Introduce warning for loopy superclass solve

Commit aed1974e completely re-engineered the treatment of loopy
superclass dictionaries in instance declarations. Unfortunately,
it has the potential to break (albeit in a rather minor way) user code.

To alleviate migration concerns, this commit re-introduces the old
behaviour. Any reliance on this old behaviour triggers a warning,
controlled by `-Wloopy-superclass-solve`. The warning text explains
that GHC might produce bottoming evidence, and provides a migration
strategy.

This allows us to provide a graceful migration period, alerting users
when they are relying on this unsound behaviour.

Fixes #22912 #22891 #20666 #22894 #22905

13 files changed, 132 insertions, 45 deletions

diff --git a/testsuite/tests/quantified-constraints/all.T b/testsuite/tests/quantified-constraints/all.T
index b3d0eb920f..d3bfe6e07e 100644
--- a/testsuite/tests/quantified-constraints/all.T
+++ b/testsuite/tests/quantified-constraints/all.T
@@ -41,3 +41,4 @@ test('T22216d', normal, compile, [''])
 test('T22216e', normal, compile, [''])
 test('T22223', normal, compile, [''])
 test('T19690', normal, compile_fail, [''])
+
diff --git a/testsuite/tests/typecheck/should_compile/T20666b.hs b/testsuite/tests/typecheck/should_compile/T20666b.hs
new file mode 100644
index 0000000000..7a655b73c9
--- /dev/null
+++ b/testsuite/tests/typecheck/should_compile/T20666b.hs
@@ -0,0 +1,12 @@
+{-# LANGUAGE UndecidableInstances #-}
+{-# LANGUAGE TypeFamilies #-}
+
+module T20666b where
+
+type family F a
+
+class Eq (F a) => D a
+class Eq (F a) => C a
+
+instance D [a] => C [a]
+
diff --git a/testsuite/tests/typecheck/should_compile/T20666b.stderr b/testsuite/tests/typecheck/should_compile/T20666b.stderr
new file mode 100644
index 0000000000..322c6850ea
--- /dev/null
+++ b/testsuite/tests/typecheck/should_compile/T20666b.stderr
@@ -0,0 +1,10 @@
+
+T20666b.hs:11:10: warning: [GHC-36038] [-Wloopy-superclass-solve (in -Wdefault)]
+    I am solving the constraint ‘Eq (F [a])’,
+      arising from the superclasses of an instance declaration,
+    in a way that might turn out to loop at runtime.
+    Future versions of GHC will turn this warning into an error.
+    See the user manual, § Undecidable instances and loopy superclasses.
+    Suggested fix:
+      Add the constraint ‘Eq (F [a])’ to the instance context,
+      even though it seems logically implied by other constraints in the context.
diff --git a/testsuite/tests/typecheck/should_compile/T22891.hs b/testsuite/tests/typecheck/should_compile/T22891.hs
new file mode 100644
index 0000000000..ad6b473395
--- /dev/null
+++ b/testsuite/tests/typecheck/should_compile/T22891.hs
@@ -0,0 +1,9 @@
+{-# LANGUAGE UndecidableInstances #-}
+
+module T22891 where
+
+class Foo f
+
+class Foo f => Bar f g
+
+instance Bar f f => Bar f (h k)
diff --git a/testsuite/tests/typecheck/should_compile/T22891.stderr b/testsuite/tests/typecheck/should_compile/T22891.stderr
new file mode 100644
index 0000000000..a0bdc1649b
--- /dev/null
+++ b/testsuite/tests/typecheck/should_compile/T22891.stderr
@@ -0,0 +1,10 @@
+
+T22891.hs:9:10: warning: [GHC-36038] [-Wloopy-superclass-solve (in -Wdefault)]
+    I am solving the constraint ‘Foo f’,
+      arising from the superclasses of an instance declaration,
+    in a way that might turn out to loop at runtime.
+    Future versions of GHC will turn this warning into an error.
+    See the user manual, § Undecidable instances and loopy superclasses.
+    Suggested fix:
+      Add the constraint ‘Foo f’ to the instance context,
+      even though it seems logically implied by other constraints in the context.
diff --git a/testsuite/tests/typecheck/should_compile/T22912.hs b/testsuite/tests/typecheck/should_compile/T22912.hs
new file mode 100644
index 0000000000..994a8022ef
--- /dev/null
+++ b/testsuite/tests/typecheck/should_compile/T22912.hs
@@ -0,0 +1,26 @@
+{-# LANGUAGE MonoLocalBinds          #-}
+{-# LANGUAGE QuantifiedConstraints   #-}
+{-# LANGUAGE ScopedTypeVariables     #-}
+{-# LANGUAGE UndecidableInstances    #-}
+{-# LANGUAGE UndecidableSuperClasses #-}
+
+module T22912 where
+
+
+class    c => Exactly c
+instance c => Exactly c
+class    c => Implies c
+
+data Dict c = c => Dict
+
+anythingDict :: forall c. Dict c
+anythingDict = go
+  where
+    go :: (Exactly (Implies c) => Implies c) => Dict c
+    go = Dict
+
+-- This is clearly wrong: we shouldn't be able to produce evidence
+-- for any constraint whatsoever! However, GHC can be tricked into
+-- producing a bottom dictionary.
+-- This test checks that it emits an appropriate warning when doing so,
+-- to allow users to adapt their code before we plug the hole completely.
diff --git a/testsuite/tests/typecheck/should_compile/T22912.stderr b/testsuite/tests/typecheck/should_compile/T22912.stderr
new file mode 100644
index 0000000000..b53614baff
--- /dev/null
+++ b/testsuite/tests/typecheck/should_compile/T22912.stderr
@@ -0,0 +1,12 @@
+
+T22912.hs:17:16: warning: [GHC-36038] [-Wloopy-superclass-solve (in -Wdefault)]
+    I am solving the constraint ‘Implies c’,
+      arising from the head of a quantified constraint
+      arising from a use of ‘go’,
+    in a way that might turn out to loop at runtime.
+    Future versions of GHC will turn this warning into an error.
+    See the user manual, § Undecidable instances and loopy superclasses.
+    Suggested fix:
+      Add the constraint ‘Implies
+                            c’ to the context of the quantified constraint,
+      even though it seems logically implied by other constraints in the context.
diff --git a/testsuite/tests/typecheck/should_compile/all.T b/testsuite/tests/typecheck/should_compile/all.T
index 07d39cebd3..ff603bbcf3 100644
--- a/testsuite/tests/typecheck/should_compile/all.T
+++ b/testsuite/tests/typecheck/should_compile/all.T
@@ -857,3 +857,6 @@ test('T22647', normal, compile, [''])
 test('T19577', normal, compile, [''])
 test('T22383', normal, compile, [''])
 test('T21501', normal, compile, [''])
+test('T20666b', normal, compile, [''])
+test('T22891', normal, compile, [''])
+test('T22912', normal, compile, [''])
diff --git a/testsuite/tests/typecheck/should_fail/T20666.stderr b/testsuite/tests/typecheck/should_fail/T20666.stderr
index bc2aad5497..46d0fafcc4 100644
--- a/testsuite/tests/typecheck/should_fail/T20666.stderr
+++ b/testsuite/tests/typecheck/should_fail/T20666.stderr
@@ -1,20 +1,20 @@
 
