[Git][ghc/ghc][wip/torsten.schmits/23612] Avoid substituting constructors for breakpoint FVs

[Torsten Schmits (@torsten.schmits)]

Torsten Schmits pushed to branch wip/torsten.schmits/23612 at Glasgow Haskell Compiler / GHC


Commits:
618372ce by Torsten Schmits at 2023-08-03T12:34:08+02:00
Avoid substituting constructors for breakpoint FVs

Fixes #23612

MR: !11026

The fingerprinting logic in Iface.Recomp gets the constructor's tycon as
an additional element of a decl group's dependency graph, although it's
not in its local OccEnv.

It's probably also generally nonsense to keep a constructor in the
inspectable breakpoint vars.

-------------------------
Metric Decrease:
    T12234
    T13035
-------------------------

- - - - -


4 changed files:

- compiler/GHC/Core/Subst.hs
- + testsuite/tests/codeGen/should_run/T23612.hs
- + testsuite/tests/codeGen/should_run/T23612.script
- testsuite/tests/codeGen/should_run/all.T


Changes:

=====================================
compiler/GHC/Core/Subst.hs
=====================================
@@ -592,9 +592,22 @@ substDVarSet subst@(Subst _ _ tv_env cv_env) fvs
 ------------------
 substTickish :: Subst -> CoreTickish -> CoreTickish
 substTickish subst (Breakpoint ext n ids modl)
-   = Breakpoint ext n (mapMaybe do_one ids) modl
+   = Breakpoint ext n (filter not_datacon (mapMaybe do_one ids)) modl
  where
     do_one = getIdFromTrivialExpr_maybe . lookupIdSubst subst
+
+    -- If a variable is substituted with a constructor, there's no point in
+    -- inspecting it anymore, and it would cause problems with the dependency
+    -- computation when fingerprinting iface decls, since it pulls in the name
+    -- of the tycon even when it's not in the decl's OccEnv.
+    not_datacon s
+      | isId s
+      = case idDetails s of
+        DataConWorkId {} -> False
+        DataConWrapId {} -> False
+        _ -> True
+      | otherwise = True
+
 substTickish _subst other = other
 
 {- Note [Substitute lazily]


=====================================
testsuite/tests/codeGen/should_run/T23612.hs
=====================================
@@ -0,0 +1,11 @@
+module Main where
+
+data X = A1
+
+foo :: X -> ()
+foo A1 = ()
+
+bar :: X -> ()
+bar y = foo y
+
+main = seq (map bar [A1]) (return ())


=====================================
testsuite/tests/codeGen/should_run/T23612.script
=====================================
@@ -0,0 +1 @@
+:load T23612


=====================================
testsuite/tests/codeGen/should_run/all.T
=====================================
@@ -237,3 +237,5 @@ test('MulMayOflo_full',
         ignore_stdout],
      multi_compile_and_run,
      ['MulMayOflo', [('MulMayOflo_full.cmm', '')], ''])
+
+test('T23612', only_ways(['ghci-opt']), ghci_script, ['T23612.script'])



View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/618372ce380ea89c744e7ae2529eadeb52cc5ef2

-- 
View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/618372ce380ea89c744e7ae2529eadeb52cc5ef2
You're receiving this email because of your account on gitlab.haskell.org.


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.haskell.org/pipermail/ghc-commits/attachments/20230803/e105bde9/attachment-0001.html>