Sven Panne writes:
> After applying any conceivable patch in this mailing list, ghc-2.05
> built (almost) fine on our Linux- and HP-boxes. A few glitches:
>
> * As usual, /tmp is much too small on our machines. The hints from
> installing.lit:
>
> The quickest way around it is \tr{setenv TMPDIR /usr/tmp} or
> even \tr{setenv TMPDIR .} (or the equivalent incantation with the
> shell of your choice).
>
> The best way around it is to say
> \begin{verbatim}
> TMPDIR=<dir>
> \end{verbatim}
> in your @build.mk@ file.
>
> But either way, that doesn't do the right thing: Setting TMPDIR
> seems to be ignored during the compilation of ghc-2.05 (the
> compilation log contains some "TMPDIR=/tmp; export TMPDIR; ..."
> lines). Hacking config.mk permanently sets TMPDIR in the installed
> ghc driver script. But having my personal ~/tmp being compiled in
> isn't a very cunning plan (to use Baldrick's words... :-)
> In a rare glimpse of ingenuity ;-) , I tried "make TMPDIR=~/tmp all",
> which seems to work. Perhaps the docs and/or Makefiles need some
> fine-tuning here.
>
A glitch - GNU make only exports variables defined in the environment
it was invoked from (or on the cmd-line), there should be an 'export'
qualifier for the above build.mk suggestion.
>
> * fptools/ghc/lib/cbits/timezone.h is a show-stopper again: On
> Linux (and any other sensible OS, I presume), the type of timezone
> is time_t. But @#$%&! HP-UX-10.20 thinks it's long. This should
> better be figured out by configure. (I think I've mailed this
> already somtime ago, but anyway...)
>
I promised to do something about this, but didn't (sorry). I've added
the necessary configure test now - hopefully you'll dare to have
another go with 2.06..
> To share my happiness with others, I make the usual unofficial snapshots
> available at
>
>
>ftp://ftp.informatik.uni-muenchen.de/local/pms/ghc-2.05-all-i386-unknown-linux.tar.gz
>
>ftp://ftp.informatik.uni-muenchen.de/local/pms/ghc-2.05-all-hppa1.1-hp-hpux10.tar.gz
>
> this afternoon. Unpack at / and enjoy...
>
Good job - I've updated the download page with links to these. If
you're cramped for disk space, let us know and we'll copy the distribs
over to Glasgow.
> Some questions:
>
> * What is this funny object-splitting all about? It seems to blow up
> a 2Mb object file into hundreds of smaller object files, using
> almost 30MB altogether!?
>
It's there to save disk space (really), gagging the linker with a
spoon produces smaller binaries. There's a known bloat problem with
splitting+profiling on systems that use ELF - we'll turn off splitting
by default for 2.06 (profiling only.)
> * I'd like to hear some success stories about booting ghc-2.05 with
> itself, especially compiler/library size/speed.
>
Me too! Feel free to mail this kind of stuff (success or otherwise) to
glasgow-haskell-bugs.
> * What's the schedule for porting Haggis to ghc-2.x-compliant code?
> (Standard Haskell code would be even better. :-))
> Up to now, one has the options of writing Ancient Haskell
> (= ghc-0.2x code) or giving up the nice GUI stuff.
>
I've got it running locally, so maybe when I've saved up enough
hacking credits again, it's time to punt..
Thanks for the feedback.
--Sigbjorn
