diff options
author | Adriaan Moors <adriaan.moors@typesafe.com> | 2014-01-28 16:43:40 -0800 |
---|---|---|
committer | Adriaan Moors <adriaan.moors@typesafe.com> | 2014-02-19 10:12:19 -0800 |
commit | c001b888b896989a2c0afa0c24d038502970151c (patch) | |
tree | b04b54e08a8ad8f95db4e14d6509a044146ef1c5 /src/compiler/scala/tools/nsc/transform/patmat/MatchAnalysis.scala | |
parent | 3dbcb1b9d4daa5cba98747bbc66f898ba0f864fd (diff) | |
download | scala-c001b888b896989a2c0afa0c24d038502970151c.tar.gz scala-c001b888b896989a2c0afa0c24d038502970151c.tar.bz2 scala-c001b888b896989a2c0afa0c24d038502970151c.zip |
SI-1503 don't assume unsound type for ident/literal patterns
The fix only kicks in under -Xfuture. We also warn under -Xlint.
What type should a variable bound to the value matched by a pattern have?
To avoid CCEs, it should be a type that's implied by the matching
semantics of the pattern.
Usually, the type implied by a pattern matching a certain value
is the pattern's type, because pattern matching implies instance-of checks.
However, Stable Identifier and Literal patterns are matched using `==`,
which does not imply a type for the binder that binds the matched value.
The change in type checking due to this fix is that programs that used to crash with a CCE
(because we blindly cast to the type of the pattern, which a `==` check does not imply)
now get a weaker type instead (and no cast). They may still type check, or they may not.
To compensate for this fix, change `case x@Foo => x` to `case x: Foo.type => x`,
if it's important that `x` have type `Foo.type`.
See also:
- SI-4577: matching of singleton type patterns uses `eq`,
not `==` (so that the types are not a lie).
- SI-5024: patmat strips unused bindings, but affects semantics
Diffstat (limited to 'src/compiler/scala/tools/nsc/transform/patmat/MatchAnalysis.scala')
-rw-r--r-- | src/compiler/scala/tools/nsc/transform/patmat/MatchAnalysis.scala | 50 |
1 files changed, 50 insertions, 0 deletions
diff --git a/src/compiler/scala/tools/nsc/transform/patmat/MatchAnalysis.scala b/src/compiler/scala/tools/nsc/transform/patmat/MatchAnalysis.scala index 2a3c631a66..cb15be79c0 100644 --- a/src/compiler/scala/tools/nsc/transform/patmat/MatchAnalysis.scala +++ b/src/compiler/scala/tools/nsc/transform/patmat/MatchAnalysis.scala @@ -16,6 +16,56 @@ trait TreeAndTypeAnalysis extends Debugging { import definitions._ import analyzer.Typer + /** Compute the type T implied for a value `v` matched by a pattern `pat` (with expected type `pt`). + * + * Usually, this is the pattern's type because pattern matching implies instance-of checks. + * + * However, Stable Identifier and Literal patterns are matched using `==`, + * which does not imply a type for the binder that binds the matched value. + * + * See SI-1503, SI-5024: don't cast binders to types we're not sure they have + * + * TODO: update spec as follows (deviation between `**`): + * + * A pattern binder x@p consists of a pattern variable x and a pattern p. + * The type of the variable x is the static type T **IMPLIED BY** the pattern p. + * This pattern matches any value v matched by the pattern p + * **Deleted: , provided the run-time type of v is also an instance of T, ** + * and it binds the variable name to that value. + * + * Addition: + * A pattern `p` _implies_ a type `T` if the pattern matches only values of the type `T`. + */ + def binderTypeImpliedByPattern(pat: Tree, pt: Type, binder: Symbol): Type = + pat match { + // because `==` decides whether these patterns match, stable identifier patterns (ident or selection) + // do not contribute any type information (beyond the pattern's expected type) + // e.g., in case x@Nil => x --> all we know about `x` is that it satisfies Nil == x, which could be anything + case Ident(_) | Select(_, _) => + if (settings.future) pt + else { + // TODO: don't warn unless this unsound assumption is actually used in a cast + // I tried annotating the type returned here with an internal annotation (`pat.tpe withAnnotation UnsoundAssumptionAnnotation`), + // and catching it in the patmat backend when used in a cast (because that would signal the unsound assumption was used), + // but the annotation didn't bubble up... + // This is a pretty poor approximation. + def unsoundAssumptionUsed = binder.name != nme.WILDCARD && !(pt <:< pat.tpe) + if (settings.lint && unsoundAssumptionUsed) + global.currentUnit.warning(pat.pos, + sm"""The value matched by $pat is bound to ${binder.name}, which may be used under the + |unsound assumption that it has type ${pat.tpe}, whereas we can only safely + |count on it having type $pt, as the pattern is matched using `==` (see SI-1503).""") + + pat.tpe + } + + + // the other patterns imply type tests, so we can safely assume the binder has the pattern's type when the pattern matches + // concretely, a literal, type pattern, a case class (the constructor's result type) or extractor (the unapply's argument type) all imply type tests + // (and, inductively, an alternative) + case _ => pat.tpe + } + // we use subtyping as a model for implication between instanceof tests // i.e., when S <:< T we assume x.isInstanceOf[S] implies x.isInstanceOf[T] // unfortunately this is not true in general: |