feat(go): resolve member calls through the receiver's declared type

`g.Greet()` where `Greeter` is declared in another file produced no edge. The
shared cross-file pass skips member calls, and the Go extractor read the
receiver's identifier only to discard it for anything that was not an imported
package — so the one shape Go dependency injection actually takes (a struct
field used from a method) contributed nothing to the call graph.

`extract_go` now exports a per-file `go_type_table` built from the four places a
Go receiver's type is written down: a struct field, a parameter (which covers the
method receiver `func (s *Server)`), `var x T`, and `x := T{}` / `x := &T{}`.
`_resolve_go_member_calls` looks the receiver up there, takes the single
declaration of that type and emits the `calls` edge to its method.

Also stamps `_callable` / `_callable_class` on Go declarations (#2438). Go is the
one extractor outside the tree-sitter engine, so its nodes carried neither, which
left them invisible to every consumer that reads the markers off the node.

Design notes:

- The receiver travels as `member_receiver`, not `receiver`: in this extractor
  `receiver` means "imported package", and the Swift/Python/Ruby member-call
  resolvers select on `is_member_call` plus `receiver` without checking the
  language, so a Go name there binds to their types in a mixed corpus.
- `s.logger.Log()` is read as a call on the `logger` field only when `s` is the
  enclosing method's own receiver. The table is flat and file-scoped, so any
  other head could be a package or a variable whose type it cannot confirm.
- Names are matched case-sensitively, unlike the sibling resolvers. Go exports by
  capitalisation, so `Run` and `run` on one type are two different methods.
- Go does not defer: a member call reaches `raw_calls` only once the bare callee
  name misses in the caller's file, so in-file behaviour is byte-identical.
- Two packages declaring the same name stay two type nodes and the
  single-definition guard bails — without import evidence neither is the answer.

(cherry picked from commit deca9fa366)

graphify/extract.py

@@ -6796,6 +6796,92 @@ def _resolve_rust_self_member_calls(
         })
 
