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

v8's `_resolve_php_member_calls` (#3872) binds scoped `Class::method()` calls
but leaves instance `$g->method()` calls unresolved. This adds the instance
tier beside it: a per-file receiver type table (`_php_receiver_type_table`
over typed properties, promoted constructor params, typed params and `new`
bindings) names the receiver's class, and the shared resolver takes the single
PHP declaration of that class and emits the `calls` edge, INFERRED.

The declaration index is gated to the PHP interop family: the index is
corpus-wide, so a same-named class in another language would both answer the
receiver and hide that no PHP file declares it. A receiver the table cannot
type to one class — untyped, union, longer chain — stays unresolved.

Co-Authored-By: Claude Code <noreply@anthropic.com>

graphify/extract.py

@@ -5710,7 +5710,7 @@ def _resolve_php_member_calls(
     type_def_nids: dict[str, list[str]] = {}
     for node in all_nodes:
         if (
-            node.get("source_file")
+            _lang_family(node.get("source_file")) == "php"
             and node.get("id") in contained
             and _is_type_like_definition(node)
         ):
@@ -5733,6 +5733,12 @@ def _resolve_php_member_calls(
         if edge.get("relation") == "inherits":
             inherits_bases.setdefault(edge["source"], []).append(edge["target"])
 
+    type_table_by_file: dict[str, dict[str, str]] = {}
+    for result in per_file:
+        tt = result.get("php_type_table")
+        if tt and tt.get("path"):
+            type_table_by_file[str(tt["path"])] = tt.get("table", {})
+
     existing_pairs = {(edge.get("source"), edge.get("target")) for edge in all_edges}
     for result in per_file:
         for raw_call in result.get("raw_calls", []):
@@ -5746,6 +5752,7 @@ def _resolve_php_member_calls(
             if not receiver or not callee or not caller:
                 continue
 
+            inferred = False
             receiver_lower = receiver.lower()
             if receiver_lower in ("self", "static"):
                 type_nid = enclosing_type.get(caller)
@@ -5756,6 +5763,25 @@ def _resolve_php_member_calls(
                 if len(bases) != 1:
                     continue
                 type_nid = bases[0]
+            elif not raw_call.get("php_scoped") and not receiver.startswith("$"):
+                # Instance receiver (`$g->m()`, `$this->prop->m()`): the type comes
+                # from the annotation table, never the call site, so this rung is
+                # INFERRED. An untyped, union or longer-chain receiver is absent from
+                # the table and stays unresolved rather than guessing a class name.
+                typed = type_table_by_file.get(
+                    str(raw_call.get("source_file", "")), {}).get(receiver)
+                if typed is None:
+                    continue
+                inferred = True
+                type_defs = type_def_nids.get(class_key(typed), [])
+                if not type_defs:
+                    _park_unresolved_member_call(
+                        node_by_id.get(caller), callee, typed, "php", raw_call,
+                    )
+                    continue
+                if len(type_defs) != 1:
+                    continue
+                type_nid = type_defs[0]
             elif not receiver.startswith("$"):
                 type_defs = type_def_nids.get(class_key(receiver), [])
                 if not type_defs:
@@ -5783,8 +5809,10 @@ def _resolve_php_member_calls(
                 "target": method_nid,
                 "relation": "calls",
                 "context": "call",
-                "confidence": "EXTRACTED",
-                "confidence_score": 1.0,
+                "confidence": "INFERRED" if inferred else "EXTRACTED",
+                # Table-typed receiver -> the discrete INFERRED rung (0.85);
+                # a source-named class scope stays EXTRACTED (1.0).
+                "confidence_score": 0.85 if inferred else 1.0,
                 "source_file": raw_call.get("source_file", ""),
                 "source_location": raw_call.get("source_location"),
                 "weight": 1.0,

graphify/extractors/engine.py

@@ -788,6 +788,77 @@ def _php_collect_type_refs(node, source: bytes, generic: bool, out: list[tuple[s
             if c.is_named:
                 _php_collect_type_refs(c, source, generic, out)
 
