Atom is currently beta testing tree-sitter grammars. tree-sitter/tree-sitter is a tool that builds complete parse trees of source code.

Using this instead of TextMate-style regex parsing will make things like #2 and #6 easier to solve.

Support for a tree-sitter grammar can be added alongside the current grammar, so I've started work on it: https://github.com/tobiasvl/tree-sitter-rgbasm

There are no standard for file extensions (besides these guidelines), so we should make it configurable instead of highjacking common assembler extensions.

Some of the scopes we're using now don't fit the current guidelines.

One example is labels. The entire label is entity.name.label, which is good, but the label itself is entity.name.function.label, which is wrong (but which I admit I scoped wrong on purpose, to get a more varied syntax highlighting).

But we should probably be consistent with existing conventions:

• https://manual.macromates.com/en/language_grammars#naming_conventions
• https://www.sublimetext.com/docs/3/scope_naming.html

We need to support operators.

There are a lot of false positives while parsing labels.

• scope1.scope2.label should not be legal label (labels can only have a global and a local scope) (partly fixed in f111876 but the parsing could still be a lot better)
• .local_label:: should not be legal (only global labels may be exported)

• Probably more

It would be nice to have some snippets – Game Boy games require a lot of boilerplate, after all – but I'm not sure what they should be.

Users can define their own character sets for use in some constant types with the OPT directive. See documentation.

I'm not sure how easy that is to implement. This Vim syntax file uses a naïve approach.