 
+def _resolve_go_member_calls(
+    per_file: list[dict],
+    all_nodes: list[dict],
+    all_edges: list[dict],
+) -> None:
+    """Resolve Go member calls (``g.Greet()``) through the receiver's declared type.
+
+    The shared cross-file pass skips member calls, so a method call on a receiver whose
+    type is declared in another file resolved to nothing. The per-file ``go_type_table``
+    names the declared type of every struct field, parameter (the method receiver
+    included), ``var x T`` and ``x := T{}`` binding; this pass looks the receiver up
+    there, takes the single declaration of that type, and emits the ``calls`` edge to its
+    method. Always INFERRED: the type comes from the table, never from the call site.
+
+    Names are matched case-sensitively, unlike the sibling resolvers. Go exports by
+    capitalisation, so `Run` and `run` on one type are two different methods with two
+    different visibilities, and folding them would pick whichever came last.
+    """
+    raw = [
+        rc
+        for result in per_file
+        for rc in result.get("raw_calls", [])
+        if rc.get("language") == "go" and rc.get("is_member_call")
+        and rc.get("member_receiver") and rc.get("callee") and rc.get("caller_nid")
+    ]
+    if not raw:
+        return
+    type_table_by_file: dict[str, dict[str, str]] = {}
+    for result in per_file:
+        tt = result.get("go_type_table")
+        if tt and tt.get("path"):
+            type_table_by_file[tt["path"]] = tt.get("table", {})
+
+    def _key(label: object) -> str:
+        return str(label or "").strip().removeprefix(".").removesuffix("()")
+
+    # A Go type node id folds in the package directory, so two packages declaring the
+    # same name stay two entries and the single-definition guard below bails: without
+    # import evidence neither one is the answer.
+    type_def_nids: dict[str, list[str]] = {}
+    node_by_id: dict[str, dict] = {}
+    for n in all_nodes:
+        node_by_id[n.get("id")] = n
+        if n.get("_callable_class") and n.get("source_file"):
+            type_def_nids.setdefault(_key(n.get("label", "")), []).append(n["id"])
+
+    method_index: dict[tuple[str, str], list[str]] = {}
+    for e in all_edges:
+        if e.get("relation") != "method":
+            continue
+        tnode = node_by_id.get(e.get("target"))
+        if tnode is not None:
+            method_index.setdefault(
+                (e.get("source"), _key(tnode.get("label", ""))), []).append(e["target"])
+
+    existing_pairs = {(e.get("source"), e.get("target")) for e in all_edges}
+    for rc in raw:
+        receiver, callee, caller = rc["member_receiver"], rc["callee"], rc["caller_nid"]
+        type_name = type_table_by_file.get(rc.get("source_file", ""), {}).get(receiver)
+        if not type_name or type_name in _LANGUAGE_BUILTIN_GLOBALS:
+            continue
+        type_defs = type_def_nids.get(_key(type_name), [])
+        if len(type_defs) != 1:  # ambiguous or absent -> bail (god-node guard)
+            continue
+        targets = method_index.get((type_defs[0], _key(callee)), [])
+        if len(targets) != 1:
+            continue
+        target = targets[0]
+        if target == caller or (caller, target) in existing_pairs:
+            continue
+        existing_pairs.add((caller, target))
+        all_edges.append({
+            "source": caller,
+            "target": target,
+            "relation": "calls",
+            "context": "call",
+            "confidence": "INFERRED",
+            # The rubric's discrete INFERRED scale (references/extraction-spec.md):
+            # a single-definition type-table hit is the high-confidence rung.
+            "confidence_score": 0.85,
+            "source_file": rc.get("source_file", ""),
+            "source_location": rc.get("source_location"),
+            "weight": 1.0,
+        })
+
+
 def _resolve_elixir_import_targets(
     per_file: list[dict],
     all_nodes: list[dict],
@@ -7145,6 +7231,12 @@ register_language_resolver(
         "kotlin_member_calls", frozenset({".kt", ".kts"}), _resolve_kotlin_member_calls
     )
 )
+# Go receiver-typed member-call resolution: `g.Greet()` where the method is declared in
+# another file. The shared pass skips member calls, so these had no edge unless the bare
+# method name happened to match a symbol in the caller's own file.
+register_language_resolver(
+    LanguageResolver("go_member_calls", frozenset({".go"}), _resolve_go_member_calls)
+)
 # C# qualified construction (#2997): `new A.B.Cache()` arrives as the bare name,
 # so a colliding `Cache` elsewhere makes it ambiguous. Runs in the tail registry
 # beside csharp_member_calls and matches the prefix against declared namespaces.

graphify/extractors/go.py

@@ -133,6 +133,58 @@ def _go_collect_type_refs(node, source: bytes, generic: bool, out: list[tuple[st
             if c.is_named:
                 _go_collect_type_refs(c, source, generic, out, skip)
 