+def _php_declared_type_name(node, source: bytes) -> str | None:
+    """The single class name a PHP type annotation names, or None.
+
+    Only `Greeter` and `?Greeter` qualify: a union, an intersection or a primitive names
+    no one class, and binding a receiver to the first arm of `A|B` would be a guess.
+    """
+    if node is None:
+        return None
+    if node.type == "optional_type":
+        node = next((c for c in node.children if c.is_named), None)
+        if node is None:
+            return None
+    if node.type != "named_type":
+        return None
+    return next((_php_name_text(c, source) for c in node.children
+                 if c.type in ("name", "qualified_name")), None)
+
+
+def _php_variable_text(node, source: bytes) -> str | None:
+    """The bare name a PHP `variable_name` binds: `$greeter` -> `greeter`."""
+    if node is None or node.type != "variable_name":
+        return None
+    return next((_read_text(c, source) for c in node.children if c.type == "name"), None)
+
+
+def _php_first_declared_type(node, source: bytes) -> str | None:
+    """The first single-class type annotation among ``node``'s direct children."""
+    return next((name for name in (_php_declared_type_name(c, source) for c in node.children)
+                 if name), None)
+
+
+def _php_receiver_type_table(root, source: bytes, table: dict[str, str]) -> None:
+    """Collect ``name -> TypeName`` for every PHP receiver whose type is written down.
+
+    Four sources, all needed: a typed property (`private Greeter $g;`), a constructor
+    promotion (`__construct(private Greeter $g)`), a typed parameter, and
+    `$g = new Greeter()`. File-scoped and flat, first binding wins — a parameter
+    shadowing a property in another method must not retype the property's own calls.
+    Children are pushed reversed so the walk yields document order and "first" means
+    first in the file.
+    """
+    stack = [root]
+    while stack:
+        n = stack.pop()
+        t = n.type
+        if t == "property_declaration":
+            type_name = _php_first_declared_type(n, source)
+            for element in n.children if type_name else ():
+                if element.type != "property_element":
+                    continue
+                name = next((_php_variable_text(c, source) for c in element.children
+                             if c.type == "variable_name"), None)
+                if name and name not in table:
+                    table[name] = type_name
+        elif t in ("property_promotion_parameter", "simple_parameter"):
+            type_name = _php_first_declared_type(n, source)
+            name = next((_php_variable_text(c, source) for c in n.children
+                         if c.type == "variable_name"), None)
+            if name and type_name and name not in table:
+                table[name] = type_name
+        elif t == "assignment_expression":
+            right = n.child_by_field_name("right")
+            if right is not None and right.type == "object_creation_expression":
+                name = _php_variable_text(n.child_by_field_name("left"), source)
+                type_name = next((_php_name_text(c, source) for c in right.children
+                                  if c.type in ("name", "qualified_name")), None)
+                if name and type_name and name not in table:
+                    table[name] = type_name
+        stack.extend(reversed(n.children))
+
+
 def _php_method_return_type_node(method_node):
     """Return the named_type/primitive_type node sitting after formal_parameters."""
     saw_params = False
@@ -8227,6 +8298,17 @@ def _extract_generic(
                             and callee_name
                             and callee_name.casefold() in _LANGUAGE_BUILTIN_GLOBALS):
                         self_receiver = "this"
+                    # `$this->greeter->greet()` types the receiver by the property name,
+                    # `$greeter->greet()` by the variable; a longer chain names neither.
+                    obj = node.child_by_field_name("object")
+                    if obj is not None and obj.type == "variable_name":
+                        member_receiver = _php_variable_text(obj, source)
+                    elif obj is not None and obj.type == "member_access_expression":
+                        inner = obj.child_by_field_name("object")
+                        if inner is not None and inner.type == "variable_name" and (
+                                _php_variable_text(inner, source) == "this"):
+                            member_receiver = _read_text(
+                                obj.child_by_field_name("name"), source)
             elif config.ts_module == "tree_sitter_cpp":
                 # C++: function field, then field_expression/qualified_identifier
                 func_node = node.child_by_field_name(config.call_function_field) if config.call_function_field else None
@@ -8603,6 +8685,12 @@ def _extract_generic(
                         # suffix sets, so a source_file suffix alone can't separate them.
                         if config.ts_module == "tree_sitter_cpp":
                             rc_entry["lang"] = "cpp"
+                        # PHP: separate a scoped class call (`Helper::m()`, source-
+                        # qualified) from an instance member call (`$g->m()`), whose
+                        # receiver is a variable the resolver must type through the
+                        # per-file table instead of reading it as a class name.
+                        if config.ts_module == "tree_sitter_php":
+                            rc_entry["php_scoped"] = is_scoped_call
                         # C#: tag the raw_call so _resolve_csharp_member_calls claims
                         # it, and stamp the receiver's type from the method's SCOPED
                         # bindings by the call's byte offset (#1609, per-method since
@@ -9166,6 +9254,8 @@ def _extract_generic(
     if config.ts_module in ("tree_sitter_javascript", "tree_sitter_typescript"):
         _ts_receiver_type_table(root, source, type_table)
         _ts_mark_public_twins(root, source, result)
+    if config.ts_module == "tree_sitter_php":
+        _php_receiver_type_table(root, source, type_table)
     if config.ts_module == "tree_sitter_swift":
         if type_table or swift_factory_bindings:
             result["swift_type_table"] = {"path": str_path, "table": type_table}
@@ -9179,6 +9269,8 @@ def _extract_generic(
             result["ts_type_table"] = {"path": str_path, "table": type_table}
         elif config.ts_module == "tree_sitter_cpp":
             result["cpp_type_table"] = {"path": str_path, "table": type_table}
+        elif config.ts_module == "tree_sitter_php":
+            result["php_type_table"] = {"path": str_path, "table": type_table}
     return result
 
