Improve module, type and function definitions

[savq]

Types and modules

• Add bare module definitions.
• Fix single line structs.
• Split tests for structs and other types.

Functions

• Add operators as names in function definitions.
• Add zero method function definitions.
• Add short form function definitions.
• Add function signature rule (shared by block and short function definitions).
• Add where clauses.
• Refactor anonymous block functions (function() ... end). Consider them function definitions, not function expressions.
• Split tests for function definitions into 4 sections.

Function parameters and scoped identifiers

• Add anonymous typed parameters.
• Add parameters to do_clause.
• Rename spread_parameter to slurp_parameter (yes, that's the real name).
• Allow _primary_expression and prefixed_string_literal as types in typed_parameter and return_type.
• Allow scoped identifiers in:
  • function definitions
  • macro identifiers
  • type parameters

Misc

• Split import rules into selected_import and _import_list. These cannot be used in the same import statement.
• Remove 29 files from known-failures.

Caveats:

• where is actually a binary operator. It can be used in places beyond function definitions. A new rule could be added in the future to handle more use cases.
• I wanted to add a _parameter rule as suggested here, but this caused more conflicts than necessary.
• I made the terminator after do_clause parameters mandatory.
• The conflicts that were added are mostly due to the fact that assignment_expression still allows call_expression in the LHS.
• The reason for calling ... a "slurp" comes from here: What does the ... operator do?

Once assignment patterns are fixed, these conflicts could be removed (and _parameter could finally be added), but I think that belongs in another PR.

---

Closes #26 Closes #40 Closes #41 Closes #42 Closes #36, but the implementation is different from #43 Closes #57 Closes #60 Closes #61 Closes #63

[ChrHorn]

Very nice, I will close https://github.com/tree-sitter/tree-sitter-julia/pull/43 in favor of getting this one through.

Could you also add support for empty functions, for example function f end. These are often used to add docstrings.

[ExpandingMan]

This is awesome, thanks!

Just wanted to report a minor issue I found in which sometimes type assertions can be parsed as symbols, see this example

[ExpandingMan]

Another minor issue, the ∈ character in for loops is not being recognized (should be @keyword.operator to match in).

[savq]

Just wanted to report a minor issue I found in which sometimes type assertions can be parsed as symbols, see this example

@ExpandingMan The PR fixes this 👍

Another minor issue, the ∈ character in for loops is not being recognized (should be @keyword.operator to match in).

This can be fixed in either the grammar or the queries, and it's a bit unrelated to the rest of the PR. I'll add a fix for this in nvim-treesitter first.

[savq]

Ok. All the updated rules and their correspoding tests should be ready.

[aviks]

Should this be merged, @maxbrunsfeld ? Seems like a significant improvement.

[tecosaur]

Regarding slurp_expression, would splat_expression be more appropriate? I see ... described as the "splat operator" more than anything else. You'll also see it called the "splat operator" (but not the "slurp operator") in Python, Ruby, and other places.

[savq]

I wrote the wrong thing in the PR comment (I updated it). For future reference:

• spread_parameter was renamed to slurp_parameter (here)
• spread_expression was renamed to splat_expression (in #83)

That's the "correct" naming convention in Julia. See What does the ... operator do?