| [#344] | project: compiler | priority: medium | category: bug | |
|---|---|---|---|---|
| submitter | assigned to | status | date submitted | |
| Vincent | Philippe | fixed | 2004-06-11 12:51:55.0 | |
| subject | nested ifs type inference | |||
| code |
object Bug {
class A;
case class A1 extends A;
case class A2 extends A;
def f: A =
if (true)
A1()
else {
val a = if (true) A1() else A2();
a
};
}
|
|||
| what happened | Compiler crash during ExplicitOuterClassesPhase. |
|||
| what expected | The program should compile normally. It 's what happen in the following cases :
|
|||
| [back to overview] | ||||
| Philippe edited on 2004-06-11 17:54:14.0 |
|
Here is a version without case classes.
class A;
trait I;
class A1 extends A with I;
class A2 extends A with I;
object Bug {
if (true)
new A1()
else {
val a = if (true) new A1() else new A2();
a
};
}
The problem comes from the inferred type for "val a" which is "A with I". This is a compound type. Currently compound types have a symbol whose owner is often uncorrect. That's probably the cause of the crash. Note that the fact that the inferred type is "A with I" instead of just A seems to be a bug. Here is an example where the inferred type is a legitimate compound type.
trait A;
trait B;
class AB1 extends A with B;
class AB2 extends A with B;
object Bug {
if (true)
new AB1()
else {
val a = if (true) new AB1() else new AB2();
a
};
}
|
| Philippe edited on 2004-06-11 18:01:57.0 |
| The remark about the uncorrect compound type is incorrect. The compound types are legitimate in all examples. |
| Lex edited on 2006-03-28 14:16:01.0 |