 def _ts_mark_public_twins(root, source: bytes, result: dict) -> None:

tests/test_php_receiver_member_calls.py

@@ -0,0 +1,184 @@
+"""PHP member calls resolve through the receiver's declared type.
+
+The shared cross-file pass skips member calls, and PHP recorded no receiver at all, so
+`$greeter->greet()` on a receiver whose class lives in another file produced no edge —
+the PHP 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: an untyped
+parameter, a union type, a longer chain, and an ambiguous class name.
+"""
+from __future__ import annotations
+
+import importlib
+
+import pytest
+
+from graphify.extract import extract
+
+pytestmark = pytest.mark.skipif(
+    importlib.util.find_spec("tree_sitter_php") is None,
+    reason="tree_sitter_php not installed",
+)
+
+GREETER = "<?php\nclass Greeter {\n    public function greet(): void {}\n}\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 "run" in src and tgt == ".greet()"), None)
+
+
+def test_a_typed_property_types_the_receiver(tmp_path):
+    calls, _ = _calls(tmp_path, {
+        "Greeter.php": GREETER,
+        "App.php": "<?php\nclass App {\n    private Greeter $greeter;\n"
+                   "    public function run(): void { $this->greeter->greet(); }\n}\n",
+    })
+    edge = _greet_edge(calls)
+    assert edge is not None, calls
+    # The type came from the table, never from the call site: `$greeter` names a
+    # variable, so there is no spelling of this call that would be exact.
+    assert edge["confidence"] == "INFERRED"
+
+
+def test_a_promoted_constructor_parameter_types_the_receiver(tmp_path):
+    # Constructor promotion declares no property, so nothing else in the walk ever
+    # names the receiver's type.
+    calls, _ = _calls(tmp_path, {
+        "Greeter.php": GREETER,
+        "App.php": "<?php\nclass App {\n"
+                   "    public function __construct(private Greeter $greeter) {}\n"
+                   "    public function run(): void { $this->greeter->greet(); }\n}\n",
+    })
+    assert _greet_edge(calls) is not None, calls
+
+
+def test_a_typed_parameter_types_the_receiver(tmp_path):
+    calls, _ = _calls(tmp_path, {
+        "Greeter.php": GREETER,
+        "App.php": "<?php\nclass App {\n"
+                   "    public function run(Greeter $greeter): void { $greeter->greet(); }\n}\n",
+    })
+    assert _greet_edge(calls) is not None, calls
+
+
+def test_a_new_binding_types_an_unannotated_local(tmp_path):
+    calls, _ = _calls(tmp_path, {
+        "Greeter.php": GREETER,
+        "App.php": "<?php\nclass App {\n    public function run(): void {\n"
+                   "        $greeter = new Greeter();\n        $greeter->greet();\n    }\n}\n",
+    })
+    assert _greet_edge(calls) is not None, calls
+
+
+def test_a_nullable_property_type_still_names_one_class(tmp_path):
+    calls, _ = _calls(tmp_path, {
+        "Greeter.php": GREETER,
+        "App.php": "<?php\nclass App {\n    private ?Greeter $greeter = null;\n"
+                   "    public function run(): void { $this->greeter->greet(); }\n}\n",
+    })
+    assert _greet_edge(calls) is not None, calls
+
+
+def test_an_untyped_parameter_resolves_to_nothing(tmp_path):
+    calls, _ = _calls(tmp_path, {
+        "Greeter.php": GREETER,
+        "App.php": "<?php\nclass App {\n"
+                   "    public function run($greeter): void { $greeter->greet(); }\n}\n",
+    })
+    assert _greet_edge(calls) is None, calls
+
+
+def test_a_union_typed_receiver_resolves_to_nothing(tmp_path):
+    # `Greeter|Other` names no one class, and binding the first arm would be a guess.
+    calls, _ = _calls(tmp_path, {
+        "Greeter.php": GREETER,
+        "Other.php": "<?php\nclass Other {\n    public function greet(): void {}\n}\n",
+        "App.php": "<?php\nclass App {\n"
+                   "    public function run(Greeter|Other $greeter): void "
+                   "{ $greeter->greet(); }\n}\n",
+    })
+    assert _greet_edge(calls) is None, calls
+
+
+def test_a_longer_chain_resolves_to_nothing(tmp_path):
+    # `$this->a->b->greet()` types neither `a` nor `b` as the receiver.
+    calls, _ = _calls(tmp_path, {
+        "Greeter.php": GREETER,
+        "App.php": "<?php\nclass App {\n    private Greeter $greeter;\n"
+                   "    public function run(): void { $this->greeter->inner->greet(); }\n}\n",
+    })
+    assert _greet_edge(calls) is None, calls
+
+
+def test_two_classes_of_the_same_name_resolve_to_neither(tmp_path):
+    # The single-definition guard: guessing one of two `Greeter`s is worse than
+    # leaving the call unresolved.
+    calls, _ = _calls(tmp_path, {
+        "a/Greeter.php": GREETER,
+        "b/Greeter.php": GREETER,
+        "App.php": "<?php\nclass App {\n    private Greeter $greeter;\n"
+                   "    public function run(): void { $this->greeter->greet(); }\n}\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 a property has to lose:
+    # otherwise `other`'s signature would redirect the property's own calls.
+    calls, result = _calls(tmp_path, {
+        "Greeter.php": GREETER,
+        "Other.php": "<?php\nclass Other {\n    public function greet(): void {}\n}\n",
+        "App.php": "<?php\nclass App {\n    private Greeter $greeter;\n"
+                   "    public function run(): void { $this->greeter->greet(); }\n"
+                   "    public function other(Other $greeter): void { $greeter->greet(); }\n}\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 "run" in str(label_of.get(e["source"]))
+                   and label_of.get(e["target"]) == ".greet()"}
+    assert len(run_targets) == 1, run_targets
+    assert run_targets.pop().endswith("Greeter.php")
+
+
+def test_a_static_call_keeps_its_own_path(tmp_path):
+    # `Helper::format()` is a scoped call: source-qualified, so it binds to the
+    # method itself and stays EXTRACTED, never routed through the type table.
+    calls, _ = _calls(tmp_path, {
+        "Helper.php": "<?php\nclass Helper {\n    public static function format(): void {}\n}\n",
+        "App.php": "<?php\nclass App {\n"
+                   "    public function run(): void { Helper::format(); }\n}\n",
+    })
+    edge = next((e for (src, tgt), e in calls.items()
+                 if src and "run" in src and tgt == ".format()"), None)
+    assert edge is not None, calls
+    assert edge["confidence"] == "EXTRACTED"
+
+
+def test_a_class_from_another_language_never_answers_a_php_receiver(tmp_path):
+    # The declaration index is corpus-wide, so a same-named Java class would both answer
+    # the receiver and hide that no PHP file declares it — the call belongs to the merge.
+    calls, result = _calls(tmp_path, {
+        "App.php": "<?php\nclass App {\n    private Greeter $greeter;\n"
+                   "    public function run(): void { $this->greeter->greet(); }\n}\n",
+        "Greeter.java": "public class Greeter { public void greet() {} }\n",
+    })
+    assert _greet_edge(calls) is None, calls
+    parked = [(n.get("metadata") or {}).get("unresolved_calls") for n in result["nodes"]
+              if "run" in str(n["label"]) and n.get("metadata")]
+    assert parked == [[{"callee": "greet", "receiver_type": "Greeter",
+                        "lang": "php", "line": "L4"}]], parked

