A bit of context and then my question.
https://gitlab.haskell.org/ghc/ghc/blob/master/compiler/typecheck/TcCanonica...
declares `rewriteEvidence` with the comment that "rewriteEvidence old_ev
new_pred co ... Returns a new_ev : new_pred, with same
wanted/given/derived flag as old_ev".
However, the `CtWanted` case just immediately calls `newWanted`, which
ignores the incoming `ctev_nosh` flag and always uses `WDeriv`.
```
rewriteEvidence ev@(CtWanted { ctev_dest = dest
, ctev_loc = loc }) new_pred co
= do { mb_new_ev <- newWanted loc new_pred
```
(This is true in 8.6.5 and also right now at
6febc444c0abea6c033174aa0e813c950b9b2877.)
Thus I'm seeing a [W] is become a [WD], which seems to contradict the
comment's "same flag" claim.
MY QUESTION ======> Is this "WD unsplitting" the intended behavior? See
canonicalization tc-trace below.
If it is intended, I'll probably write a follow-up question asking for more
advice. The constraint solver is looping unproductively because my plugin
keeps discarding the (re-)emitted [D] shadow as a tautology, then the [W]
"unsplits" during canonicalization (even though my plugin didn't touch the
[W]), this re-emits the same [D] (via a fundep), and so on.
My continual disclaimer recently is that I'm seeing this behavior with my
in-development plugin activated, so it's entirely I'm pushing GHC into
unfamiliar waters or even just plain violating some invariants without
realizing it. But this trace looks rather isolated from plugins.
Thanks! -Nick
Here's a -ddump-tc-trace of a canonicalization that goes from [W] to [WD]
happen (with GHC 8.6.5):
runStage canonicalization {
workitem = [W] $dMonadFree_a35V {0}:: MonadFree
((fs_a2UF[sk:1] :- Const
e_a2UE[sk:1])
:+ Const e_a2UE[sk:1])
(Free fs_a2UF[sk:1]) (CDictCan)
flatten_args {
(fs_a2UF[sk:1] :- Const e_a2UE[sk:1]) :+ Const e_a2UE[sk:1]
Free fs_a2UF[sk:1]
matchFamTcM
Matching: (fs_a2UF[sk:1] :- Const e_a2UE[sk:1])
:+ Const e_a2UE[sk:1]
Match failed
matchFamTcM
Matching: fs_a2UF[sk:1] :- Const e_a2UE[sk:1]
Match failed
Unfilled tyvar fs_a2UF[sk:1]
Unfilled tyvar e_a2UE[sk:1]
flatten/flat-cache hit
:- [* -> *, fs_a2UF[sk:1], Const e_a2UE[sk:1]]
s_a35Z[fmv:1]
Unfilled tyvar s_a35Z[fmv:1]
Unfilled tyvar e_a2UE[sk:1]
matchFamTcM
Matching: s_a35Z[fmv:1] :+ Const e_a2UE[sk:1]
Match failed
New coercion hole: co_a365
Emitting new coercion hole
{co_a365} :: (s_a35Z[fmv:1] :+ Const e_a2UE[sk:1])
GHC.Prim.~# s_a364[fmv:1]
extendFlatCache
:+ [* -> *, s_a35Z[fmv:1], Const e_a2UE[sk:1]]
[WD]
s_a364[fmv:1]
flatten/flat-cache miss
:+ [* -> *, s_a35Z[fmv:1], Const e_a2UE[sk:1]]
s_a364[fmv:1]
[WD] hole{co_a365} {0}:: (s_a35Z[fmv:1] :+ Const e_a2UE[sk:1])
GHC.Prim.~# s_a364[fmv:1]
Unfilled tyvar fs_a2UF[sk:1]
flatten }
s_a364[fmv:1]
Free fs_a2UF[sk:1]
Emitting new wanted
$dMonadFree_a366 :: MonadFree s_a364[fmv:1] (Free fs_a2UF[sk:1])
arising from a use of ‘wrap’ at tests/ill-typed/T7.hs:77:14-17
addTcEvBind
a2V1
[W] $dMonadFree_a35V
= $dMonadFree_a366
`cast` ((MonadFree
(Sym {co_a365} ; ((:+)
<* -> *>_N (Sym {co_a360})