convert remaining Node.get* methods to use paths

[ianstormtaylor]

Do you want to request a feature or report a bug?

Improvement.

What's the current behavior?

In #1997 many (almost all) of the Node.get* models were converted to use paths, which are often much more performant. But a few commonly used ones remain key-based. This is unfortunate since the more often key-based getters are used, the more often the expensive to calculate key -> path mapping has to be created.

What's the expected behavior?

If we can get Slate's core internal usage of keys to be almost all gone, performance should increase a good amount.

The remaining methods that overtly take key arguments are:

• Node.getNextBlock
• Node.getPreviousBlock
• Node.getOffset

But then there are also the *AtRange methods which are currently key-based under the covers as well:

• Node.getActiveMarksAtRange
• Node.getCharactersAtRange
• Node.getFragmentAtRange
• Node.getInlinesAtRangeAsArray
• Node.getInsertMarksAtRange
• Node.getOrderedMarksAtRange

And then some of the less standardized ones which we can either convert to using paths or try to remove entirely:

• Node.getFirstInvalidDescendant
• Node.getSelectionIndexes
• Node.getOrderedMarksBetweenPositions
• Node.getMarksAtPosition
• Node.getTextsBetweenPositionsAsArray

[ianstormtaylor]

I'd go further to say that this kind of optimization (using paths instead of keys) is probably the biggest place for performance gains in the codebase, at least as far as the data model is concerned. (There may be other ones at the DOM level in slate-react.) So any help is much appreciated!

For an example of places where the key -> path map is used in ways that degrade performance, check out this comment: #1993 (comment)

[alanchrt]

I've been doing some profiling on large documents, and it's really looking like just improving/eliminating getSelectionIndexes alone could be a major performance win to start, depending on how much it could be improved. I'm not sure if the path-based optimization will be the ticket, but it seems likely since it does so much ancestor and descendant discovery.