feat(r): extract roxygen cross-references and S3 dispatch

CHANGELOG.md

@@ -4,6 +4,8 @@ Full release notes with details on each version: [GitHub Releases](https://githu
 
 ## 0.9.32 (unreleased)
 
+- Feat: R roxygen blocks and S3 dispatch are extracted. `@seealso` and `@inheritParams` name functions the documented one never calls, and `@template` names a `man-roxygen/` file nothing sources — neither is reachable from the AST, and both become `references` edges when the corpus defines the target exactly once. `@export` marks the public API on the node and `@family` is recorded. S3 methods link to their generic: a dotted name is treated as a method only where the corpus evidences the split (a `UseMethod` generic, or a site assigning that class), so `is.null` and `compute.stuff` fabricate nothing; when several sites assign one class, R's `new_<class>` constructor convention breaks the tie. Non-call records carry `ref_name` rather than `callee`, because the shared cross-file pass turns any `callee` into a `calls` edge by name — the trap Ruby's mixin markers hit in #1668 — which had `@seealso` shipping as a phantom call and blocking the real `references` edge as a duplicate.
+- Feat: R (`.R`/`.r`, and `Rscript` shebangs) is extracted with tree-sitter instead of being counted as code and silently dropped — it was the language the #1689 no-AST-extractor warning named as its example. R has no named-function syntax (every definition is an assignment whose RHS is an anonymous `function_definition`), so this is a bespoke extractor rather than a `LanguageConfig`; it covers `<-`/`=`/`<<-`/right-assign bindings, `\(x)` lambdas, nested definitions, `library`/`require`/`requireNamespace` and `pkg::fn` imports, and `source()`. Calls resolve corpus-wide in a registered `LanguageResolver` rather than per file: R has one shared namespace and no name-binding import, so `paste0(x)` and a sibling file's `compute(x)` are indistinguishable within a file. Anything the corpus does not define is dropped, which keeps base R out of the graph with no hardcoded list of ~1,300 base names and emits no dangling edges. The r-lib grammar has no standalone PyPI wheel, so it comes from `tree-sitter-language-pack` under a new optional `[r]` extra.
 - Fix: incremental extraction and `_rebuild_code` no longer drop a file's other tier (#2333, #2334, #2336). Node/edge ownership was keyed on `source_file` alone, so a semantic re-extract deleted a doc's AST headings and a full rebuild deleted document AST nodes. Merge is now tier-aware (an AST re-extract replaces only AST nodes and keeps the semantic layer, and vice versa), the `_origin` provenance marker is backfilled on load so old graphs self-heal, and the full-rebuild drop is scoped to sources actually regenerated.
 - Fix: `graphify update` preserves the graph's `directed` flag instead of rebuilding it undirected (#2342, thanks @Rishet11), so God-node / path ranking keeps its direction on both the clustered and `--no-cluster` rebuild paths.
 - Fix: a numeric or otherwise non-string node id from an LLM fragment no longer aborts the build with a TypeError (#2326, thanks @Rishet11); ids are coerced consistently across nodes, edges, and hyperedges.

graphify/extract.py

@@ -23,6 +23,7 @@
 )
 from .ruby_resolution import resolve_ruby_member_calls
 from .pascal_resolution import resolve_pascal_inherited_calls
+from .r_resolution import resolve_r_calls
 
 # --- migrated to graphify/extractors/ (see graphify/extractors/MIGRATION.md) ---
 from graphify.extractors.base import (  # noqa: F401
@@ -48,6 +49,7 @@
 from graphify.extractors.markdown import extract_markdown  # noqa: F401
 from graphify.extractors.pascal_forms import extract_delphi_form, extract_lazarus_form  # noqa: F401
 from graphify.extractors.powershell import extract_powershell, extract_powershell_manifest  # noqa: F401
+from graphify.extractors.r import extract_r  # noqa: F401
 from graphify.extractors.razor import extract_razor  # noqa: F401
 from graphify.extractors.rust import extract_rust  # noqa: F401
 from graphify.extractors.sln import extract_sln  # noqa: F401
@@ -3111,6 +3113,14 @@ def _key(label: str) -> str:
     )
 )
 
