Delete global constant pool from the parser #2826

soutaro · 2026-01-23T02:48:36Z

The parser currently uses global constant pool and updates the pool during parsing through RBS_KEYWORD node. This is not thread-safe and we cannot run the parser in parallel. This PR is to remove the global constant pool and RBS_KEYWORD node by:

Simplify the location structure from the nested one to flat struct attributes
Introducing C enums for symbol literal unions (for example, method/attribute visibility is :public | :private)

This is not a breaking change in Ruby library -- the RBS::Location structure nor the union type attributes are not changed -- but introduces breaking changes in C level API.

Flat location attributes

The nested location attributes are expanded to flat struct attributes.

Instead of having one location attribute and sub-locations are stored under it, node structs directly have sub-locations are its attributes.

    // before: sub-locations are saved under location attribute.
    rbs_location_t *location;

    // after: The main location is saved as `location`, but sub-locations are saved to their own attributes.
    rbs_location _range location;
    rbs_location_range keyword_range;     /* Required */
    rbs_location_range name_range;        /* Required */
    rbs_location_range end_range;         /* Required */
    rbs_location_range type_params_range; /* Optional */
    rbs_location_range colon_range;       /* Optional */
    rbs_location_range self_types_range;  /* Optional */

The nested locations are then reconstructed in ast_translation.c.

        VALUE location = rbs_location_range_to_ruby_location(ctx, node->base.location);
        rbs_loc *loc = rbs_check_location(location);
        rbs_loc_legacy_alloc_children(loc, 5);
        rbs_loc_legacy_add_required_child(loc, rb_intern("keyword"), (rbs_loc_range) { .start = node->keyword_range.start, .end = node->keyword_range.end });
        rbs_loc_legacy_add_required_child(loc, rb_intern("name"), (rbs_loc_range) { .start = node->name_range.start, .end = node->name_range.end });
        rbs_loc_legacy_add_required_child(loc, rb_intern("end"), (rbs_loc_range) { .start = node->end_range.start, .end = node->end_range.end });
        rbs_loc_legacy_add_optional_child(loc, rb_intern("type_params"), (rbs_loc_range) { .start = node->type_params_range.start, .end = node->type_params_range.end });
        rbs_loc_legacy_add_optional_child(loc, rb_intern("lt"), (rbs_loc_range) { .start = node->lt_range.start, .end = node->lt_range.end });
        rb_hash_aset(h, ID2SYM(rb_intern("location")), location);

C `enum`s for symbol literal unions

The union attributes have their own enum types.

enum rbs_attribute_visibility {
    RBS_ATTRIBUTE_VISIBILITY_UNSPECIFIED, /* unspecified (nil) */
    RBS_ATTRIBUTE_VISIBILITY_PUBLIC,      /* public (:public) */
    RBS_ATTRIBUTE_VISIBILITY_PRIVATE,     /* private (:private) */
};

The values are translated to Ruby symbols (and nil).

Uses `ID` directly instead of constant_id.

soutaro · 2026-01-23T04:06:08Z

rust/ruby-rbs/src/node/mod.rs

    #[must_use]
    pub fn start(&self) -> i32 {
-        unsafe { (*self.pointer).rg.start.byte_pos }
+        self.range.start


This should be fixed. The rbs_location_range has character index, but should have byte position here.

alexcrocha · 2026-01-23T20:47:38Z

rust/ruby-rbs-sys/src/lib.rs


    use rbs_encoding_type_t::RBS_ENCODING_UTF_8;

    static INIT: Once = Once::new();


We can remove this now since we no longer use it.

alexcrocha · 2026-01-23T20:47:52Z

rust/ruby-rbs-sys/src/lib.rs

-            rbs_constant_pool_init(RBS_GLOBAL_CONSTANT_POOL, 26);
-        });
-    }
+    fn setup() {}


We can also remove this as it's empty.

alexcrocha · 2026-01-23T21:03:26Z

rust/ruby-rbs/src/node/mod.rs

    }
 }

 pub struct RBSLocation {


Since rbs_location_t has been renamed to rbs_location_range_t, we can rename this to RBSLocationRange. It still needs to hold a pointer.

