Optimize Color::find_named_color()

[dalexeev]

Use binary search instead of linear.

Note that GDScript already caches constant values, unlike RichTextLabel. If we still decide to use a HashMap, then we need to store two names (for example, ALICE_BLUE and ALICEBLUE), since the name with underscore is needed for constant binding.

[RedworkDE]

This and your unicode pr makes me think that we should maybe introduce a proper binary_search function that takes a three way comparison and return the complement of the would be index when not found, to for cases like these where information about the existence of the value is also reqired and the existing bsearch function is not useful.

[lawnjelly]

This should probably just use some kind of HashMap as said imo. I don't see the advantage of binary search here, perhaps I'm missing something.

You could probably have a collision free hash even with the duplicates.

Also this kind of stuff should likely be parsed once then reused on subsequent frames, so there's only a cost if the text is constantly changing. I'm assuming this is the case, but I haven't read any of this code.

[dalexeev]

This should probably just use some kind of HashMap as said imo. I don't see the advantage of binary search here, perhaps I'm missing something.

Done.

[AThousandShips]

Haven't tested but LGTM

[akien-mga]

The optimization looks good.

---

I'm a bit puzzled at the general implementation of that feature though, it feels pretty convoluted, especially with the "normalization" done to support apparently arbitrary variants like "yel_lo.w'G R'-een"...

Seems like unnecessary complexity, but well, changing it now would break compat.
With a simpler design, we could just have a hashmap for the colors directly instead of storing an intermediate map of indices.

[akien-mga]

Thanks!