Cyclic Reference errors for recursive toplevel match types #9000
Comments
odersky
added a commit
to dotty-staging/dotty
that referenced
this issue
May 18, 2020
I still don't know why the match type ``` type A2B[Xs <: Tuple] <: Tuple = Xs match case Unit => Unit case a *: as => B *: A2B[as] ``` compiled OK if it appeared in a user defined object, but failed with a cyclic reference when written on the top level. But the logic for cyclic detection was clearly very fragile, and unadapted to the case at hand. Essentially, it's a type map that inserts LazyRefs when it detects legal cycles. The problem is that these insertions cause recomputations of applied types in the TypeMap. And these fail since the type constructor (A2B in this case) has not yet been completed. So after a LazyRef was inserted, the cycle checker fails again in an unrecoverable way because it tries to get the type parameters of an uncompleted type constructor. So the question would be rather - why did this work if the match type appears in a user-defined object? I could not answer that question, but I could fix the logic to handle these cases.
odersky
added a commit
to dotty-staging/dotty
that referenced
this issue
May 19, 2020
The only difference between checking cyclicity of a match type in an object O and on the toplevel is that the matchtype itself has the prefix O.this in the first case, and TermRef(_, pkgObject) in the second case (this is because the match type is actually looked up in the package scope, not in the package object scope). This made a difference in the isTrivial check when computing type parameters. And this in turn caused spurious cycles to be reported. The fix is to add the missing case to isTrivial.
odersky
added a commit
to dotty-staging/dotty
that referenced
this issue
May 19, 2020
The only difference between checking cyclicity of a match type in an object O and on the toplevel is that the matchtype itself has the prefix O.this in the first case, and TermRef(_, pkgObject) in the second case (this is because the match type is actually looked up in the package scope, not in the package object scope). This made a difference in the isTrivial check when computing type parameters. And this in turn caused spurious cycles to be reported. The fix is to add the missing case to isTrivial.
smarter
added a commit
that referenced
this issue
May 19, 2020
…yclic Fix #9000: Avoid spurious cyclic errors for toplevel matches
rethab
pushed a commit
to rethab/dotty
that referenced
this issue
May 25, 2020
The only difference between checking cyclicity of a match type in an object O and on the toplevel is that the matchtype itself has the prefix O.this in the first case, and TermRef(_, pkgObject) in the second case (this is because the match type is actually looked up in the package scope, not in the package object scope). This made a difference in the isTrivial check when computing type parameters. And this in turn caused spurious cycles to be reported. The fix is to add the missing case to isTrivial.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Minimized example
Output
Error: cyclic reference
Expectation
Should compile. The type
A2Bcompiles OK if it is put inside an objecttest.The text was updated successfully, but these errors were encountered: