[MLton] cvs commit: improved analysis for ref flattening

[Stephen Weeks]

> refFlatten seems to have gotten a bunch of stuff in the CML library.
> Nothing jumped out at me as being obvious as to what was being flattened;
> CML uses a lot of refs in different places; I'd hope that the imperative
> queues (represented as two list refs) got flattened.

They are.  For example, look at variable x_1457 when compiling
primes.cm.  That variable is a pair of list_5 refs in the input, and
both are flattened into the pair.