[MLton] MLton.TextIO.mkstemps

[Matthew Fluet]

> Something that will make the problem (with the assembler unlinking the file)
> MUCH less likely to cause a problem is to include the process ID as part
> of the file name (as part of the random bytes).  That way collisions are
> much less common (even though the race condition still exists).

Right.  That seems reasonable.

> I was looking at the code trying to make sure that we (MLton) was never
> incorrectly unlinking the file until all uses were done.  I didn't see any,
> but it wasn't completely clear to me.

I'm pretty sure we're o.k.  Looking at an strace of a compilation shows 
all the ulinking of .S and .c files happens at the end.