There was an error in the way that we calculate the reference
scope sequences that are used as the basis for assertions about
changed ranges in randomized tests. The error caused some
characters' scopes to not be checked. This corrects the reference
implementation and fixes a previously uncaught bug in the
implementation of `tree_path_get_changed_ranges`.
Previously, when iterating over the old and new trees, we would
only perform comparisons of visible nodes. This resulted in a failure
to do any comparison for portions of the text in which there were
trailing invisible child nodes (e.g. trailing `_line_break` nodes
inside `statement` nodes in the JavaScript grammar).
Now, we additionally perform comparisons at invisible leaf nodes,
based on their lowest visible ancestor.
The parser spends the majority of its time allocating and freeing trees and stack nodes.
Also, the memory footprint of the AST is a significant concern when using tree-sitter
with large files. This library is already unlikely to work very well with source files
larger than 4GB, so representing rows, columns, byte lengths and child indices as
unsigned 32 bit integers seems like the right choice.
The `pos` and `size` functions for Nodes now return TSLength structs,
which contain lengths in both characters and bytes. This is important
for knowing the number of unicode characters in a Node.
