[MLton] Callback functions: how?

[Henry Cejtin]

Yes,  the  connection  id ends up just being the word which is the key in the
hash table, so you can remove the entry.

Note, if the time of the hash lookup is really to expensive (and it is really
pretty  cheap  because  the  hash is delivered as the void * argument, so you
just and it with a mask and then use that as a lookup into a list), then  you
could  use  a  trick  like this: you use an array instead of a hash table, as
discussed, but the void * argument isn't an  integer  but  a  pointer  to  an
integer  (pointing into the C heap).  This way you can re-number the integers
pointed to when the array becomes too sparsely occupied.  It  costs  you  one
level of indirection (to get the int) but is safe for space.

In  MLton,  a  `closure'  is  just  an element of a datatype, with the choice
corresponding to the code part of the closure and the rest being  the  closed
over  variables.   My  guess  is that this pretty much always ends up getting
boxed.

Re _export, note that MLton never ever puts any code in the heap.  As  above,
the  closure  is just an element of the datatype.  The choice of the datatype
tells you which static piece  of  code  to  use.   Thanks  to  whole  program
compilation,  all  those  bits  of  code are known and fixed at compile time.
Only the closed over variables cause a potentially unbounded number of cases,
and those are just kept in the data carried by that choice.