Improve exprOkForSpeculation for classopswip/T22745

This patch fixes #22745 and #15205, which are about GHC's
failure to discard unnecessary superclass selections that
yield coercions.  See
  GHC.Core.Utils Note [exprOkForSpeculation and type classes]

The main changes are:

* Write new Note [NON-BOTTOM_DICTS invariant] in GHC.Core, and
  refer to it

* Define new function isTerminatingType, to identify those
  guaranteed-terminating dictionary types.

* exprOkForSpeculation has a new (very simple) case for ClassOpId

* ClassOpId has a new field that says if the return type is
  an unlifted type, or a terminating type.

This was surprisingly tricky to get right.  In particular note
that unlifted types are not terminating types; you can write an
expression of unlifted type, that diverges.  Not so for dictionaries
(or, more precisely, for the dictionaries that GHC constructs).

Metric Decrease:
    LargeRecord

1 files changed, 40 insertions, 11 deletions

diff --git a/compiler/GHC/Core.hs b/compiler/GHC/Core.hs
index b18ec7acda..809332d395 100644
--- a/compiler/GHC/Core.hs
+++ b/compiler/GHC/Core.hs
@@ -292,7 +292,7 @@ data AltCon
 -- This instance is a bit shady. It can only be used to compare AltCons for
 -- a single type constructor. Fortunately, it seems quite unlikely that we'll
 -- ever need to compare AltCons for different type constructors.
--- The instance adheres to the order described in [Core case invariants]
+-- The instance adheres to the order described in Note [Case expression invariants]
 instance Ord AltCon where
   compare (DataAlt con1) (DataAlt con2) =
     assert (dataConTyCon con1 == dataConTyCon con2) $
@@ -466,6 +466,45 @@ we need to allow lots of things in the arguments of a call.
 
 TL;DR: we relaxed the let/app invariant to become the let-can-float invariant.
 
+Note [NON-BOTTOM-DICTS invariant]
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+It is a global invariant (not checkable by Lint) that
+
+     every non-newtype dictionary-typed expression is non-bottom.
+
+These conditions are captured by GHC.Core.Type.isTerminatingType.
+
+How are we so sure about this?  Dictionaries are built by GHC in only two ways:
+
+* A dictionary function (DFun), arising from an instance declaration.
+  DFuns do no computation: they always return a data constructor immediately.
+  See DFunUnfolding in GHC.Core.  So the result of a call to a DFun is always
+  non-bottom.
+
+  Exception: newtype dictionaries.
+
+  Plus: see the Very Nasty Wrinkle in Note [Speculative evaluation]
+  in GHC.CoreToStg.Prep
+
+* A superclass selection from some other dictionary. This is harder to guarantee:
+  see Note [Recursive superclasses] and Note [Solving superclass constraints]
+  in GHC.Tc.TyCl.Instance.
+
+A bad Core-to-Core pass could invalidate this reasoning, but that's too bad.
+It's still an invariant of Core programs generated by GHC from Haskell, and
+Core-to-Core passes maintain it.
+
+Why is it useful to know that dictionaries are non-bottom?
+
+1. It justifies the use of `-XDictsStrict`;
+   see `GHC.Core.Types.Demand.strictifyDictDmd`
+
+2. It means that (eq_sel d) is ok-for-speculation and thus
+     case (eq_sel d) of _ -> blah
+   can be discarded by the Simplifier.  See these Notes:
+   Note [exprOkForSpeculation and type classes] in GHC.Core.Utils
+   Note[Speculative evaluation] in GHC.CoreToStg.Prep
+
 Note [Case expression invariants]
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 Case expressions are one of the more complicated elements of the Core
@@ -556,10 +595,6 @@ substitutions until the next run of the simplifier.
   Note [Equality superclasses in quantified constraints]
   in GHC.Tc.Solver.Canonical
 
-Note [Core case invariants]
-~~~~~~~~~~~~~~~~~~~~~~~~~~~
-See Note [Case expression invariants]
-
 Note [Representation polymorphism invariants]
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 GHC allows us to abstract over calling conventions using **representation polymorphism**.
@@ -627,12 +662,6 @@ representation: we check whether bound variables and function arguments have a
 See Note [Representation polymorphism checking] in GHC.Tc.Utils.Concrete
 for an overview of how we enforce these invariants in the typechecker.
 
-Note [Core let goal]
-~~~~~~~~~~~~~~~~~~~~
-* The simplifier tries to ensure that if the RHS of a let is a constructor
-  application, its arguments are trivial, so that the constructor can be
-  inlined vigorously.
-
 Note [Empty case alternatives]
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 The alternatives of a case expression should be exhaustive.  But