mlprof bug

[Matthew Fluet]

> > I would vote for 4.  Why not just do it in place?
> 
> I don't really mind that either.  It is slower, but who cares.  

Well, I only discovered this bug because I wanted to do profiling of a
self-compile.  Coloring a over a thousand .dot files isn't going to be
quick in any case, but the faster the better.  And some of those files
are decent sized.  (As a side note, -keep ssa -keep dot -keep-pass backend
-keep g on a self compile yields close to 200Meg of data!)

> But if
> there is a general problem with File.move, (2) might be nice to do
> independently of (4).

Yet another solution that struck me would be simply to deal with this
particular idom; i.e., that of rewriting via a temporary file.  That has
particular properties, like it's unlikely that anythink linked to the
temporary file was it was in existence.  So, it's reasonably to try harder
to accomplish the task with less feedback (i.e., exceptions) to the user.

But, I dont otherwise have a strong opinion on handling the error in
File.move or just in mlprof.  I've rarely used the File structure, so I
don't have a particular mental model.