Incorrect callstack trace when function is used multiple times in different branches

[SeungheonOh]

Consider

foo :: forall s. HasCallStack => Term s (PInteger :--> PInteger)
foo = plam $ \(_x :: Term s PInteger) -> (pthrow "nope") :: Term s PInteger
bob :: forall s. HasCallStack => Term s (PInteger :--> PInteger)
bob = plam $ \(x :: Term s PInteger) -> foo # x

rob :: forall s. HasCallStack => Term s (PInteger :--> PInteger)
rob = plam $ \(x :: Term s PInteger) -> foo # x

mine :: forall s. HasCallStack => Term s (PInteger :--> PInteger)
mine = plam $ \(x :: Term s PInteger) ->
  pif
    (x #== 10)
    (bob # x)
    (rob # x)

-- and execute below

λ let Left a = evalTerm mempty (mine # pconstant @PInteger 11) in putStrLn $ unpack a
CallStack (from HasCallStack):
  pthrow', called at ./Plutarch/Internal/Term.hs:632:24 in plutarch-1.10.1-inplace:Plutarch.Internal.Term
  pthrow, called at <interactive>:52:43 in interactive:Ghci20
  foo, called at <interactive>:54:41 in interactive:Ghci20
  bob, called at <interactive>:60:58 in interactive:Ghci20
  mine, called at <interactive>:66:31 in interactive:Ghci25

nope

λ let Left a = evalTerm mempty (mine # pconstant @PInteger 10) in putStrLn $ unpack a
CallStack (from HasCallStack):
  pthrow', called at ./Plutarch/Internal/Term.hs:632:24 in plutarch-1.10.1-inplace:Plutarch.Internal.Term
  pthrow, called at <interactive>:52:43 in interactive:Ghci20
  foo, called at <interactive>:54:41 in interactive:Ghci20
  bob, called at <interactive>:60:58 in interactive:Ghci20
  mine, called at <interactive>:67:31 in interactive:Ghci26

nope

Notice how they have exactly same callstack trace where first must show foo -> rob -> mine. This is not particularly unexpected and I think it will consistently take the first branch to generate callstack. It will be difficult to make it generate correct callstack since that means we need a different copies of foo for each "branches." Generating "correct" callstack would also mean duplicating a ton of code obviously.

So, I don't think we should necessarily fix this by making correct behavior default. We should at least

• Make note of current shortcoming clearly
• (if we have time and passion) Add new compilation option that will generate correct callstack trace with note that will severely bloat amount of generated code.

[SeungheonOh]

This is the correct behavior: IntersectMBO/plutus#7148 (comment)

This was implemented by having a runtime callstack bindings, which does bring some overhead on execution cost, but callstack is for testing anyways so who cares.