[commit: ghc] master: Fix a bug in ApplicativeDo (#11612) (3259bf6)
git at git.haskell.org
git at git.haskell.org
Sat Feb 20 09:51:41 UTC 2016
Repository : ssh://git@git.haskell.org/ghc
On branch : master
Link : http://ghc.haskell.org/trac/ghc/changeset/3259bf658662e7052ae91de2fa27baae8c84b7fa/ghc
>---------------------------------------------------------------
commit 3259bf658662e7052ae91de2fa27baae8c84b7fa
Author: Simon Marlow <marlowsd at gmail.com>
Date: Sat Feb 20 07:23:37 2016 +0000
Fix a bug in ApplicativeDo (#11612)
In some cases ApplicativeDo would miss some opportunities, due to a
wrong calculation of free variables in RnExpr.segments.
>---------------------------------------------------------------
3259bf658662e7052ae91de2fa27baae8c84b7fa
compiler/rename/RnExpr.hs | 27 +++++++++++++++++++++------
testsuite/tests/ado/ado001.hs | 12 ++++++++++++
testsuite/tests/ado/ado001.stdout | 1 +
3 files changed, 34 insertions(+), 6 deletions(-)
diff --git a/compiler/rename/RnExpr.hs b/compiler/rename/RnExpr.hs
index 616f259..9d1200a 100644
--- a/compiler/rename/RnExpr.hs
+++ b/compiler/rename/RnExpr.hs
@@ -1549,24 +1549,36 @@ segments stmts = map fst $ merge $ reverse $ map reverse $ walk (reverse stmts)
_otherwise -> (seg,all_lets) : rest
where
rest = merge segs
- all_lets = all (not . isBindStmt . fst) seg
+ all_lets = all (isLetStmt . fst) seg
+ -- walk splits the statement sequence into segments, traversing
+ -- the sequence from the back to the front, and keeping track of
+ -- the set of free variables of the current segment. Whenever
+ -- this set of free variables is empty, we have a complete segment.
+ walk :: [(ExprLStmt Name, FreeVars)] -> [[(ExprLStmt Name, FreeVars)]]
walk [] = []
walk ((stmt,fvs) : stmts) = ((stmt,fvs) : seg) : walk rest
- where (seg,rest) = chunter (fvs `intersectNameSet` allvars) stmts
+ where (seg,rest) = chunter fvs' stmts
+ (_, fvs') = stmtRefs stmt fvs
chunter _ [] = ([], [])
chunter vars ((stmt,fvs) : rest)
| not (isEmptyNameSet vars)
= ((stmt,fvs) : chunk, rest')
where (chunk,rest') = chunter vars' rest
- evars = fvs `intersectNameSet` allvars
- pvars = mkNameSet (collectStmtBinders (unLoc stmt))
+ (pvars, evars) = stmtRefs stmt fvs
vars' = (vars `minusNameSet` pvars) `unionNameSet` evars
chunter _ rest = ([], rest)
- isBindStmt (L _ BindStmt{}) = True
- isBindStmt _ = False
+ stmtRefs stmt fvs
+ | isLetStmt stmt = (pvars, fvs' `minusNameSet` pvars)
+ | otherwise = (pvars, fvs')
+ where fvs' = fvs `intersectNameSet` allvars
+ pvars = mkNameSet (collectStmtBinders (unLoc stmt))
+
+isLetStmt :: LStmt a b -> Bool
+isLetStmt (L _ LetStmt{}) = True
+isLetStmt _ = False
-- | Find a "good" place to insert a bind in an indivisible segment.
-- This is the only place where we use heuristics. The current
@@ -1576,6 +1588,9 @@ splitSegment
:: [(ExprLStmt Name, FreeVars)]
-> ( [(ExprLStmt Name, FreeVars)]
, [(ExprLStmt Name, FreeVars)] )
+splitSegment [one,two] = ([one],[two])
+ -- there is no choice when there are only two statements; this just saves
+ -- some work in a common case.
splitSegment stmts
| Just (lets,binds,rest) <- slurpIndependentStmts stmts
= if not (null lets)
diff --git a/testsuite/tests/ado/ado001.hs b/testsuite/tests/ado/ado001.hs
index 9f8f8da..e452cdd 100644
--- a/testsuite/tests/ado/ado001.hs
+++ b/testsuite/tests/ado/ado001.hs
@@ -109,6 +109,17 @@ test10 = do
x5 <- e
return (const () (x3,x4,x5))
+-- (a | b)
+-- This demonstrated a bug in RnExpr.segments (#11612)
+test11 :: M ()
+test11 = do
+ x1 <- a
+ let x2 = x1
+ x3 <- b
+ let x4 = c
+ x5 = x4
+ return (const () (x1,x2,x3,x4))
+
main = mapM_ run
[ test1
, test2
@@ -120,6 +131,7 @@ main = mapM_ run
, test8
, test9
, test10
+ , test11
]
-- Testing code, prints out the structure of a monad/applicative expression
diff --git a/testsuite/tests/ado/ado001.stdout b/testsuite/tests/ado/ado001.stdout
index 93e300c..f7c48ca 100644
--- a/testsuite/tests/ado/ado001.stdout
+++ b/testsuite/tests/ado/ado001.stdout
@@ -8,3 +8,4 @@ a; (b | (c; (d; (e | (f; g)))))
a; ((b | c) | d)
((a | (b; c)) | d) | e
((a | b); (c | d)) | e
+a | b
More information about the ghc-commits
mailing list