+def _go_single_type_name(node, source: bytes) -> str | None:
+    """The one declared type a Go type expression names, or None.
+
+    Only `Greeter` and `*Greeter` qualify. A slice, map or channel element is not the
+    receiver of `x.M()`, and a `qualified_type` (`pkg.Greeter`) names a type no bare
+    local name can stand for.
+    """
+    if node is not None and node.type == "pointer_type":
+        node = next((c for c in node.children if c.is_named), None)
+    if node is None or node.type != "type_identifier":
+        return None
+    name = _read_text(node, source)
+    return name if name and name not in _GO_PREDECLARED_TYPES else None
+
+
+def _go_receiver_type_table(root, source: bytes) -> dict[str, str]:
+    """Collect ``name -> TypeName`` for every Go receiver whose type is written down.
+
+    Four sources: a struct field, a parameter (which covers the method receiver
+    `func (s *Server)`), `var x T`, and `x := T{}` / `x := &T{}`. File-scoped and flat,
+    first binding wins — a parameter shadowing a field in another function must not
+    retype the field's own calls. Children are pushed reversed so the walk yields
+    document order and "first" means first in the file.
+    """
+    table: dict[str, str] = {}
+    stack = [root]
+    while stack:
+        n = stack.pop()
+        t = n.type
+        if t in ("field_declaration", "parameter_declaration", "var_spec"):
+            type_name = _go_single_type_name(n.child_by_field_name("type"), source)
+            for c in n.children if type_name else ():
+                if c.type in ("field_identifier", "identifier"):
+                    table.setdefault(_read_text(c, source), type_name)
+        elif t == "short_var_declaration":
+            left, right = n.child_by_field_name("left"), n.child_by_field_name("right")
+            if left is not None and right is not None:
+                names = [c for c in left.children if c.type == "identifier"]
+                values = [c for c in right.children if c.is_named]
+                for name_node, value in zip(names, values):
+                    if value.type == "unary_expression":
+                        value = value.child_by_field_name("operand") or value
+                    if value.type != "composite_literal":
+                        continue
+                    type_name = _go_single_type_name(
+                        value.child_by_field_name("type"), source)
+                    if type_name:
+                        table.setdefault(_read_text(name_node, source), type_name)
+        stack.extend(reversed(n.children))
+    return table
+
+
 def extract_go(path: Path) -> dict:
     """Extract functions, methods, type declarations, and imports from a .go file."""
     try:
@@ -158,7 +210,13 @@ def extract_go(path: Path) -> dict:
     nodes: list[dict] = []
     edges: list[dict] = []
     seen_ids: set[str] = set()
-    function_bodies: list[tuple[str, object]] = []
+    # `(caller nid, body, receiver variable name)` — the receiver name is what makes
+    # `s.logger.Log()` readable as a call on the `logger` field rather than on `s`.
+    function_bodies: list[tuple[str, object, str | None]] = []
+    # Fed to the `_callable` / `_callable_class` stamp below (#2438): the indirect-call
+    # guard and the cross-repo member-call pass both read them off the node.
+    callable_nids: set[str] = set()
+    callable_class_nids: set[str] = set()
     # local package name (including aliases) -> written Go import path
     go_imported_pkgs: dict[str, str] = {}
 
@@ -351,22 +409,27 @@ def extract_go(path: Path) -> dict:
                 line = node.start_point[0] + 1
                 func_nid = symbol_nid(_make_id(stem, func_name), func_name)
                 add_node(func_nid, f"{func_name}()", line)
+                callable_nids.add(func_nid)
                 add_edge(file_nid, func_nid, "contains", line)
                 emit_go_method_refs(node, func_nid, line)
                 body = node.child_by_field_name("body")
                 if body:
-                    function_bodies.append((func_nid, body))
+                    function_bodies.append((func_nid, body, None))
             return
 
         if t == "method_declaration":
             receiver = node.child_by_field_name("receiver")
             receiver_type: str | None = None
+            receiver_name: str | None = None
             if receiver:
                 for param in receiver.children:
                     if param.type == "parameter_declaration":
                         type_node = param.child_by_field_name("type")
                         if type_node:
                             receiver_type = _read_text(type_node, source).lstrip("*").strip()
+                        name_node = param.child_by_field_name("name")
+                        if name_node:
+                            receiver_name = _read_text(name_node, source)
                         break
             name_node = node.child_by_field_name("name")
             if not name_node:
@@ -377,6 +440,7 @@ def extract_go(path: Path) -> dict:
             if receiver_type:
                 parent_nid = _make_id(pkg_scope, receiver_type)
                 add_node(parent_nid, receiver_type, line)
+                callable_class_nids.add(parent_nid)
                 method_nid = symbol_nid(_make_id(parent_nid, method_name), method_name)
                 add_node(method_nid, f".{method_name}()", line)
                 add_edge(parent_nid, method_nid, "method", line)
@@ -384,11 +448,12 @@ def extract_go(path: Path) -> dict:
                 method_nid = symbol_nid(_make_id(stem, method_name), method_name)
                 add_node(method_nid, f"{method_name}()", line)
                 add_edge(file_nid, method_nid, "contains", line)
+            callable_nids.add(method_nid)
 
             emit_go_method_refs(node, method_nid, line)
             body = node.child_by_field_name("body")
             if body:
-                function_bodies.append((method_nid, body))
+                function_bodies.append((method_nid, body, receiver_name))
             return
 