-T20666.hs:13:10: error: [GHC-39999]
-    • Could not deduce ‘Show (T c)’
-        arising from the superclasses of an instance declaration
-      from the context: (D d, c ~ S d)
-        bound by the instance declaration at T20666.hs:13:10-31
-      Possible fix:
-        If the constraint looks soluble from a superclass of the instance context,
-        read 'Undecidable instances and loopy superclasses' in the user manual
-    • In the instance declaration for ‘C1 c’
+T20666.hs:13:10: warning: [GHC-36038] [-Wloopy-superclass-solve (in -Wdefault)]
+    I am solving the constraint ‘Show (T c)’,
+      arising from the superclasses of an instance declaration,
+    in a way that might turn out to loop at runtime.
+    Future versions of GHC will turn this warning into an error.
+    See the user manual, § Undecidable instances and loopy superclasses.
+    Suggested fix:
+      Add the constraint ‘Show (T c)’ to the instance context,
+      even though it seems logically implied by other constraints in the context.
 
-T20666.hs:17:10: error: [GHC-39999]
-    • Could not deduce ‘Show (T c)’
-        arising from the superclasses of an instance declaration
-      from the context: (D d, c ~ S d, c' ~ c)
-        bound by the instance declaration at T20666.hs:17:10-40
-      Possible fix:
-        If the constraint looks soluble from a superclass of the instance context,
-        read 'Undecidable instances and loopy superclasses' in the user manual
-    • In the instance declaration for ‘C2 c'’
+T20666.hs:17:10: warning: [GHC-36038] [-Wloopy-superclass-solve (in -Wdefault)]
+    I am solving the constraint ‘Show (T c)’,
+      arising from the superclasses of an instance declaration,
+    in a way that might turn out to loop at runtime.
+    Future versions of GHC will turn this warning into an error.
+    See the user manual, § Undecidable instances and loopy superclasses.
+    Suggested fix:
+      Add the constraint ‘Show (T c)’ to the instance context,
+      even though it seems logically implied by other constraints in the context.
diff --git a/testsuite/tests/typecheck/should_fail/T20666a.stderr b/testsuite/tests/typecheck/should_fail/T20666a.stderr
index 4192b88807..bed585dcd2 100644
--- a/testsuite/tests/typecheck/should_fail/T20666a.stderr
+++ b/testsuite/tests/typecheck/should_fail/T20666a.stderr
@@ -1,10 +1,10 @@
 
