Problem

We're using the alias function a lot in several grammars, including tree-sitter-html. It's been really useful, but it has two downsides:

1. It increases the size of the parse table somewhat because otherwise-identical REDUCE actions can have different alias_sequence_id values.
2. It causes tokens to appear with different types when they occur within ERROR nodes, because in that case, the aliases are not applied.

The second problem came to my attention when adapting Atom's autocomplete-html package to work with Tree-sitter.

Solution

In this PR, I introduce an internal concept called simple aliases. A grammar symbol is considered to have a simple alias if the symbol is always used with that alias: it's never used without an alias, and it's never used with a different alias.

Simple aliases can be dealt with in a way that is, well, simpler than how we normally implement aliases 😁. This results in a smaller parse table. More importantly though, it causes the rule to always appear with the same name, even when it occurs inside of an ERROR, which makes the syntax tree more consistent and useful overall.