tree-sitter/todo.md

TODO
====

### Handling ambiguity (GLR)
* Add a simple way to specify syntactic ambiguity resolutions in the Grammar (e.g. 'prefer declarations to statements' in C)
* Optimize the lexical DFA so that all compatible starting lex states are merged. Then, when the parse stack has multiple heads, check at runtime that each head's state corresponds to the same starting lex state.
* Fix memory leaks in the graph-structured parse stack.

### Runtime System
* Add a NodeIterator class that can be used to traverse the tree efficiently.
* Stop storing nodes' visible children in a separate array. Instead, allow traversing the tree at two different levels of abstraction: one which includes anonymous tokens (the concrete syntax tree) and one that does not (the sufficiently-abstract syntax tree).

### Testing / Quality
* Start running the clang-analyzer on the codebase on travis-CI.
* Use the Valgrind leak checker to fix the memory leaks in the runtime library.
* Randomize the editing in the language tests, using a seed that can be specified in order to reproduce failures.
* Verify that the tree's total size is as expected after each edit.

### Ubiquitous token handling
* Fix the unintuitive tree that results when ubiquitous tokens are last child of their parent node.

### Error handling
* Use information about nesting depth of tokens like '(' and ')' to make error recovery more accurate.

### Grammar Features
* Regexp assertions
  - [ ] '^'
  - [ ] '$'
  - [ ] '\b'
* Composing languages
  - [ ] Rule for referencing named grammar
  - [ ] Grammar registry object in runtime
  - [ ] Parsing returns control to parent language
* Indentation tokens

### Performance
* Optimize grammar compilation