[CI] Fail lint when a registered file's TestCase classes never run (#32735)

scripts/ci/check_registered_tests.py

@@ -21,6 +21,7 @@ Reuses ut_parse_one_file() from ci_register.py (AST-based parsing)
 to match the same logic used by run_suite.py's collect_tests().
 """
 
+import ast
 import glob
 import importlib.util
 import os
@@ -39,6 +40,37 @@ _MODERN_SHAPE = re.compile(r"^(.+)-test-(.+)$")
 _LEGACY_CUDA_PREFIXES = ("nightly", "stress", "weekly")
 
 
+def _defines_testcase(tree: ast.AST) -> bool:
+    """True if the file defines unittest classes, statically or via type()."""
+    for node in ast.walk(tree):
+        if isinstance(node, ast.ClassDef):
+            if any("TestCase" in ast.unparse(b) for b in node.bases):
+                return True
+        elif isinstance(node, ast.Call):
+            if (
+                isinstance(node.func, ast.Name)
+                and node.func.id == "type"
+                and len(node.args) >= 2
+                and isinstance(node.args[1], ast.Tuple)
+                and any("TestCase" in ast.unparse(e) for e in node.args[1].elts)
+            ):
+                return True
+    return False
+
+
+def _main_runs_tests(tree: ast.Module) -> bool:
+    for stmt in tree.body:
+        if not (
+            isinstance(stmt, ast.If)
+            and ast.unparse(stmt.test).replace("'", '"') == '__name__ == "__main__"'
+        ):
+            continue
+        body = ast.unparse(ast.Module(body=stmt.body, type_ignores=[]))
+        if "unittest.main" in body or "pytest.main" in body:
+            return True
+    return False
+
+
 def main() -> int:
     # Import ci_register directly to avoid pulling in all of sglang
     spec = importlib.util.spec_from_file_location(
@@ -61,6 +93,7 @@ def main() -> int:
     missing = []
     legacy_shape = []  # (file, suite, stage, runner_config) -- has a -test- split
     non_dispatchable = []  # (file, suite) -- legacy CUDA suite no workflow invokes
+    dead_tests = []  # (file) -- TestCase classes that `python3 file.py` never runs
     for f in files:
         try:
             registries, _has_main_entry = ci_register.ut_parse_one_file(f)
@@ -70,6 +103,12 @@ def main() -> int:
         if len(registries) == 0:
             missing.append(f)
             continue
+        # TestCase classes are dead unless __main__ runs them (CI does
+        # `python3 file.py`); the ERROR text below explains the fix.
+        with open(f, "r", encoding="utf-8") as fh:
+            tree = ast.parse(fh.read(), filename=f)
+        if _defines_testcase(tree) and not _main_runs_tests(tree):
+            dead_tests.append(f)
         for r in registries:
             # Pure legacy form on a CUDA registry: suite set, stage/runner unset.
             if not (
@@ -127,6 +166,19 @@ def main() -> int:
             )
         print()
         exit_code = 1
+    if dead_tests:
+        print(
+            "ERROR: Test file(s) define TestCase classes that CI never runs: "
+            "the registered file is executed as `python3 file.py`, but its "
+            '`if __name__ == "__main__"` block does not call unittest.main() '
+            "or pytest.main(), so the classes are silently skipped while the "
+            "file reports success. Make __main__ run the tests (put any CLI "
+            "entry point behind an explicit flag):\n"
+        )
+        for f in dead_tests:
+            print(f"  {f}")
+        print()
+        exit_code = 1
 
     return exit_code
 

test/registered/unit/mem_cache/test_unified_radix_cache_bench.py

@@ -1,7 +1,7 @@
 """Large-scale benchmark + fuzz correctness tests for UnifiedRadixCache.
 
 Usage (standalone):
-    bench: python3 test/registered/unit/mem_cache/test_unified_radix_cache_bench.py --num-seqs 5000 --verify --components mamba legacy-mamba swa legacy-swa
+    bench: python3 test/registered/unit/mem_cache/test_unified_radix_cache_bench.py --bench --num-seqs 5000 --verify --components mamba legacy-mamba swa legacy-swa
     CI Test: python -m pytest test/registered/unit/mem_cache/test_unified_radix_cache_bench.py -v -s
 """
 
@@ -10,6 +10,7 @@ import gc
 import logging
 import random
 import statistics
+import sys
 import time
 import unittest
 from array import array
@@ -239,7 +240,7 @@ def create_bench_cache(
     _rid = [0]
 
     def make_req():
-        from sglang.srt.managers.schedule_batch import Req
+        from sglang.srt.managers.schedule_batch import Req, ReqKvInfo
         from sglang.srt.sampling.sampling_params import SamplingParams
 
         req = Req(
@@ -250,6 +251,8 @@ def create_bench_cache(
         )
         _rid[0] += 1
         req_to_token_pool.alloc([req])
+        # fabricated reqs bypass alloc_for_extend, the normal creator of req.kv
+        req.kv = ReqKvInfo(kv_allocated_len=0, swa_evicted_seqlen=0)
         return req
 
     return tree, allocator, req_to_token_pool, make_req
@@ -821,7 +824,8 @@ _TREE_CONFIGS = {
     "legacy-swa": ((ComponentType.FULL, ComponentType.SWA), SWARadixCache),
 }
 
-if __name__ == "__main__":
+
+def _run_bench_cli():
     parser = argparse.ArgumentParser(description="UnifiedRadixCache benchmark")
     parser.add_argument("--num-seqs", type=int, default=5000)
     parser.add_argument("--chunk-len", type=int, default=256)
@@ -857,3 +861,12 @@ if __name__ == "__main__":
             tree_cls=tree_cls,
             page_size=args.page_size,
         )
+
+
+if __name__ == "__main__":
+    # CI runs `python3 file.py`; it must execute the TestBench_* classes
+    if "--bench" in sys.argv:
+        sys.argv.remove("--bench")
+        _run_bench_cli()
+    else:
+        unittest.main()