Nonrecursive contiguous subschedule

[erichulburd]

Rebases #693 which should fix tests.

Closes #697

[stylewarning]

Both @stylewarning (me!) and @kilimanjaro will take a look.

[braised-babbage]

It would help if we added another test with more than "H" gates (just using "H 0" and "H 1" does not allow us to tell whether the "H 0" gates get reordered accidentally), and perhaps some multi-qubit ops and classical instructions as well.

[braised-babbage]

This is different from the earlier version -- previously, we had

(labels ((mark-predecessors-seen (instr)
               (cond ((resource= (instruction-resources instr)
                                 resource)
                      (setf (gethash instr seen) t))
                     ((resource-subsetp (instruction-resources instr)
                                        resource)
                      (map nil #'mark-predecessors-seen (gethash instr earlier-instrs)))))
...

which only marked those predecessors of instr which acted on a resource set equal to resource. Here you are marking everything -- is this the intent?

[braised-babbage]

Does the ordering here matter?

It seems that previously there was an explicit reverse, but it wasn't entirely clear to me how the ordering of earlier-instructions was exploited (and e.g. I'm not sure whether the logical scheduler "officially" makes any promise about this ordering)? I realize that your code has gotten rid of the above reverse, but the use of a stack has the same effect (i.e. you push them one by one and then pop them in a reversed order), so if we are doing something which depends on the order I would ask that we leave a comment to that affect (an easy check for this is just to try the above iteration in reverse order).

[notmgsk]

I'm a bit sketched out by the original code, and also by the changes here.

Unnecessarily unfriendly language IMO

[braised-babbage]

I'm a bit sketched out by the original code, and also by the changes here.

Unnecessarily unfriendly language IMO

I didn't mean it as a strong statement, but rather that it's hard for me to look at the code (e.g. even as originally written) and say if it is correct or not (or if it would break if the logical scheduler code changed, etc).