@encukou @corona10 could you take a look? Thanks!

Thank you!
My main worry here is that we might get into a situation like sys.version/sys.version_info, where the earlier “stringly typed” attribute has a better name.
Are we likely to need something like backend.name & backend.version in the future? If we do, will it be OK to keep backend and a new backend_version?
Or am I overthinking this?

Could you add a test to assert backend is either "readline" or "editline"?

Thank you! My main worry here is that we might get into a situation like sys.version/sys.version_info, where the earlier “stringly typed” attribute has a better name. Are we likely to need something like backend.name & backend.version in the future? If we do, will it be OK to keep backend and a new backend_version? Or am I overthinking this?

Could you add a test to assert backend is either "readline" or "editline"?

Very helpful insight! From my observation, the difference between different versions of the same library is significantly less important than the difference between the two libraries. I can't guarantee that the version of the backend will not ever be used at all in the future, but I would guess that's a very rare case. Both readline and editline seems to be relatively consistent between versions.

That being said, I think using a single string type for readline.backend is reasonable, and it probably makes the life easier for the users compared to having a (name, version) tuple for the attribute and the users would have to do readline.backend[0] == 'readline' in most of the cases.

If in the future, readline or editline has a breaking change in some version that affects many users, having a readline.backend_version is not the end of the world.

I think we are done with the review for this PR? We should merge this before #107748 so it can use the backend instead of checking __doc__

One more thing: the docs entry needs a versionadded. I updated the PR directly to avoid a round of review ping-pong; hope you don't mind.

(And I fell into the trap of thinking this is simple enough to use GitHub UI rather than a local clone...)

Thanks for the quick review!