alexcrocha · 2026-01-23T21:09:02Z

rust/ruby-rbs/build.rs

+                    "alias_kind"
+                    | "attribute_kind"
+                    | "attribute_visibility"
+                    | "method_definition_kind"
+                    | "method_definition_visibility"
+                    | "type_param_variance" => {


We should create Rust enums and return the enums for these instead of u32. This would allow us to leverage Rust's type system and provide better ergonomics.

alexcrocha · 2026-01-23T21:12:44Z

rust/ruby-rbs/src/node/mod.rs

+    }
+}
+
 pub struct RBSLocationListIter {


We can remove this struct since we now use RBSLocationRangeList

alexcrocha · 2026-01-23T21:13:16Z

rust/ruby-rbs/build.rs

+                            writeln!(file, "    #[must_use]")?;
+                            writeln!(
+                                file,
+                                "    pub fn {}(&self) -> Option<rbs_location_range> {{",


This should return the Rust RBSLocationRange struct.

alexcrocha · 2026-01-23T21:13:35Z

rust/ruby-rbs/build.rs

+                            writeln!(file, "    #[must_use]")?;
+                            writeln!(
+                                file,
+                                "    pub fn {}(&self) -> rbs_location_range {{",


This should also return the Rust RBSLocationRange struct.

alexcrocha · 2026-01-23T21:37:11Z

rust/ruby-rbs/build.rs

+                        writeln!(file)?;
+                    }
+                    "rbs_location_range_list" => {
+                        write_node_field_accessor(&mut file, field, "RBSLocationRangeList")?;


The write_node_field_accessor helper function is meant to add accessors to other nodes that need the parser. Location doesn't need it, so we need to treat it differently.

soutaro added this to the RBS 4.0 milestone Jan 23, 2026

soutaro added 16 commits January 23, 2026 12:24

Rename rbs_loc_*** types to rbs_location_***

3303647

Introduce rbs_loc_*** for *legacy* locations

fe9799d

Uses `ID` directly instead of constant_id.

Generate AST with flat location ranges

3f19884

Add locations: in config

76fe7d7

Migrate explicitly declared rbs_location fields

914e60d

Migrate rbs_location_list fields

2090a51

Make the location attribute rbs_location_range

2f567a0

Delete unused rbs_location types/functions

495d51a

Rename Type to Node in template.rb

0691840

Overhaul config.yml

d5dee4c

Introduces enums in config.yml

d4e6ec2

Delete RBS_AST_KEYWORD and rbs_ast_keyword_t

ecaa57b

Add rbs_ivar_name type

d3645cd

Delete global constant pool

62affc0

Delete optional: true for enum fields from config.yml

a3207a9

Fix rust crates

15d181e

soutaro force-pushed the flat-locations branch from 1384b7d to 15d181e Compare January 23, 2026 03:24

soutaro added 3 commits January 23, 2026 12:25

cargo fmt

bdf6147

Fix rubocop

4b603ed

fix test

92d9009

soutaro force-pushed the flat-locations branch from 1f3e7c9 to 92d9009 Compare January 23, 2026 03:39

cargo fmt

8e49bbc

soutaro commented Jan 23, 2026

View reviewed changes

alexcrocha reviewed Jan 23, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Delete global constant pool from the parser #2826

Delete global constant pool from the parser #2826

soutaro commented Jan 23, 2026 •

edited

Loading

Uh oh!

soutaro Jan 23, 2026 •

edited

Loading

Uh oh!

alexcrocha Jan 23, 2026

Uh oh!

alexcrocha Jan 23, 2026

Uh oh!

alexcrocha Jan 23, 2026

Uh oh!

alexcrocha Jan 23, 2026

Uh oh!

alexcrocha Jan 23, 2026

Uh oh!

alexcrocha Jan 23, 2026

Uh oh!

alexcrocha Jan 23, 2026

Uh oh!

alexcrocha Jan 23, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants


		use rbs_encoding_type_t::RBS_ENCODING_UTF_8;

		static INIT: Once = Once::new();

Delete global constant pool from the parser #2826

Are you sure you want to change the base?

Delete global constant pool from the parser #2826

Conversation

soutaro commented Jan 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Flat location attributes

C enums for symbol literal unions

Uh oh!

soutaro Jan 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

soutaro commented Jan 23, 2026 •

edited

Loading

C `enum`s for symbol literal unions

soutaro Jan 23, 2026 •

edited

Loading