feat: Support tree-sitter Rust crate 0.26.x

[pboling]

This PR updates tree_stump to support the tree-sitter Rust crate 0.26.x, enabling grammars with LANGUAGE_VERSION 15 to be loaded and used. It also fixes the unsafe memory transmute that extended the LanguageRef lifetime from the borrowed scope.

Changes

ext/tree_stump/Cargo.toml

[dependencies]
tree-sitter = "0.26.3"
streaming-iterator = "0.1"

ext/tree_stump/src/language.rs

• Replace self.raw_language_ref.version() with self.raw_language_ref.abi_version()
• The Ruby Language#version method now calls abi_version() internally

ext/tree_stump/src/parser.rs

• Remove timeout_micros() and set_timeout_micros() methods
• Handle Parser::language() returning Option<LanguageRef<'_>> instead of
  Option<Language> - return None to Ruby since we can't convert LanguageRef
  back to Language without cloning

ext/tree_stump/src/tree.rs

• Convert usize to u32 when calling Node::child(index) using .try_into().unwrap()
• Convert usize to u32 when calling Node::named_child(index)
• Update child_containing_descendant to use child_with_descendant (renamed in 0.26)

ext/tree_stump/src/query.rs

• Use streaming_iterator::StreamingIterator for QueryMatches iteration:

  use streaming_iterator::StreamingIterator;
  
  while let Some(m) = matches.next() {
      // process match
  }

• This replaces the previous for m in matches { ... } pattern

Breaking Changes

• Ruby API: The following methods are removed from TreeStump::Parser:

  • timeout_micros / timeout_micros=

  These were deprecated in tree-sitter 0.25 and removed in 0.26. Users should use
  alternative cancellation mechanisms (e.g., Ruby's Timeout module).

• Ruby API: Parser#language now returns nil when called after language=.
  This is because tree-sitter 0.26's Parser::language() returns a borrowed reference
  that cannot be converted back to an owned Language. If you need to track the language,
  store it separately before setting it on the parser.

Backward Compatibility

• Grammars with LANGUAGE_VERSION 13-15 are supported
• Older grammars may work but are not guaranteed

Testing

# Load tree_stump
require 'tree_stump'

# Register a LANGUAGE_VERSION 15 grammar
TreeStump.load_language("/path/to/libtree-sitter-toml.so", "tree_sitter_toml")

# Create parser and parse
parser = TreeStump::Parser.new
parser.set_language("toml")
tree = parser.parse('key = "value"')

puts tree.root_node.kind        # => "document"
puts tree.root_node.child_count # => 1
puts tree.root_node.text        # => 'key = "value"'

Closes #16, #17

Additional Fixes

Eliminate unsafe memory transmute and prevent integer truncation.

• Remove unsound unsafe transmute in Parser::language() that extended LanguageRef lifetime from borrowed scope to 'static. The safety comment claimed languages outlive parsers, but this wasn't enforced by the type system and could lead to use-after-free.
• Redesign Parser to store language_name and look up the language directly from the global LANG_LANGUAGES map in build_query(), keeping proper lifetime bounds tied to the lock guard.
• Replace as u32 casts with .try_into().ok()? in Node::child() and Node::named_child() to prevent silent value truncation on 64-bit systems where usize can exceed u32::MAX.

Fix build errors and deprecation warnings for tree-sitter 0.26

Rust source changes (ext/tree_stump/src/*.rs):

• Fix compilation error in parser.rs by changing Query::new to accept
  &tree_sitter::Language instead of &tree_sitter::LanguageRef
• Replace deprecated magnus::exception::runtime_error() calls with
  build_error() helper throughout lib.rs, query.rs, and tree.rs
• Replace deprecated magnus::exception::type_error() with
  ruby.exception_type_error() in query.rs
• Fix deprecated into_symbol() -> into_symbol_with(&ruby) in query.rs
• Fix deprecated into_value() -> into_value_with(ruby) in query.rs
• Add #[allow(deprecated)] for unavoidable fallback case in util.rs
• Fix lifetime elision warnings by adding explicit <'_> lifetime
  parameters to Node and TreeCursor return types in tree.rs

