diff options
| author | Simon Peyton Jones <simonpj@microsoft.com> | 2015-12-31 09:07:33 +0000 |
|---|---|---|
| committer | Simon Peyton Jones <simonpj@microsoft.com> | 2015-12-31 14:43:59 +0000 |
| commit | 0579fe99b933384172d19beb6a00dc8a1238101a (patch) | |
| tree | 273ecea533838ccbe61702f5b7a393d830a15fee /compiler/simplCore | |
| parent | 75851bf930067ae7c57bee3c6feea456534eafed (diff) | |
| download | haskell-0579fe99b933384172d19beb6a00dc8a1238101a.tar.gz | |
Improve exprIsBottom
This fixes Trac #11290, by being sligthtly cleverer about finding
what expressions are bottom. Actually this might have minor
other side benefits.
Diffstat (limited to 'compiler/simplCore')
| -rw-r--r-- | compiler/simplCore/SetLevels.hs | 13 |
1 files changed, 12 insertions, 1 deletions
diff --git a/compiler/simplCore/SetLevels.hs b/compiler/simplCore/SetLevels.hs index b742a291fc..b84d67b496 100644 --- a/compiler/simplCore/SetLevels.hs +++ b/compiler/simplCore/SetLevels.hs @@ -542,7 +542,7 @@ See Maessen's paper 1999 "Bottom extraction: factoring error handling out of functional programs" (unpublished I think). When we do this, we set the strictness and arity of the new bottoming -Id, *immediately*, for two reasons: +Id, *immediately*, for three reasons: * To prevent the abstracted thing being immediately inlined back in again via preInlineUnconditionally. The latter has a test for bottoming Ids @@ -551,6 +551,17 @@ Id, *immediately*, for two reasons: * So that it's properly exposed as such in the interface file, even if this is all happening after strictness analysis. + * In case we do CSE with the same expression that *is* marked bottom + lvl = error "urk" + x{str=bot) = error "urk" + Here we don't want to replace 'x' with 'lvl', else we may get Lint + errors, e.g. via a case with empty alternatives: (case x of {}) + Lint complains unless the scrutinee of such a case is clearly bottom. + + This was reported in Trac #11290. But since the whole bottoming-float + thing is based on the cheap-and-cheerful exprIsBottom, I'm not sure + that it'll nail all such cases. + Note [Bottoming floats: eta expansion] c.f Note [Bottoming floats] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Tiresomely, though, the simplifier has an invariant that the manifest |