feat(csharp): emit a member node per property (#3006)

C# properties got no graph node, so a property was invisible as a class member (only
fields and methods appeared). Emit a member node per property_declaration with the same id
scheme, ownership, and defines/field-context edge as the C++ data-member branch; all
property shapes (auto, expression-bodied, read-only) are covered, the #2913 property-type
references walk is preserved, and no builtin types are fabricated.
This commit is contained in:
durmazoguzhan
2026-08-24 14:08:25 +01:00
committed by safishamsi
parent da8ca1684c
commit 3ef625b3b6
2 changed files with 173 additions and 0 deletions
+16
View File
@@ -3750,6 +3750,22 @@ def _extract_generic(
# field. Use _csharp_collect_type_refs (like the Java/PHP/Kotlin
# siblings) so `List<Widget>` yields both the List field ref and the
# Widget generic_arg ref.
# A property becomes a node, the way a C++ data member does. Fields
# stay out: the id recipe casefolds and strips leading underscores, so
# `_count` and `Count` normalize to the same id, and emitting both
# would hand the node to whichever the parser reached first — in
# practice the private backing field, hiding the public member behind
# it. See #3006 for the follow-up.
prop_node_name = node.child_by_field_name("name")
if prop_node_name is not None:
property_name = _read_text(prop_node_name, source)
if property_name:
property_line = node.start_point[0] + 1
property_nid = _make_id(parent_class_nid, property_name)
if property_nid not in seen_ids:
add_node(property_nid, property_name, property_line)
add_edge(parent_class_nid, property_nid, "defines",
property_line, context="field")
type_node = node.child_by_field_name("type")
if type_node is not None:
# Record the property's declared type for the method-scoped
+157
View File
@@ -0,0 +1,157 @@
"""C# properties get a node, like C++ data members (#3006).
`_CSHARP_CONFIG` emitted a `references` edge to a member's *type* and no node for
the member, so C# was the only language with a class-shaped type layer whose
state was absent from the graph. C++ emits a node per data member with `defines`,
#2971 is adding the same for C structs, and #2220 gave Swift computed properties
nodes. This brings C# to that line.
Properties only, not fields. `ids.normalize_id` casefolds and strips leading
underscores, so `_count` and `Count` are the same id: emitting both would hand
the node to whichever the parser reached first, which is the private backing
field, hiding the public member behind it. In C# the property is the state's
public identity, so it is the half worth having first.
"""
from __future__ import annotations
import os
import tempfile
from pathlib import Path
from graphify.extract import extract
def _extract(tmp_path, files: dict[str, str]):
for name, body in files.items():
p = tmp_path / name
p.parent.mkdir(parents=True, exist_ok=True)
p.write_text(body)
old = os.getcwd()
try:
os.chdir(tmp_path)
r = extract([Path(n) for n in files], cache_root=Path(tempfile.mkdtemp()))
finally:
os.chdir(old)
defines = {(e["source"], e["target"]) for e in r["edges"] if e["relation"] == "defines"}
return defines, r
def _find(r, label):
return next(n["id"] for n in r["nodes"] if n["label"] == label)
def _labels(r):
return [n["label"] for n in r["nodes"]]
def test_auto_property_becomes_a_member_node(tmp_path):
defines, r = _extract(tmp_path, {"S.cs": (
"public class Variant {\n"
" public bool IsShowCategory { get; set; }\n"
"}\n"
)})
assert (_find(r, "Variant"), _find(r, "IsShowCategory")) in defines
def test_every_property_on_a_class_gets_its_own_node(tmp_path):
defines, r = _extract(tmp_path, {"S.cs": (
"public class Variant {\n"
" public bool IsShowCategory { get; set; }\n"
" public bool ShowOnWeb { get; set; }\n"
"}\n"
)})
variant = _find(r, "Variant")
assert (variant, _find(r, "IsShowCategory")) in defines
assert (variant, _find(r, "ShowOnWeb")) in defines
def test_a_primitive_property_still_gets_a_node(tmp_path):
# `int` produces no member-worthy type reference, and the property is still
# part of the class's state.
defines, r = _extract(tmp_path, {"S.cs": (
"public class Counter {\n"
" public int Count { get; set; }\n"
"}\n"
)})
assert (_find(r, "Counter"), _find(r, "Count")) in defines
def test_a_generic_parameter_typed_property_still_gets_a_node(tmp_path):
# The type-reference path returns early for a type parameter, which is right
# for a reference and wrong for the member itself.
defines, r = _extract(tmp_path, {"S.cs": (
"public class Box<T> {\n"
" public T Value { get; set; }\n"
"}\n"
)})
assert (_find(r, "Box"), _find(r, "Value")) in defines
def test_a_backing_field_does_not_take_the_property_node(tmp_path):
# `_count` and `Count` normalize to one id. The public member is the one to
# keep, and there is exactly one edge rather than two onto a shared node.
defines, r = _extract(tmp_path, {"S.cs": (
"public class Counter {\n"
" private int _count;\n"
" public int Count { get; set; }\n"
"}\n"
)})
assert (_find(r, "Counter"), _find(r, "Count")) in defines
assert len(defines) == 1
assert "_count" not in _labels(r)
def test_a_field_alone_makes_no_member_node_but_keeps_its_type_reference(tmp_path):
# Fields are out for now, and the reference to a field's type is untouched.
defines, r = _extract(tmp_path, {"S.cs": (
"public class Runner {\n"
" private readonly Worker _worker;\n"
"}\n"
"public class Worker { }\n"
)})
assert defines == set()
references = {(e["source"], e["target"], e.get("context")) for e in r["edges"]
if e["relation"] == "references"}
assert (_find(r, "Runner"), _find(r, "Worker"), "field") in references
def test_property_type_references_are_kept(tmp_path):
# Regression guard for #1591: the member node is additive, the reference to
# the property's type stays.
_, r = _extract(tmp_path, {"S.cs": (
"public class Holder {\n"
" public Widget Main { get; set; }\n"
"}\n"
"public class Widget { }\n"
)})
references = {(e["source"], e["target"], e.get("context")) for e in r["edges"]
if e["relation"] == "references"}
assert (_find(r, "Holder"), _find(r, "Widget"), "field") in references
assert "Main" in _labels(r)
def test_methods_are_still_methods(tmp_path):
# A property is not a method: it lands on `defines`, the method on `method`.
defines, r = _extract(tmp_path, {"S.cs": (
"public class Variant {\n"
" public bool Flag { get; set; }\n"
" public void Touch() { }\n"
"}\n"
)})
variant = _find(r, "Variant")
methods = {(e["source"], e["target"]) for e in r["edges"] if e["relation"] == "method"}
assert (variant, _find(r, "Flag")) in defines
assert (variant, _find(r, ".Touch()")) in methods
assert (variant, _find(r, ".Touch()")) not in defines
def test_case_only_sibling_properties_do_not_duplicate_an_edge(tmp_path):
# Legal C#, and one id after normalization. One node, one edge, rather than a
# second edge hung on the first property's node.
defines, r = _extract(tmp_path, {"S.cs": (
"public class Odd {\n"
" public int Count { get; set; }\n"
" public int count { get; set; }\n"
"}\n"
)})
assert len(defines) == 1