Unnecessary String construction in `MatchesIterator` reduces query performance

[Emmeral]

In the tryAdvance method of the MatchesIterator class there is the following line

var hasNoText = tree.getText() == null;

This looks to be a simple predicate, but under the hood the Tree#getText() method will to a conversion from a byte array to a String object

    public @Nullable String getText() {
        return charset != null ? new String(source, charset) : null;
    }

This needs some time for Trees with a big underlying text which is a pity as we are not using the constructed string value in the end.

Ideas to solve this

Either add a method hastText() to the Tree class that avoids the unnecessary conversion or reopen #65 to retrieve the raw source bytes (potentially as a read only byte buffer ) in MatchesIterator.

What do you think @ObserverOfTime ?

[ObserverOfTime]

We could instead remove this check and collect the bytes similar to tree-sitter/py-tree-sitter#343 but I don't know how much of a performance impact this will have.

[Emmeral]

Your proposal would be an improvement for query evaluation if the source bytes are only available with a callable. This looks like a bigger change (and the solution to a different issue).

My issue would already be solved if the determination whether a tree has text available was faster (e.g. by not converting bytes to a string if the string is not needed anyway). So I think only fixing this will be simpler.

[maloewe-ona]

CapturesIterator is probably affected in the same way:

java-tree-sitter/src/main/java/io/github/treesitter/jtreesitter/CapturesIterator.java

Line 41 in a2fabed

var hasNoText = tree.getText() == null;