Spec changes:

• Update spec_helper.rb to use TREE_SITTER_RUBY_PATH env var for
  grammar path, falling back to local path for CI
• Add is_missing? and missing? method tests for Node class
• Add helper methods find_missing_nodes and check_all_nodes for
  tree traversal in tests

[Copilot]

Pull request overview

This PR updates the tree_stump Ruby gem to support tree-sitter Rust crate version 0.26.x, enabling compatibility with grammars using LANGUAGE_VERSION 15. The update involves adapting to breaking API changes in tree-sitter, including removal of timeout methods, API signature changes, and adoption of streaming iterators.

Key changes:

• Updated tree-sitter dependency from 0.22 to 0.26 with new dependencies (tree-sitter-language, streaming-iterator)
• Removed deprecated timeout methods from the Parser API
• Added new Node methods (is_missing) and compatibility aliases for better API consistency
• Updated CI workflow to test additional Ruby implementations and versions

Reviewed changes

Copilot reviewed 7 out of 8 changed files in this pull request and generated 6 comments.

Show a summary per file

File	Description
ext/tree_stump/Cargo.toml	Updated tree-sitter to 0.26, added tree-sitter-language and streaming-iterator dependencies
Cargo.lock	Updated dependency versions and added new transitive dependencies for tree-sitter 0.26
ext/tree_stump/src/lib.rs	Updated language loading to use LanguageFn, removed timeout method definitions, added Node method aliases
ext/tree_stump/src/language.rs	Updated version() to call abi_version(), added new abi_version() method
ext/tree_stump/src/parser.rs	Removed timeout methods, added unsafe lifetime extension for LanguageRef
ext/tree_stump/src/tree.rs	Added is_missing() method, converted usize to u32 for child access, renamed method call to child_with_descendant
ext/tree_stump/src/query.rs	Updated to use StreamingIterator for query matches iteration
.github/workflows/rspec.yml	Enhanced CI with additional Ruby versions, experimental implementations, and improved workflow configuration

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[pboling]

The last build failed because ruby-head tag in setup ruby now points at ruby 4.1.0. The last time it ran it worked because it was still using Ruby 4.0.0, and there is a breaking change in the C API. I have fixed it in rbs-sys and magnus, but rbs-sys-env and magnus both still need to be released:

• rb-sys-env: Release Ruby 4.0, 4.1 version support (COMPARABLE_RUBY_MINORS) oxidize-rb/rb-sys#699
• Ruby 4.1 (ruby-head) Support oxidize-rb/rb-sys#696
• ✨ feat: add Ruby 4.1 version support for ruby-head compatibility oxidize-rb/rb-sys#697
• ✨ feat: Support Ruby 4.1 (ruby-head) w/ handle_weak_references matsadler/magnus#164
• Support Ruby 4.1 (ruby-head) - handle_weak_references replaces reserved field matsadler/magnus#163

@joker1007 This can be merged, but the ruby-head build will continue to fail until new versions are released for rbs-sys-env and magnus.

[pboling]

Also, I added, and then removed, CI for JRuby and Truffleruby. I just wanted to see if it was a reasonable lift to add support for them. Turns out it was more than I can handle right now.

[pboling]

@joker1007 The new rbs-sys-env has been released, but still waiting on magnus.

[pboling]

@joker1007 I think we need to make rb-sys a runtime dependency in the gemspec, because the gem seems to not build without it in some scenarios. I'm not sure why yet. I've added this, but happy to take it back out if it doesn't belong and I've misdiagnosed my build failures.

  spec.add_dependency "rb_sys", "~> 0.9.119"

Oh, and magnus will not be released for ruby-head support. We'll have to target main until there is a new release with the support...

Oh, lol, maybe that's why the build was failing... 😆