Flatten lifts out anonymous module vals without making sure that lifted names do not conflict

[DarkDimius]

Shown by ./tests/pos/zoo.scala

checking ./tests/pos/zoo.scala after phase TreeTransform:{LazyVals, seqLiterals, intercepted, literalize, getters, elimByName, resolveSuper}
checking ./tests/pos/zoo.scala after phase TreeTransform:{mixin, memoize, capturedVars, constructors}
checking ./tests/pos/zoo.scala after phase TreeTransform:{lambdaLift, flatten, restoreScopes}
[error] module val defined twice module $anon 9922 9917
[error] module val defined twice module $anon 9927 9922
[error] module val defined twice module $anon 9932 9927

[odersky]

This should be solved by #420 @DarkDimius can you check and close the issue if that's the case?

[DarkDimius]

After applying 7b6964a the problem persists.

[odersky]

After some investigation, it looks like the test checks the wrong condition. Module vals are not lifted out; they remain in the scope where they were defined first. So there's no reason why they should have unique names.

[DarkDimius]

The check suggests that they start to clash only after flatten, that is, before flatten the names were unique. If they are not lifted out, what made them clash?

[odersky]

This: (in TreeChecker)

• if* (this.flatClasses || !(sym.isAnonymousFunction ||
  sym.isAnonymousClass || sym.isAnonymousModuleVal))

printError(s"$typ defined twice $sym ${sym.id} ${registry(name).id}"

)

On Sun, Mar 22, 2015 at 7:00 PM, Dmitry Petrashko [email protected]
wrote:

The check suggests that they start to clash with names only after flatten,
that is, before flatten the names were unique. If they are not lifted out,
what made them clash?

—
Reply to this email directly or view it on GitHub
#415 (comment).

Martin Odersky
EPFL

[DarkDimius]

ahh, I get it.
I'll remove this check.
Thanks for investigation.