Unreproducible bug

[Stephen Weeks]

> The files that I've seen it occur on were:
> regression/array.sml
> regression/klife_eq.sml
> and a self-compile.
> 
> The only other thing I can possibly think of might be that I was
> (SML/NJ) recompiling MLton while the regressions were going.  I don't know
> what NJ does if it's image file is rewritten while it's running -- I would
> have assumed it loads it all in at invocation.

I would too, but who knows.  I'm willing to discount array and klife based on
this.

> But I wouldn't have done
> that on the self-compile, since I set that going before I left last night.

The self compile error should be reproducible then.  This is entirely plausible,
since you're using a new mlton.sml as input to the simplifier, and hence may be
tickling new bugs.  When you get such an error, please save mlton.sml and see if
you can reproduce the bug.  Thanks.