Rename allow_redefinition to allow_redefinition_old

[chdemko]

Fix #372

[cgoldberg]

This change would require pinning Mypy in refurb to >=1.20.0 or else it will break environments that have an older version installed.

Maybe:

try:
    opt.allow_redefinition_old = True
except AttributeError:
    opt.allow_redefinition = True

[dosisod]

@cgoldberg your try/catch method would be preferable, as lots of people use older versions of MyPy.

In regards to Refurb's use of MyPy's internal classes, see python/mypy#16497 (comment):

There is also the bigger question of how 3rd parties should safely use Mypy internals (or if they even should in the first place). Currently Mypy internals are not versioned and can change with any release. In addition, certain parts of Mypy are hard/impossible to use outside of Mypy itself, whether that's because they don't work, crash, or require lots of moving parts because they weren't meant to be used in a stand-alone environment. I know that standardizing/stabilizing Mypy's internals so that 3rd parties can use them is probably not a major priority, but I thought I would bring it up to gauge how you all feel about it. I could elaborate more but I want to keep this short. I can open a separate issue for this if need be.

MyPy does not -- nor should they, necessarily -- use semantic versioning for their internal APIs, because they aren't meant to be used directly by 3rd party libraries. This means that tools like Refurb will have to be on the ball with new releases. There should probably be a cronjob that runs CI against the latest MyPy commit, that way these sorts of issues are detected before a MyPy release is published.

[cgoldberg]

There should probably be a cronjob that runs CI against the latest MyPy commit, that way these sorts of issues are detected before a MyPy release is published.

I added that in #374

[chdemko]

Do I change my PR to a try ... except ?

[cgoldberg]

Do I change my PR to a try ... except ?

yes, or else it will break on older versions of Mypy

[cgoldberg]

@chdemko this looks like a nicer option, and should work for old and new versions: #372 (comment)

[cgoldberg]

@chdemko if this works, it's probably better than setting the attribute directly on opt: #372 (comment)

[dosisod]

Sorry @chdemko , I went down the --allow-redefinitions route to avoid the headache of using the internal class directly.