+# R resolves every call at runtime against one shared namespace and has no import
+# statement binding a name to a file, so the per-file extractor cannot tell a base
+# R call from a sibling-file one. Lives in graphify.r_resolution; registered here
+# as a consumer of the framework, same as the Ruby and Pascal resolvers above.
+register_language_resolver(
+    LanguageResolver("r_calls", frozenset({".r", ".R"}), resolve_r_calls)
+)
+
 
 # Inline markdown link: [text](target "optional title"). The negative lookbehind
 # excludes images (![alt](src)). The target stops at whitespace/closing paren so
@@ -4163,6 +4173,11 @@ def add_existing_edge(edge: dict) -> None:
     ".m": extract_objc,
     ".mm": extract_objc,
     ".jl": extract_julia,
+    # Both cases are listed, as for Fortran below: R sources are conventionally
+    # `.R`, and _DISPATCH.keys() is read as the supported-suffix set (a case-fold
+    # only at _get_extractor lookup would leave `.R` out of it).
+    ".r": extract_r,
+    ".R": extract_r,
     ".f": extract_fortran,
     ".F": extract_fortran,
     ".f90": extract_fortran,
@@ -4223,6 +4238,8 @@ def add_existing_edge(edge: dict) -> None:
 # rather than falling back like Pascal does. Used by the #1745 warning in
 # extract() to tell the user which extra restores the language.
 _EXTRA_FOR_EXTENSION = {
+    ".r": "r",
+    ".R": "r",
     ".sql": "sql",
     ".tf": "terraform",
     ".tfvars": "terraform",
@@ -4237,7 +4254,7 @@ def add_existing_edge(edge: dict) -> None:
 # routes them to the CODE path via _shebang_interpreter; _get_extractor must
 # honor the same signal or these files are classified as code and then silently
 # dropped by extraction. Only interpreters with a real extractor are mapped —
-# detect's wider set (perl, fish, tcsh, Rscript) stays unmapped and skipped.
+# detect's wider set (perl, fish, tcsh) stays unmapped and skipped.
 _SHEBANG_DISPATCH: dict[str, Any] = {
     "python": extract_python,
     "python2": extract_python,
@@ -4253,6 +4270,7 @@ def add_existing_edge(edge: dict) -> None:
     "lua": extract_lua,
     "php": extract_php,
     "julia": extract_julia,
+    "Rscript": extract_r,
 }
 
 
@@ -4711,7 +4729,7 @@ def extract(
         )
 
     # #1689: a file counted as code (extension in CODE_EXTENSIONS) but with no AST
-    # extractor wired up (e.g. .r/.R — there is no tree-sitter-r dispatch) silently
+    # extractor wired up (e.g. .ejs/.ets — there is no dispatch for either) silently
     # contributes zero nodes. The #1666 warning above deliberately skips these (it
     # only fires when an extractor exists), so surface them explicitly, grouped by
     # extension, rather than reporting success as if the language were mapped.

graphify/extractors/__init__.py

@@ -25,6 +25,7 @@
 from graphify.extractors.pascal import extract_pascal
 from graphify.extractors.pascal_forms import extract_delphi_form, extract_lazarus_form
 from graphify.extractors.powershell import extract_powershell, extract_powershell_manifest
+from graphify.extractors.r import extract_r
 from graphify.extractors.razor import extract_razor
 from graphify.extractors.rust import extract_rust
 from graphify.extractors.sln import extract_sln
@@ -54,6 +55,7 @@
     "pascal": extract_pascal,
     "powershell": extract_powershell,
     "powershell_manifest": extract_powershell_manifest,
+    "r": extract_r,
     "razor": extract_razor,
     "rust": extract_rust,
     "sln": extract_sln,