tests/test_php_static_call_resolution.py

@@ -158,10 +158,10 @@ def test_method_name_case_insensitivity(tmp_path):
     assert any(s == ".run()" and t.lower() == ".sanitize()" for s, t in calls)
 
 
-def test_instance_call_is_still_unresolved_no_regression(tmp_path):
-    """`$obj->method()` was never resolved before this fix (no receiver-typed
-    PHP resolver existed) and must remain so - this fix only targets scoped
-    (`::`) calls, not instance (`->`) calls."""
+def test_instance_call_resolves_through_the_receivers_declared_type(tmp_path):
+    """`$h->sanitize()` on a `Helper $h` parameter now resolves through the
+    receiver's declared type (INFERRED), the instance-receiver tier that extends
+    the scoped (`::`) resolver."""
     calls, _ = _extract(tmp_path, {
         "Helper.php": (
             "<?php\nnamespace App;\nclass Helper {\n"
@@ -170,4 +170,7 @@ def test_instance_call_is_still_unresolved_no_regression(tmp_path):
             "<?php\nnamespace App;\nclass Caller {\n"
             "    public function run(Helper $h, $x) { return $h->sanitize($x); }\n}\n"),
     })
-    assert not any(s == ".run()" for s, _t in calls)
+    edge = next((e for (s, t), e in calls.items()
+                 if s == ".run()" and t == ".sanitize()"), None)
+    assert edge is not None, calls
+    assert edge["confidence"] == "INFERRED"