Hi @rhettinger, was there anything wrong with #11781 ?

Hmm, I didn't see that PR until after this one was made.

Looking at it now, I prefer this one which has cleaner code, better docs updates, and focuses on key and leaves out reversed which was not requested by the OP (if that turns out to be needed, it can be added later; for now, it is better to avoid doubling the code complexity).

I don't care which version gets merged but regarding the reversed argument you said in the review:

FWIW, if key-function support is added, it is inevitable that reversed support will be requested (since sorted() and bisect() are often used together).

Also, some of the tests should be combined. Both PRs have tests that the others don't.

Am unsure about reversed. It seems like a nice to have, but looking at the other PR, it doubles the code complexity and that spills into the docs/tests as well. What do you think?

For reversed, most users that know they need bisect should be able to write the key function so that it gives the correct order directly and don't need the help of reversed. It's not part of the original request and complicates both the Python and C code.

If it's ever needed you could add it later but since it increases the complexity without adding functionality, it may be best not to add it.

There is a note recommending lru_cache() (maybe just cache() now) which I probably would not have thought of and may be worth keeping:

When specifying a custom key function, you should wrap it with
:func:functools.lru_cache if the key function is not already fast.

Edit: Nevermind, I read the doc better and it's actually here.

BTW, do you know how to appease the buildbots for the doc build. I'm unclear on what they deem "suspicious" and how to clear the blocker.

You can fix the spurious error os "suspicious" by removing lines 114 and 115 of Doc/tools/susp-ignored.csv:

library/bisect,32,:hi,all(val >= x for val in a[i:hi])
library/bisect,42,:hi,all(val > x for val in a[i:hi])

Thanks.

Apologies again for duplicating work. I completely forget about the other PR.

Somehow bisect(lst, fn(x), key=fn) feels very uncomfortable. After all, we're looking to find the insertion point for x in lst, rather than fn(x). I'd really expect the syntax to be bisect(lst, x, key=fn). Would the Python devs be open for such a change?