Frontend/Closure Convert questions...

Suresh Jagannathan suresh@research.nj.nec.com
Tue, 8 Feb 2000 19:16:36 -0500


I'd be inclined to think that (3) - your own Sxml -> IL
closure converter - would be the cleanest choice.  It would
give you a fair degree of independence from MLton specific
implementation details, and you'd be able to take advantage
of properties of your IL in the translation.  It is
especially appealing if you can use flow-analysis.sig along
the same lines that the current closure converter does.

I also agree that choice (1) would be the easiest to
implement in the early going, but if the CPS code spit out
by MLton upon compiling an SML program employs features of
the CPS language that your IL doesn't support (such as
variant types), it could be a problem that may require
hacking your IL.  This may not be too bad, but may be
unnecessary.

I'd discourage you from considering choice (2) - hacking
closure-convert.fun - since that's a complex piece of code
that is closely tied to the structure of the SXML and CPS
ILs.

	-- sj