Report error on dynamic metaclass

[gvanrossum]

Previously the fast parser would crash ("assert False" in stringify_name()). Now it forces an error.

I didn't see a decent way to report an error from fastparser.py, so I set the metaclass name to "<error>" and check for that in semanal.py which will report a decent error.

Note that the old parser still reports a syntax error on the '(', I think that's fine. Also there doesn't seem to be a corresponding issue with Python 2.

Fixes #2273.

[gvanrossum]

@JukkaL would there be a better way to report this as an error?

[JukkaL]

Raise an exception and catch it at the top-level parse function? We already have TypeCommentParseError, and if there's a parse error, we give up and don't return the AST. This error is probably very rare, so losing errors from the following passes seems like a non-issue.

[gvanrossum]

I'm not so sure that it's okay to treat this as a parse error. This is existing non-annotation code and people may have a good reason for it. We should produce an error explaining that we can't process the metaclass, and ignore it. So I don't think an exception would be better than what I have.

[JukkaL]

Yeah, that's a good point. A better fix would be to make ClassDef.metaclass a node instead of a string.

[gvanrossum]

I thought I had responded here but don't see my response.

I tried making it a node but the refactor was too complex given the simple problem. Can we just merge this? It's a crash bug (with the new parser).

[JukkaL]

Agreed, this isn't worth a big refactor.

[gvanrossum]

I'm sorry, but that's quite a big refactoring. Can we just go ahead with what I have now? If it ever breaks due to some future changes it should be quite easy to track down the breakage.

[JukkaL]

Were your comments delayed by several days? Odd.

[gvanrossum]

Yeah, looks I replied by email and the email only showed up an hour ago. Weird.