BUG: declare and use correctly self.unique_check

[toobaz]

I am not sure I understood the last comment in the previous pull request, but anyway this is a rebased version.

In the previous PR I proposed to simply make the __get__ of is_unique directly rely on self.initialized, but I am no more so sure it is a good idea, since the current status is slightly redundant but clearer (and coherent with the use of "is_monotonic").

[jreback]

ok, what exactly does this fix?

[toobaz]

First, self.unique_check (which follows the same logic of self.monotonic_check) is now initialized.

Then, when is_unique is accessed, self.unique_check is checked. But,self.unique_check is (was) not set when the index is not unique. So on a non-unique index, every access to is_unique results in a call to _do_unique_check, which calls then _ensure_mapping_populated, which finally does nothing (in calls following the first) because self.initialized is now 1. So the performance loss is minimal, but still it doesn't make any sense.

[jreback]

is unique_check really a duplicate of initialized? if so, just remove it. check that everything passes and perf is unchanged.

[toobaz]

See #10229

[jorisvandenbossche]

@toobaz just for future reference, you don't have to open a new PR every time you update. Just update your branch and push again will do!

[toobaz]

@jorisvandenbossche : since I was indeed trying to understand this... by "update" do you mean "push a new commit" or "rebase and push -f"? Because I tried the latter and it didn't seem to work (but I may have been mistaken).

[jorisvandenbossche]

As long as you stay on the same branch, it does not matter. Adding a new commit, rebasing, etc all work and the PR will be updated with that if you push (force) again.

[jorisvandenbossche]

But I see that the new PR uses another branch name. So you only need to stay on the same branch to update the PR

[toobaz]

I changed branch name precisely because I didn't think using the same would work! Thanks for the tip.