expandf : (tactic -> unit)
STRUCTURE
SYNOPSIS
Applies a tactic to the current goal, stacking the resulting subgoals.
DESCRIPTION
The function expandf is a faster version of expand. It does not use a validated version of the tactic. That is, no check is made that the justification of the tactic does prove the goal from the subgoals it generates. If an invalid tactic is used, the theorem ultimately proved may not match the goal originally set. Alternatively, failure may occur when the justifications are applied in which case the theorem would not be proved. For a description of the subgoal package, see under set_goal.
FAILURE
Calling expandf tac fails if the tactic tac fails for the top goal. It will diverge if the tactic diverges for the goal. It will fail if there are no unproven goals. This could be because no goal has been set using set_goal or because the last goal set has been completely proved. If an invalid tactic, whose justification actually fails, has been used earlier in the proof, expandf tac may succeed in applying tac and apparently prove the current goal. It may then fail as it applies the justifications of the tactics applied earlier.
EXAMPLE
   - g `HD[1;2;3] = 1`;

   `HD[1;2;3] = 1`

   () : void

   - expandf (REWRITE_TAC[HD;TL]);;
   OK..
   goal proved
   |- HD[1;2;3] = 1

   Previous subproof:
   goal proved
   () : void
The following example shows how the use of an invalid tactic can yield a theorem which does not correspond to the goal set.
   - set_goal([], Term `1=2`);
   `1 = 2`

   () : void

   - expandf (REWRITE_TAC[ASSUME (Term`1=2`)]);
   OK..
   goal proved
   . |- 1 = 2

   Previous subproof:
   goal proved
   () : void
The proof assumed something which was not on the assumption list. This assumption appears in the assumption list of the theorem proved, even though it was not in the goal. An attempt to perform the proof using expand fails. The validated version of the tactic detects that the justification produces a theorem which does not correspond to the goal set. It therefore fails.
USES
Saving CPU time when doing goal-directed proofs, since the extra validation is not done. Redoing proofs quickly that are already known to work.
COMMENTS
The CPU time saved may cause misery later. If an invalid tactic is used, this will only be discovered when the proof has apparently been finished and the justifications are applied.
SEEALSO
HOL  Kananaskis-10