-T20666a.hs:11:10: error: [GHC-39999]
-    • Could not deduce ‘Eq (F [a])’
-        arising from the superclasses of an instance declaration
-      from the context: D [a]
-        bound by the instance declaration at T20666a.hs:11:10-23
-      Possible fix:
-        If the constraint looks soluble from a superclass of the instance context,
-        read 'Undecidable instances and loopy superclasses' in the user manual
-    • In the instance declaration for ‘C [a]’
+T20666a.hs:11:10: warning: [GHC-36038] [-Wloopy-superclass-solve (in -Wdefault)]
+    I am solving the constraint ‘Eq (F [a])’,
+      arising from the superclasses of an instance declaration,
+    in a way that might turn out to loop at runtime.
+    Future versions of GHC will turn this warning into an error.
+    See the user manual, § Undecidable instances and loopy superclasses.
+    Suggested fix:
+      Add the constraint ‘Eq (F [a])’ to the instance context,
+      even though it seems logically implied by other constraints in the context.
diff --git a/testsuite/tests/typecheck/should_fail/T6161.stderr b/testsuite/tests/typecheck/should_fail/T6161.stderr
index 71c7455abd..c1daab09df 100644
--- a/testsuite/tests/typecheck/should_fail/T6161.stderr
+++ b/testsuite/tests/typecheck/should_fail/T6161.stderr
@@ -1,7 +1,10 @@
 
-T6161.hs:19:10: error: [GHC-39999]
-    • Could not deduce ‘Super (Fam a)’
-        arising from the superclasses of an instance declaration
-      from the context: Foo a
-        bound by the instance declaration at T6161.hs:19:10-31
-    • In the instance declaration for ‘Duper (Fam a)’
+T6161.hs:19:10: warning: [GHC-36038] [-Wloopy-superclass-solve (in -Wdefault)]
+    I am solving the constraint ‘Super (Fam a)’,
+      arising from the superclasses of an instance declaration,
+    in a way that might turn out to loop at runtime.
+    Future versions of GHC will turn this warning into an error.
+    See the user manual, § Undecidable instances and loopy superclasses.
+    Suggested fix:
+      Add the constraint ‘Super (Fam a)’ to the instance context,
+      even though it seems logically implied by other constraints in the context.
diff --git a/testsuite/tests/typecheck/should_fail/all.T b/testsuite/tests/typecheck/should_fail/all.T
index a7e9f6a678..f8a3db350c 100644
--- a/testsuite/tests/typecheck/should_fail/all.T
+++ b/testsuite/tests/typecheck/should_fail/all.T
@@ -241,7 +241,7 @@ test('tcfail215', normal, compile_fail, [''])
 test('tcfail216', normal, compile_fail, [''])
 test('tcfail217', normal, compile_fail, [''])
 test('tcfail218', normal, compile_fail, [''])
-test('tcfail223', normal, compile_fail, [''])
+test('tcfail223', normal, compile, ['']) # To become compile_fail after migration period (see #22912)
 test('tcfail224', normal, compile_fail, [''])
 test('tcfail225', normal, compile_fail, [''])
 
@@ -294,7 +294,7 @@ test('T19187a', normal, compile_fail, [''])
 test('T2534', normal, compile_fail, [''])
 test('T7175', normal, compile_fail, [''])
 test('T7210', normal, compile_fail, [''])
-test('T6161', normal, compile_fail, [''])
+test('T6161', normal, compile, [''])       # To become compile_fail after migration period (see #22912)
 test('T7368', normal, compile_fail, [''])
 test('T7264', normal, compile_fail, [''])
 test('T6069', normal, compile_fail, [''])
@@ -665,5 +665,5 @@ test('T21530a', normal, compile_fail, [''])
 test('T21530b', normal, compile_fail, [''])
 test('T22570', normal, compile_fail, [''])
 test('T22645', normal, compile_fail, [''])
-test('T20666', normal, compile_fail, [''])
-test('T20666a', normal, compile_fail, [''])
+test('T20666', normal, compile, [''])   # To become compile_fail after migration period (see #22912)
+test('T20666a', normal, compile, [''])  # To become compile_fail after migration period (see #22912)
diff --git a/testsuite/tests/typecheck/should_fail/tcfail223.stderr b/testsuite/tests/typecheck/should_fail/tcfail223.stderr
index d3173adde7..8201c8839f 100644
--- a/testsuite/tests/typecheck/should_fail/tcfail223.stderr
+++ b/testsuite/tests/typecheck/should_fail/tcfail223.stderr
@@ -1,9 +1,10 @@
 
-tcfail223.hs:10:10: error: [GHC-39999]
-    • Could not deduce ‘Class1 a’
-        arising from the superclasses of an instance declaration
-      from the context: Class3 a
-        bound by the instance declaration at tcfail223.hs:10:10-29
-      Possible fix:
-        add (Class1 a) to the context of the instance declaration
-    • In the instance declaration for ‘Class2 a’
+tcfail223.hs:10:10: warning: [GHC-36038] [-Wloopy-superclass-solve (in -Wdefault)]
+    I am solving the constraint ‘Class1 a’,
+      arising from the superclasses of an instance declaration,
+    in a way that might turn out to loop at runtime.
+    Future versions of GHC will turn this warning into an error.
+    See the user manual, § Undecidable instances and loopy superclasses.
+    Suggested fix:
+      Add the constraint ‘Class1 a’ to the instance context,
+      even though it seems logically implied by other constraints in the context.