diff options
author | Paolo Giarrusso <p.giarrusso@gmail.com> | 2012-09-05 02:53:26 +0200 |
---|---|---|
committer | Paolo Giarrusso <p.giarrusso@gmail.com> | 2012-09-12 19:11:52 +0200 |
commit | 9568dc9d1a8e69812f4ec0b72a11a38edaa51f2a (patch) | |
tree | 847ef29de2566c2ab892bc07cf96deda4d06990e /src | |
parent | 5415272018114bb2e15036c5d6f9ae9c5af625d2 (diff) | |
download | scala-9568dc9d1a8e69812f4ec0b72a11a38edaa51f2a.tar.gz scala-9568dc9d1a8e69812f4ec0b72a11a38edaa51f2a.tar.bz2 scala-9568dc9d1a8e69812f4ec0b72a11a38edaa51f2a.zip |
SI-6306 Remove incorrect eta-expansion optimization in Uncurry
Fix SI-6306 by removing the offending code. Moreover, clarify comment
and add testcase.
This pattern match only matches on testcases triggering SI-6306; the
transformation it performs in that case is unsafe.
The intended optimization is to undo eta-expansion of nullary functions,
that is, transform `() => foo()` to `foo`. But that's only valid when
`foo` is an instance of `Function0`, so the optimization is unsafe.
Moreover, the pattern match will fail because at the end of typer that
code has become `() => foo.apply()`, and `isExprSafeToInline(foo.apply)`
always (correctly) fails the isExprSafeToInline test. The pattern match
should thus be different - this code was dead even when it was
introduced (45bcd02f6ba099277bedbf83ec2bda07435c7797), since it was not
invoked either when building the compiler or when compiling function
attempt2() in the included testcase.
Thanks to all commenters on SI-6306 and
https://github.com/scala/scala/pull/1255, in particular to Jason Zaugg
for diagnosing the underlying fault and Lukas Rytz for understanding the
goal of the code.
Diffstat (limited to 'src')
-rw-r--r-- | src/compiler/scala/tools/nsc/transform/UnCurry.scala | 5 | ||||
-rw-r--r-- | src/reflect/scala/reflect/internal/TreeInfo.scala | 2 |
2 files changed, 1 insertions, 6 deletions
diff --git a/src/compiler/scala/tools/nsc/transform/UnCurry.scala b/src/compiler/scala/tools/nsc/transform/UnCurry.scala index 181463657b..f961685dd4 100644 --- a/src/compiler/scala/tools/nsc/transform/UnCurry.scala +++ b/src/compiler/scala/tools/nsc/transform/UnCurry.scala @@ -212,11 +212,6 @@ abstract class UnCurry extends InfoTransform /** Undo eta expansion for parameterless and nullary methods */ def deEta(fun: Function): Tree = fun match { - case Function(List(), Apply(expr, List())) if treeInfo.isExprSafeToInline(expr) => - if (expr hasSymbolWhich (_.isLazy)) - fun - else - expr case Function(List(), expr) if isByNameRef(expr) => noApply += expr expr diff --git a/src/reflect/scala/reflect/internal/TreeInfo.scala b/src/reflect/scala/reflect/internal/TreeInfo.scala index 3a930a195b..6ef4c3f660 100644 --- a/src/reflect/scala/reflect/internal/TreeInfo.scala +++ b/src/reflect/scala/reflect/internal/TreeInfo.scala @@ -67,7 +67,7 @@ abstract class TreeInfo { /** Is tree an expression which can be inlined without affecting program semantics? * - * Note that this is not called "isExprSafeToInline" since purity (lack of side-effects) + * Note that this is not called "isExprPure" since purity (lack of side-effects) * is not the litmus test. References to modules and lazy vals are side-effecting, * both because side-effecting code may be executed and because the first reference * takes a different code path than all to follow; but they are safe to inline |