[decompiler] Get rid of FormStack peek and fix virtual method calls

[water111]

I was expecting this to be much harder, so I'm a little suspicious that this isn't going to work 100% of the time.

[coveralls]

Pull Request Test Coverage Report for Build 845297566

• 25 of 25 (100.0%) changed or added relevant lines in 2 files are covered.
• 6 unchanged lines in 1 file lost coverage.
• Overall coverage decreased (-0.01%) to 69.178%

---

Files with Coverage Reduction	New Missed Lines	%
decompiler/IR2/FormExpressionAnalysis.cpp	6	80.56%

Totals
Change from base Build 845084620:	-0.01%
Covered Lines:	34425
Relevant Lines:	49763

---

💛 - Coveralls

[water111]

Yeah this didn't work. The issue would be if the method call is falsely "split".

(defmacro set-vector! (vec x y z w)
  `(begin
     (set! (-> vec x) ,x)
     (set! (-> vec y) ,y)
     (set! (-> vec z) ,z)
     (set! (-> vec w) ,w)
   )
  )

;; method call with a macro used to generate an argument
(some-method obj (set-vector! (new 'stack 'vector) 1. 2. 3. 4.))

(let ((x (method-of-object obj some-method))
      (vec (new 'stack 'vector)))
  (set! (-> vec x) 1.)
  (set! (-> vec y) 2.)
  (set! (-> vec z) 3.)
  (set! (-> vec w) 4.)
  (x vec) ;; missing obj
  )

I really don't like the peek system. It seems like it would be better to ditch it and treat a method call as normal use of variables. Then, if it makes a mess with multiply accessing an object for the vtable and the this argument, we can clean that up manually in a later pass (or part of let pass). When it was first implemented, there was no let detection, which would make the clean up much harder. Now I think this is the better approach, and would let us delete all the "peek" junk.

[water111]

Will close #255

This finally ditches the awful peek system and weird special casing for virtual methods. The new strategy is to treat a virtual method just like any other method, but when popping the object, we allow it to be used twice. This wasn't possible in the original framework because we didn't have SSA based use/def counts.

I originally tried to fix this up at the let stage, but that didn't work when the method call was embedded in another expression - the temp var for the this blocked other unrelated variables from propagating properly.