Context Navigation

Changes between Initial Version and Version 1 of Ticket #35399, comment 6

initial	v1
1	1	Hi, that is already in the changeset: in case the item is not hashable, it will raise a type error: we catch that, and then fallback on the original way of doing it, without much delay. So it is more a boosting mechanism that when fails, is "disabled" when running through it.
2	2
3		The hash of an expression indeed is take more time than the one of an `int`, but the idea is that this often will still pay off, since the cascade of ~~Case`-~~`When` essentially forces the database into linear search, while it might be possible that a certain database finds a way, if that is not the case, it will first look at the first `When`, if that fails, the next one, etc. so for ~1000 `When`s, it will on average take 500 items per row.
	3	The hash of an expression indeed is take more time than the one of an `int`, but the idea is that this often will still pay off, since the cascade of `Case` - `When` essentially forces the database into linear search, while it might be possible that a certain database finds a way, if that is not the case, it will first look at the first `When`, if that fails, the next one, etc. so for ~1000 `When`s, it will on average take 500 items per row.
4	4
5	5	I agree that a temporary table is probably more ideal.