[m-rev.] for review: Document experience on AIX.

[Peter Wang]

On Mon, 14 Jul 2014 14:06:36 +1000 (EST), Julien Fischer <jfischer at opturion.com> wrote:
> 
> Hi Peter,
> 
> On Mon, 14 Jul 2014, Peter Wang wrote:
> 
> > On Fri, 11 Jul 2014 18:15:38 +1000, Julien Fischer <jfischer at opturion.com> wrote:
> >>>
> >>> You may need to use the IBM C compiler instead of gcc.  Pass
> >>> "--with-cc=xlc"
> >>> when configuring the Mercury installation, and use high-level C grades
> >>> for better performance.
> >>>
> >>
> >> One issue here is that the configure script, Mercury.options, mgnuc script,
> >> compiler/compiile_target_code.m
> >> and probably other places won't recognise xlc as a type of C compiler --
> >> the type should default
> >> to unknown, but may incorrectly default to GCC in spots.
> >
> > I found one instance with grep -i for "compiler[-_ ]type",
> > in reserve_stack_size_flags.
> 
> I'm not sure what you grepped there.  I count at least six uses of
> get_c_compiler_type in compiler/compile_target_code.m (admittedly, most
> of them are workarounds for Visual C weirdness).

git grep -i 'compiler[-_ ]type'

> 
> I suggest adding a cc_xlc alternative to the type
> globals.c_compiler_type and also recognising the options variable
> XLC_FLAGS (we already support GCC_FLAGS, CLANG_FLAGS etc).
> 
> > Here is an updated document for review.
> 
> That's fine although most of the various workarounds could (and should
> IMO) be applied by automatically if we recognised xlc as a separate sort
> of C compiler.  (Incidentally, I believe xlc has a GCC compatibility
> wrapper.
> 
> For GCC 4.2 on AIX, I think you should make the configure script print
> an error message saying that the version of GCC is too ancient and
> broken and then abort.

Nah.

Peter