[3.x] Make AnimationNodeBlendTree use OrderedHashMap insteads Map

[TokageItLab]

3.x version of #79595. Supersedes #79531.

An ordered map must be used.

...I am not familiar with OrderedHashMap, does its order depend on the key? If it does not depend, then StringName is sufficient without making to be a String.

[AThousandShips]

You can use StringName::AlphCompare here as well, check the template for the map

Edit: the "ordered" part seems to mean that it preserves insertion order, not comparison, I think

I'd suggest Map<StringName, Node, StringName::AlphCompare>

[TokageItLab]

It is sufficient if the order of the inserted order, or the order of the index list of the map depending on the key is unique.

At least from testing #79595 it seems to work well, but does unique comparison results mean that the order of the index list is unique? @AThousandShips

[AThousandShips]

Pointer comparison means unique order, in one run, and unless the names are freed and reallocated they'll remain ordered, the order doesn't persist between runs though

But in cases where the nose is removed and re-added it'll have a new order, if using hash map, and it'll have if the node is removed, and the string name is unallocated because it is freed

Using pointer comparison is more performant, so if possible it'd be better

I suspect the issue with pointers is only that it doesn't remain consistent between runs, and isn't consistent if the StringName is freed (completely, i.e. ref-count reaches zero)

So if you add "ABC" and "BCD" they'll have one order, if you then remove "ABC" and re-add it, after "ABC" is freed, it'll be unknown where it'll end up

Sorry if I gave a confusing answer 😅

[TokageItLab]

Looking at the code for Map/RBMap, I see that the comparison by Comparator uniquely determines the index to be stored.

When StringName is used as a key of the Map/RBMap:

• When using AlphCompare, the index is uniquely determined by the string
• If nothing is specified, the index is determined by comparison of memory address, so if the memory address changes due to restart, etc., the index will change

@AThousandShips Is this correct?

[AThousandShips]

That sounds correct from my experience and reading of the code

My suggestion would be then:

• If the order between runs is important (like seems to be the case in 4.x) AlphCompare is good
• Otherwise pointer comparison should be preferred as it is significantly more performant

[AThousandShips]

LGTM, haven't tested the code but usage is correct

[akien-mga]

Thanks!