Tactic.ASSUME

Given a goal g of the form {x = y, y = z} ?- P, where x, y and z have type :'a, the theorem x = y, y = z |- x = z can, first, be inferred by forward proof

   let val eq1 = Term `(x:'a) = y`
       val eq2 = Term `(y:'a) = z`
   in
   TRANS (ASSUME eq1) (ASSUME eq2)
   end;

and then added to the assumptions. This process requires the explicit text of the assumptions, as well as invocation of the rule ASSUME:

   let val eq1 = Term `(x:'a) = y`
       val eq2 = Term `(y:'a) = z`
       val goal = ([eq1,eq2],Parse.Term `P:bool`)
   in
   ASSUME_TAC (TRANS (ASSUME eq1) (ASSUME eq2)) goal
   end;

   val it = ([([`x = z`, `x = y`, `y = z`], `P`)], fn) : tactic_result

This is the naive way of manipulating assumptions; there are more advanced proof styles (more elegant and less transparent) that achieve the same effect, but this is a perfectly correct technique in itself.

Alternatively, the axiom EQ_TRANS could be added to the assumptions of g:

   let val eq1 = Term `(x:'a) = y`
       val eq2 = Term `(y:'a) = z`
       val goal = ([eq1,eq2], Term `P:bool`)
   in
   ASSUME_TAC EQ_TRANS goal
   end;

   val it =
     ([([`!x y z. (x = y) /\ (y = z) ==> (x = z)`, 
         `x = y`,`y = z`],`P`)],fn) : tactic_result

A subsequent resolution (see RES_TAC) would then be able to add the assumption x = z to the subgoal shown above. (Aside from purposes of example, it would be more usual to use IMP_RES_TAC than ASSUME_TAC followed by RES_TAC in this context.)