         if t == "type_declaration":
@@ -402,6 +467,7 @@ def extract_go(path: Path) -> dict:
                 line = child.start_point[0] + 1
                 type_nid = _make_id(pkg_scope, type_name)
                 add_node(type_nid, type_name, line)
+                callable_class_nids.add(type_nid)
                 add_edge(file_nid, type_nid, "contains", line)
                 # A generic type's own parameters (`type Box[T any]`) are local
                 # bindings; a field or method using `T` must not mint a stub.
@@ -533,6 +599,16 @@ def extract_go(path: Path) -> dict:
     _scan_declarations(root)
     walk(root)
 
+    # A type carries both markers, as in the tree-sitter engine: `_callable_class` is
+    # the narrowing, not a separate kind.
+    for n in nodes:
+        if n["id"] in callable_nids or n["id"] in callable_class_nids:
+            n["_callable"] = True
+            if n["id"] in callable_class_nids:
+                n["_callable_class"] = True
+
+    type_table = _go_receiver_type_table(root, source)
+
     label_to_nid: dict[str, str] = {}
     for n in nodes:
         raw = n["label"]
@@ -542,7 +618,7 @@ def extract_go(path: Path) -> dict:
     seen_call_pairs: set[tuple[str, str]] = set()
     raw_calls: list[dict] = []
 
-    def walk_calls(node, caller_nid: str) -> None:
+    def walk_calls(node, caller_nid: str, self_name: str | None) -> None:
         if node.type in ("function_declaration", "method_declaration"):
             return
         if node.type == "call_expression":
@@ -551,6 +627,7 @@ def extract_go(path: Path) -> dict:
             is_member_call: bool = False
             is_bare_identifier: bool = False
             package_receiver: str | None = None
+            member_receiver: str | None = None
             import_path: str | None = None
             if func_node:
                 if func_node.type == "identifier":
@@ -566,6 +643,17 @@ def extract_go(path: Path) -> dict:
                     if not is_member_call:
                         package_receiver = receiver_name
                         import_path = go_imported_pkgs[receiver_name]
+                    elif operand is not None and operand.type == "identifier":
+                        member_receiver = receiver_name
+                    elif operand is not None and operand.type == "selector_expression":
+                        # `s.logger.Log()` names the receiver by the field only when `s` is
+                        # this method's own receiver; any other head could be a package or
+                        # a variable whose type the flat table does not know.
+                        inner = operand.child_by_field_name("operand")
+                        if (self_name and inner is not None and inner.type == "identifier"
+                                and _read_text(inner, source) == self_name):
+                            member_receiver = _read_text(
+                                operand.child_by_field_name("field"), source)
                     if field:
                         callee_name = _read_text(field, source)
             if is_bare_identifier and callee_name in _GO_PREDECLARED_FUNCS:
@@ -599,15 +687,20 @@ def extract_go(path: Path) -> dict:
                         "is_member_call": is_member_call,
                         "language": "go",
                         "receiver": package_receiver,
+                        # A key of its own: `receiver` means "imported package" here, and
+                        # the Swift/Python/Ruby member-call resolvers select on
+                        # `is_member_call` plus `receiver` without checking the language,
+                        # so a Go name placed there binds to their types in a mixed corpus.
+                        "member_receiver": member_receiver,
                         "import_path": import_path,
                         "source_file": str_path,
                         "source_location": f"L{node.start_point[0] + 1}",
                     })
         for child in node.children:
-            walk_calls(child, caller_nid)
+            walk_calls(child, caller_nid, self_name)
 
