simonmar:
>
> > As the external-cmm syntax isn't fully tied down, we'll have to
> > synchronise PprCmm.hs with the Cmm parser. What is the best way to do
> > this, so that we emerge with a well-defined external
> > representation for
> > Cmm?
> >
> > Also, while we're here, a couple of questions:
> >
> > * So we want both an external Cmm representation, as well as
> > a true C--
> > code generator? So PprCmm is the external-cmm printer, but we also
> > need a C-- generator?
>
> Ok, so we need these output passes:
>
> (a) Cmm -> .cmm
> (b) Cmm -> .s
> (c) Cmm -> .hc (registerised, mangled)
> (d) Cmm -> .c (unreg'd, not mangled)
>
> Now, we think that (c) and (d) can be the same (they currently are,
> after all). However they will be considerably simpler than the current
> PprAbsC, because Cmm is "macro-expanded" relative to the old AbstractC.
Ok. Yes, I see a lot of macro fiddling work in PprAbsC.
> What you currently have as PprCmm could turn into the C-- code generator
> in due course (and we'll also use it for general ppr activities).
Yes, this is what I thought it would be for.
> We also need the parser:
>
> (f) .??? -> Cmm
>
> The question is, what syntax should we use for the parser: preferably it
> should be a *superset* of one of the output formats, so that we can feed
> the output back in as input again. I say superset, because the input
> syntax will need to be able to use macros for info table construction,
> closure construction, and many other things, which are expanded using
> the Cmm "smart constructors". These won't be present in the output, of
> course.
Yup.
> We have a choice between using .hc or .cmm for the input syntax. Since
> we're not going to be feeding these files into either a C-- compiler or
> a C compiler, it doesn't really matter which we choose. Given the C--
> flavour of the backend choosing a C--ish syntax might be more
> appropriate.
Sure. Thanks, that clarifies things.
Cheers,
Don
_______________________________________________
Cvs-ghc mailing list
[EMAIL PROTECTED]
http://www.haskell.org/mailman/listinfo/cvs-ghc
