> A typo:
>
> $ diff fptools-orig/ghc/rts/RtsFlags.c fptools/ghc/rts/RtsFlags.c
> 219c219
> < " -M<size> Sets the maximum heap size (default 64M) Egs:
> -M256k -M1G",
> ---
> > " -M<size> Sets the maximum heap size (default 64M) Egs:
> -H256k -H1G",
Thanks, fixed.
> Then, ghc refused to compile because of lacking heap. (Although this
> seems a little strange: how did anybody compile it then?)
>
> $ diff fptools-orig/ghc/compiler/Makefile
> fptools/ghc/compiler/Makefile
> 193c193
> < rename/ParseIface_HC_OPTS += -Onot -H75m
> -fno-warn-incomplete-patterns
> ---
> > rename/ParseIface_HC_OPTS += -Onot -H45m
> -fno-warn-incomplete-patterns
Firstly, use -dcore-lint when compiling this module (it seems to alleviate a
space leak). Then ParseIface will compile in about 55M, which is still
very, very bad, but not *as* bad.
The options I use most often to bootstrap the compiler are
GhcHcOpts = -O -dcore-lint -H24m
and possibly -W for fun.
> To increase the heap to 75m, I head to increase the heap limit (-M) as
> well, but the ghc-4.02 from the binary distribution claimed
> not to know
> about this option. So I browsed through the driver and found the
> following workaround. (Maybe it's clear to you what is going
> wrong here
> and you know a proper fix? Is it me again?)
Use -optCrts-M<size> to pass this option to the compiler proper.
Unfortunately, we've already used -M for something else (to run mkdependHS).
And we really don't anticipate needing to pass -M to the compiler very
often.
Just gotta fix that space leak <sigh>.
Cheers,
Simon
