[MLton] cvs commit: added flag -sequence-unit {false|true}

[Stephen Weeks]

> By the "one spurious error in the MLton library", do you mean that
> you didn't see any other type errors in the MLton library or that
> all the other ones were acceptable (i.e., weren't a bug)?

I meant that with -sequence-unit true, hash-set.sml fails to compile
because the lookupOrInsert is forced to return type unit, which means
that the hash set must be of type "unit t".

> Because I see a bunch of errors in various lib/mlton/* files.

I see those too.  Unlike the one in hash-set.sml, all of those are
reported as warnings.


I was certainly unclear and inconsistent on whether -sequence-unit
introduces warnings or errors.  Right now, it unifies the type with
unit, which may succeed and cause a later unification error (as with
hash-set.sml), or it may fail and issue a warning.  I guess it would
be best to report an error, not a warning, in the latter case.