For sequence __contains__ and other scenarios, identity is checked before equality, which I've heard is so that "for x in y: assert x in y" doesn't ever fail with an AssertionError (even with NaN and so on). This is not the case for collections.abc-based sequences, which is a jarring inconsistency.

See Raymond Hettinger's comments in http://bugs.python.org/issue4296 for details on why the usual sequence behavior is deliberate.

While this hasn't proven to be an issue to date (non-reflexive objects don't arise much in practice), the proposed change would make it easier to use the Sequence ABC to create classes that are interoperable with other sequences and that have the clean invariant, "anything added to the sequence will be "in" the sequence".

That said, I don't know if this is really needed.

Thanks for the patch! I was secretly hoping to write it though. :P

Not sure how the code review tool works. The patch looks good to me, except it needs tests IMO. I can write them if you think this is not worth the effort.

Devin, you're welcome to whip-up a patch for the tests for this one.  I don't know if it will go forward though.  I'm leaving it open for a while to see if anyone has objections.

If you want to work another straight-forward ABC patch, have a look at http://bugs.python.org/issue23086

Since both patches will clobber each other (same test method), complicating the review process, I've written the patch for issue23086 first. It's currently waiting on committee review from my employer. Will upload to other issue, and once it's submitted or rejected, I'll add a test patch to this issue too.

Can you submit a contributor agreement as well?

I think that such a thing is meaningless, as I don't own copyright to the patches, my employer (Google) does. But, it can't hurt, so, done.