Merge pull request #95 from pythonbpf/vmlinux-gen-fix

Fix make test: regenerate vmlinux.py fixture and stop mis-collecting it

Author: varun-r-mallya

Makefile

@@ -6,18 +6,29 @@ clean:
 	rm -rf examples/*.ll examples/*.o
 	rm -rf htmlcov .coverage
 
-test:
+# Regenerate the master vmlinux.py from the running kernel's BTF, then
+# symlink it into every directory under tests/ so both pytest (which
+# resolves "import vmlinux" via pythonpath=["."]) and any test file run
+# standalone from its own directory see the same, always-fresh fixture.
+vmlinux:
+	python3 tools/vmlinux-gen.py -o vmlinux.py
+	@find tests -type d -not -path '*/__pycache__*' | while read -r d; do \
+		target=$$(python3 -c "import os,sys; print(os.path.relpath('vmlinux.py', sys.argv[1]))" "$$d"); \
+		ln -sf "$$target" "$$d/vmlinux.py"; \
+	done
+
+test: vmlinux
 	pytest tests/ -W ignore::DeprecationWarning -v --tb=short -m "not verifier"
 
-test-cov:
+test-cov: vmlinux
 	pytest tests/ -W ignore::DeprecationWarning -v --tb=short -m "not verifier" \
 		--cov=pythonbpf --cov-report=term-missing --cov-report=html
 
-test-verifier:
+test-verifier: vmlinux
 	@echo "NOTE: verifier tests shell out to 'sudo bpftool'; run 'sudo -v' first so"
 	@echo "      the timestamp does not lapse mid-run. bpftool must be installed."
 	pytest tests/test_verifier.py -W ignore::DeprecationWarning -v --tb=short -m verifier
 
 all: clean install
 
-.PHONY: all clean install test test-cov test-verifier
+.PHONY: all clean install test test-cov test-verifier vmlinux

pythonbpf/functions/functions_pass.py

@@ -403,7 +403,7 @@ def process_bpf_chunk(func_node, compilation_context, return_type):
     if func_node.args.args:
         # Only look at the first argument for now
         param = func.args[0]
-        param.add_attribute("nocapture")
+        param.add_attribute("captures(none)")
 
     probe_string = get_probe_string(func_node)
     if probe_string is not None:

tests/framework/collector.py

@@ -35,6 +35,10 @@ def collect_all_test_files() -> list[BpfTestCase]:
     cases = []
     for subdir in ("passing_tests", "failing_tests"):
         for py_file in sorted((TESTS_DIR / subdir).rglob("*.py")):
+            if py_file.name == "vmlinux.py":
+                # Not a test case: the per-directory symlink to the master
+                # vmlinux fixture module, kept alongside tests that import it.
+                continue
             rel = str(py_file.relative_to(TESTS_DIR))
             needs_vmlinux = _is_vmlinux_test(rel)
 

tools/vmlinux-gen.py

@@ -261,11 +261,109 @@ def repl(m):
         for name in invalid_ctypes:
             data = re.sub(rf"\bctypes\.{name}\b", name, data)
 
+        data = self.disambiguate_anonymous_field_names(data)
+
         with open(self.output_file, "w") as f:
             f.write(data)
 
         self.log(f"Saved final output to {self.output_file}")
 
+    def disambiguate_anonymous_field_names(self, data):
+        """Make clang2py's generic '_N' field names globally unique per struct/union.
+
+        clang2py names both (a) unnamed anonymous struct/union members that go
+        into `_anonymous_`, and (b) other unnamed/reserved members, using the
+        same per-struct sequential scheme ('_0', '_1', ...). ctypes flattens
+        anonymous members by promoting their field names onto the parent class,
+        so if a struct has an anonymous member (say '_0') whose type itself has
+        a field also named '_1', and that same struct has ANOTHER anonymous
+        member named '_1', the promoted name collides with the parent's own
+        field key. ctypes then fails with a cryptic
+        "type object 'c_ulong' has no attribute '_fields_'" (or similar) when
+        `_fields_` is assigned. Prefixing every generic '_N' key with its
+        owning struct/union name keeps these names unique across the whole
+        file, so promoted names can never collide with a sibling's own key.
+        """
+        self.log("Disambiguating generic anonymous-field names...")
+
+        def sanitize(name):
+            return re.sub(r"[^0-9A-Za-z_]", "_", name)
+
+        lines = data.split("\n")
+        class_re = re.compile(r"^class\s+(\w+)\(")
+        stmt_fields_re = re.compile(r"^(\w+)\._fields_\s*=\s*\[\s*$")
+        inline_fields_re = re.compile(r"^\s+_fields_\s*=\s*\[\s*$")
+        anon_stmt_re = re.compile(r"^(\w+)\._anonymous_\s*=\s*\((.*)\)\s*$")
+        anon_inline_re = re.compile(r"^(\s+)_anonymous_\s*=\s*\((.*)\)\s*$")
+        key_re = re.compile(r"(\('_)([0-9]+)(',)")
+
+        last_class = None
+        out = []
+        i = 0
+        n = len(lines)
+        renamed_count = 0
+        while i < n:
+            line = lines[i]
+
+            m_class = class_re.match(line)
+            if m_class:
+                last_class = m_class.group(1)
+                out.append(line)
+                i += 1
+                continue
+
+            m_anon = anon_stmt_re.match(line)
+            if m_anon:
+                cls, body = m_anon.group(1), m_anon.group(2)
+                prefix = sanitize(cls)
+                new_body, cnt = re.subn(
+                    r"'_([0-9]+)'", lambda mm: f"'_{prefix}_{mm.group(1)}'", body
+                )
+                renamed_count += cnt
+                out.append(f"{cls}._anonymous_ = ({new_body})")
+                i += 1
+                continue
+
+            m_anon_inline = anon_inline_re.match(line)
+            if m_anon_inline and last_class:
+                indent, body = m_anon_inline.group(1), m_anon_inline.group(2)
+                prefix = sanitize(last_class)
+                new_body, cnt = re.subn(
+                    r"'_([0-9]+)'", lambda mm: f"'_{prefix}_{mm.group(1)}'", body
+                )
+                renamed_count += cnt
+                out.append(f"{indent}_anonymous_ = ({new_body})")
+                i += 1
+                continue
+
+            m_stmt = stmt_fields_re.match(line)
+            m_inline = inline_fields_re.match(line) if not m_stmt else None
+            if m_stmt or m_inline:
+                cls = m_stmt.group(1) if m_stmt else last_class
+                prefix = sanitize(cls) if cls else None
+                out.append(line)
+                i += 1
+                while i < n and lines[i].strip() != "]":
+                    fl = lines[i]
+                    if prefix:
+                        fl, cnt = key_re.subn(
+                            lambda mm: f"{mm.group(1)}{prefix}_{mm.group(2)}{mm.group(3)}",
+                            fl,
+                        )
+                        renamed_count += cnt
+                    out.append(fl)
+                    i += 1
+                if i < n:
+                    out.append(lines[i])  # closing ']'
+                    i += 1
+                continue
+
+            out.append(line)
+            i += 1
+
+        self.log(f"Renamed {renamed_count} generic field keys")
+        return "\n".join(out)
+
     def cleanup(self):
         """Remove temporary files if not keeping them."""
         if not self.keep_intermediate and self.temp_dir != ".":