-    for caller_nid, body_node in function_bodies:
-        walk_calls(body_node, caller_nid)
+    for caller_nid, body_node, self_name in function_bodies:
+        walk_calls(body_node, caller_nid, self_name)
 
     valid_ids = seen_ids
     clean_edges = []
@@ -621,4 +714,5 @@ def extract_go(path: Path) -> dict:
         "edges": clean_edges,
         "raw_calls": raw_calls,
         "go_imports": dict(go_imported_pkgs),
+        "go_type_table": {"path": str_path, "table": type_table},
     }

tests/test_go_receiver_member_calls.py

@@ -0,0 +1,161 @@
+"""Go member calls resolve through the receiver's declared type.
+
+The shared cross-file pass skips member calls and the Go extractor read the receiver's
+name only to throw it away, so `g.Greet()` on a receiver whose type lives in another
+file produced no edge — the Go twin of the Swift gap in #1356. Each case below pins one
+source of the receiver's type, and the negative cases pin what must stay unresolved: a
+constructor return, a chain that does not start at the method's own receiver, a
+package-qualified type, and an ambiguous type name.
+"""
+from __future__ import annotations
+
+import importlib
+
+import pytest
+
+from graphify.extract import extract
+
+pytestmark = pytest.mark.skipif(
+    importlib.util.find_spec("tree_sitter_go") is None,
+    reason="tree_sitter_go not installed",
+)
+
+GREETER = "package svc\n\ntype Greeter struct{}\n\nfunc (g *Greeter) Greet() {}\n"
+
+
+def _calls(tmp_path, files: dict[str, str]):
+    for name, body in files.items():
+        path = tmp_path / name
+        path.parent.mkdir(parents=True, exist_ok=True)
+        path.write_text(body, encoding="utf-8")
+    result = extract([tmp_path / n for n in files],
+                     cache_root=tmp_path / "graphify-out", parallel=False)
+    label = {n["id"]: n["label"] for n in result["nodes"]}
+    calls = {(label.get(e["source"]), label.get(e["target"])): e
+             for e in result["edges"] if e["relation"] == "calls"}
+    return calls, result
+
+
+def _greet_edge(calls: dict) -> dict | None:
+    return next((e for (src, tgt), e in calls.items()
+                 if src and "un()" in src and tgt == ".Greet()"), None)
+
+
+def test_a_typed_parameter_types_the_receiver(tmp_path):
+    calls, _ = _calls(tmp_path, {
+        "svc/greeter.go": GREETER,
+        "svc/app.go": "package svc\n\nfunc Run(g *Greeter) { g.Greet() }\n",
+    })
+    edge = _greet_edge(calls)
+    assert edge is not None, calls
+    # The type came from the table, never from the call site: `g` names a variable, so
+    # there is no spelling of this call that would be exact.
+    assert edge["confidence"] == "INFERRED"
+
+
+def test_a_struct_field_types_the_receiver_through_the_methods_own_receiver(tmp_path):
+    # The dominant Go shape: the dependency is held as a field and used from a method.
+    calls, _ = _calls(tmp_path, {
+        "svc/greeter.go": GREETER,
+        "svc/app.go": "package svc\n\ntype App struct {\n\tgreeter *Greeter\n}\n\n"
+                      "func (a *App) Run() { a.greeter.Greet() }\n",
+    })
+    assert _greet_edge(calls) is not None, calls
+
+
+def test_a_var_declaration_types_the_receiver(tmp_path):
+    calls, _ = _calls(tmp_path, {
+        "svc/greeter.go": GREETER,
+        "svc/app.go": "package svc\n\nfunc Run() {\n\tvar g Greeter\n\tg.Greet()\n}\n",
+    })
+    assert _greet_edge(calls) is not None, calls
+
+
+def test_a_composite_literal_binding_types_the_receiver(tmp_path):
+    calls, _ = _calls(tmp_path, {
+        "svc/greeter.go": GREETER,
+        "svc/app.go": "package svc\n\nfunc Run() {\n\tg := Greeter{}\n\tg.Greet()\n}\n",
+    })
+    assert _greet_edge(calls) is not None, calls
+
+
+def test_an_address_of_composite_literal_types_the_receiver(tmp_path):
+    calls, _ = _calls(tmp_path, {
