: 1st point, what does 'gmake --version' report?
--Sigbjorn
