Allow a custom type comment prefix.

[matthiaskramm]

Instead of hardcoding it, we expect callers to explicitly do something like

ast27.register_type_comment_prefix("# type: ")

before parsing.

This is the typed_ast counterpiece to python/mypy#2030.

[ddfisher]

Thanks for the PR! Before getting down to a finer-grained review, I have a few big picture comments:

• This library is intended to parse PEP 484-compliant typed code, so it should always parse # type: comments without any additional configuration.
• I'd like for this library to be able to parse comments with arbitrary prefixes (when configured to do so) but I don't think they should be parsed as type comments, because -- per PEP 484 -- they aren't (and because this would severely limit what you can do with them). Instead, I'd like to see a new ast node in the style of TYPE_IGNORE (i.e. stored at a file level by line number) with the prefix and contents of the comment as fields.

[matthiaskramm]

So are you saying that you want two kinds of nodes for comments, one for # type: ignore and one for free-form comments?

Or can we map the former to the latter, and supply the default configuration?

[ddfisher]

Yes. I think that # type: ignores and freeform comments are quite different beasts, and should therefore be different nodes.

[matthiaskramm]

It feels odd to me to make # type: ignore and # mypy: ignore different nodes. Maybe the "ignore" comments should have a configurable prefix, and then on top of that we have a pattern for freeform comments?

[ddfisher]

It should turn out simpler overall to just have a pattern for freeform comments. My understanding is that the PEP doesn't discuss custom type prefixes (type ignore or otherwise), so I'm reluctant to add those explicitly to typed_ast.

[ddfisher]

This is still likely not the approach we want. Closing due to inactivity.