+        "svc/greeter.go": GREETER,
+        "svc/app.go": "package svc\n\nfunc Run() {\n\tg := &Greeter{}\n\tg.Greet()\n}\n",
+    })
+    assert _greet_edge(calls) is not None, calls
+
+
+def test_a_constructor_return_resolves_to_nothing(tmp_path):
+    # `g := NewGreeter()` types `g` only by reading the constructor's return type, which
+    # this pass does not do; nothing in the file writes `Greeter` next to `g`.
+    calls, _ = _calls(tmp_path, {
+        "svc/greeter.go": GREETER + "\nfunc NewGreeter() *Greeter { return &Greeter{} }\n",
+        "svc/app.go": "package svc\n\nfunc Run() {\n\tg := NewGreeter()\n\tg.Greet()\n}\n",
+    })
+    assert _greet_edge(calls) is None, calls
+
+
+def test_a_chain_not_rooted_at_the_methods_receiver_resolves_to_nothing(tmp_path):
+    # `p.greeter.Greet()` on a parameter: the flat table cannot say that `greeter` is a
+    # field of `p`'s type rather than of some other type declared in the file.
+    calls, _ = _calls(tmp_path, {
+        "svc/greeter.go": GREETER,
+        "svc/app.go": "package svc\n\ntype App struct {\n\tgreeter *Greeter\n}\n\n"
+                      "func Run(p *App) { p.greeter.Greet() }\n",
+    })
+    assert _greet_edge(calls) is None, calls
+
+
+def test_a_package_qualified_type_resolves_to_nothing(tmp_path):
+    # `other.Greeter` names a type in another package; binding it to the local `Greeter`
+    # by bare name would ignore the qualifier.
+    calls, _ = _calls(tmp_path, {
+        "svc/greeter.go": GREETER,
+        "svc/app.go": "package svc\n\nimport \"example.com/other\"\n\n"
+                      "func Run(g *other.Greeter) { g.Greet() }\n",
+    })
+    assert _greet_edge(calls) is None, calls
+
+
+def test_two_packages_declaring_the_same_type_resolve_to_neither(tmp_path):
+    # The single-definition guard: without import evidence, guessing one of two
+    # `Greeter`s is worse than leaving the call unresolved.
+    calls, _ = _calls(tmp_path, {
+        "a/greeter.go": GREETER.replace("package svc", "package a"),
+        "b/greeter.go": GREETER.replace("package svc", "package b"),
+        "svc/app.go": "package svc\n\nfunc Run(g *Greeter) { g.Greet() }\n",
+    })
+    assert _greet_edge(calls) is None, calls
+
+
+def test_the_first_binding_of_a_name_wins(tmp_path):
+    # The table is flat per file, so a parameter named like an earlier one has to lose:
+    # otherwise `other`'s signature would redirect the first function's calls.
+    calls, result = _calls(tmp_path, {
+        "svc/greeter.go": GREETER,
+        "svc/other.go": "package svc\n\ntype Other struct{}\n\nfunc (o *Other) Greet() {}\n",
+        "svc/app.go": "package svc\n\nfunc Run(g *Greeter) { g.Greet() }\n\n"
+                      "func Also(g *Other) { g.Greet() }\n",
+    })
+    source_of = {n["id"]: str(n.get("source_file") or "") for n in result["nodes"]}
+    label_of = {n["id"]: n["label"] for n in result["nodes"]}
+    run_targets = {source_of[e["target"]] for e in result["edges"]
+                   if e["relation"] == "calls"
+                   and label_of.get(e["source"]) == "Run()"
+                   and label_of.get(e["target"]) == ".Greet()"}
+    assert len(run_targets) == 1, run_targets
+    assert run_targets.pop().endswith("greeter.go")
+
+
+def test_a_same_file_method_call_keeps_its_extracted_edge(tmp_path):
+    # In-file resolution still happens by bare label before this pass, so the edge must
+    # stay EXTRACTED and must not be doubled.
+    calls, _ = _calls(tmp_path, {
+        "svc/app.go": "package svc\n\ntype App struct{}\n\nfunc (a *App) Greet() {}\n\n"
+                      "func Run(a *App) { a.Greet() }\n",
+    })
+    edges = [e for (src, tgt), e in calls.items() if src == "Run()" and tgt == ".Greet()"]
+    assert len(edges) == 1, calls
+    assert edges[0]["confidence"] == "EXTRACTED"