ZJIT: We are missing opportunities to elide inline frames

[tekknolagi]

def foo(x) = x + 1
def bar(x) = x + 1
def test = foo(3) + bar(4)
test
test

This gets rid of the frame for bar but not foo even though they are the same:

Optimized HIR:
fn test@inlineframe.rb:3:
bb1():
  EntryPoint interpreter
  v1:BasicObject = LoadSelf
  Jump bb3(v1)
bb2():
  EntryPoint JIT(0)
  v4:BasicObject = LoadArg :self@0
  Jump bb3(v4)
bb3(v6:BasicObject):
  v11:Fixnum[3] = Const Value(3)
  PatchPoint MethodRedefined(Object@0x101a30f00, foo@0xa801, cme:0x101af6c50)
  v28:ObjectSubclass[class_exact*:Object@VALUE(0x101a30f00)] = GuardType v6, ObjectSubclass[class_exact*:Object@VALUE(0x101a30f00)] recompile
  PushInlineFrame :foo, v28 (0x101a45540), num_args=1
  v45:Fixnum[1] = Const Value(1)
  PatchPoint MethodRedefined(Integer@0x101a344c0, +@0x2b, cme:0x101a1c280)
  v84:Fixnum[4] = Const Value(4)
  PopInlineFrame
  v16:Fixnum[4] = Const Value(4)
  PatchPoint MethodRedefined(Object@0x101a30f00, bar@0xa811, cme:0x101af6c78)
  v86:Fixnum[9] = Const Value(9)
  CheckInterrupts
  Return v86

[k0kubun]

Copying my reply on Slack:

---

PatchPoint was previously a blocker and I don't think we have worked on fixing that yet.

The problem for this ISEQ seems like PatchPoint on Integer#+, not the ones on foo or bar. Because the duplicated Integer#+ PatchPoint can be eliminated for the bar call, it works only for bar right now.