wip: allow providing the key at insertion time for EntryRef

[djugei]

Either explicitly through insert_kv or implicitly through insert_clone.

I have added an assert in insert_kv just to be sure, but if it is only a logic error to insert a different key at an entry or is caught deeper in the stack it should be removed/changed to debug_assert.

If this is something the maintainers are interested in i can implement the EntryRef and OccupiedEntryRef variants and document them.

[cuviper]

I have added an assert in insert_kv just to be sure, but if it is only a logic error to insert a different key at an entry or is caught deeper in the stack it should be removed/changed to debug_assert.

I expect we do want a full assert, just like we did in HashSet::get_or_insert_with:

hashbrown/src/set.rs

Line 957 in eea9804

assert!(value.equivalent(&new), "new value is not equivalent");

[cuviper]

I think this should use Q: Equivalent<K> instead, just like entry_ref did when it was created.

[djugei]

ah yes that covers exactly what is needed there. had not seen that trait so far and tbh i have not read a lot of hashbrowns code, i just needed one of those methods to avoid either copying my key more than needed or doing a double lookup.

[cuviper]

Not your problem, but this is an issue too, since downstream Q might implement a bad From<&Q> for K.

[djugei]

yeah i noticed that too, that was half the reason i asked about the assert

[djugei]

i think i will drop the insert_clone as insert_kv already covers the usecase.
if passing the wrong item is just a logic error i do think that using debug_assert is the correct move, though it is somewhat likely for the compiler to be smart even about the full assert.

should i make the full pr?

[cuviper]

I'll defer to @Amanieu to decide on the API design.