chore: cleanup garbage code (#29770)

python/sglang/test/ascend/e2e/test_npu_multi_node_utils.py

@@ -779,7 +779,7 @@ def wait_server_ready(url, timeout=LOCAL_TIMEOUT):
                 logger.info(
                     f"Server {url} returned status code: {response.status_code}"
                 )
-        except Exception as e:
+        except Exception:
             # logger.error(f"Server {url} request error: {e}, retrying...")
             pass
 

python/sglang/test/kits/attention_unittest/attention_methods/dsv4_attention.py

@@ -1104,7 +1104,7 @@ def _seed_c4_if_needed(
     fixture: DSV4AttentionFixture, *, num_entries: int = _DSV4_EXTRA_ENTRIES
 ) -> None:
     """For compress_ratio=4, seed the C4 metadata the exercised path consumes
-    (the C4Indexer would normally populate it; the smoke fixture skips the
+    (the C4Indexer would normally populate it; the compact fixture skips the
     indexer): `c4_sparse_page_indices` for the dense extend path,
     `c4_sparse_raw_indices` for sparse prefill. No-op for other compress_ratios.
     """
@@ -1383,7 +1383,7 @@ def _seed_c4_sparse_indices(
 ) -> None:
     """For compress_ratio=4 the production `init_flashmla_related` initializes
     `c4_sparse_page_indices` to all `-1` (the C4Indexer fills it in later).
-    Since the smoke fixture does not run the indexer, the C4 path attends to
+    Since the compact fixture does not run the indexer, the C4 path attends to
     zero extra entries unless we seed the indices ourselves. Seed each query
     row to point to `[0, 1, ..., num_entries - 1]` so the backend reads the
     same `num_entries` C4 K's that the reference also reads, exercising the
@@ -1614,7 +1614,7 @@ def run_dsv4_compress_attention_case(
     assert case.compress_ratio in (
         4,
         128,
-    ), f"smoke runner requires compress_ratio in (4, 128); got {case.compress_ratio}"
+    ), f"DSV4 compact runner requires compress_ratio in (4, 128); got {case.compress_ratio}"
     if sparse_prefill:
         assert (
             case.forward_mode.is_extend_without_speculative()

python/sglang/test/kits/basic_decode_correctness_kit.py

@@ -2,7 +2,7 @@
 
 Probes that catch the model producing wrong output: weight load
 failure, sampling path bugs, KV / attention corruption, and cuda graph
-edge cases. Single-prompt smoke only -- dataset-driven accuracy gates
+edge cases. Single-prompt coverage only -- dataset-driven accuracy gates
 belong to the consuming test class, not this kit.
 
 Mix into any ``CustomTestCase`` subclass that exposes ``self.base_url``

python/sglang/test/simple_eval_mgsm.py

@@ -175,7 +175,7 @@ class MGSMEval(Eval):
             ]
             try:
                 response_text = sampler(prompt_messages)
-            except Exception as e:
+            except Exception:
                 response_text = ""
 
             answer_prefix = LANG_TO_ANSWER_PREFIX[language]