[Frontend] Remove deprecated -O.xx flag (#29991)

Signed-off-by: Yanan Cao <gmagogsfm@gmail.com>

docs/design/debug_vllm_compile.md

@@ -86,7 +86,7 @@ LLM(model, enforce_eager=True)
 ```
 
 To turn off just torch.compile, pass `mode = NONE` to the compilation config.
-(`-cc` is short for `--compilation_config`; `-O.*` dotted syntax is deprecated):
+(`-cc` is short for `--compilation_config`):
 
 ```sh
 # Online

tests/utils_/test_argparse_utils.py

@@ -460,23 +460,20 @@ def test_flat_product():
     ]
 
 
-def test_o_legacy_syntax_deprecation(caplog_vllm):
-    """Test that -O.* dotted syntax emits warnings and converts correctly to -cc syntax."""
+def test_o_dotted_syntax_error():
+    """Test that -O.* dotted syntax raises a clear error message."""
     parser = FlexibleArgumentParser()
     parser.add_argument("-cc", "--compilation-config", type=json.loads)
 
-    # Test that -O.backend gets converted correctly AND emits warning
-    args = parser.parse_args(["-O.backend=eager"])
-    assert args.compilation_config == {"backend": "eager"}
+    # Test that -O.* syntax raises a clear ValueError
+    with pytest.raises(ValueError, match=r"The -O\.\* syntax is no longer supported"):
+        parser.parse_args(["-O.backend=eager"])
 
-    # Check that deprecation warning was logged
-    assert len(caplog_vllm.records) >= 1
-    assert (
-        "The -O.* dotted syntax for --compilation-config is deprecated"
-        in caplog_vllm.text
-    )
+    with pytest.raises(ValueError, match=r"Please use -cc\.\* instead"):
+        parser.parse_args(["-O.mode=2"])
 
-    # Test that -O.mode gets converted correctly
-    # Note: warning_once won't emit again in same session
-    args = parser.parse_args(["-O.mode=2"])
-    assert args.compilation_config == {"mode": 2}
+    with pytest.raises(
+        ValueError,
+        match=r"replace '-O\.cudagraph_mode=NONE' with '-cc\.cudagraph_mode=NONE'",
+    ):
+        parser.parse_args(["-O.cudagraph_mode=NONE"])

vllm/utils/argparse_utils.py

@@ -244,9 +244,15 @@ class FlexibleArgumentParser(ArgumentParser):
                 else:
                     key = pattern.sub(repl, arg, count=1)
                     processed_args.append(key)
-            elif arg.startswith("-O") and arg != "-O" and arg[2] != ".":
+            elif arg.startswith("-O."):
+                # Provide clear error for deprecated -O.* syntax
+                raise ValueError(
+                    f"The -O.* syntax is no longer supported. "
+                    f"Please use -cc.* instead. "
+                    f"For example, replace '{arg}' with '{arg.replace('-O', '-cc', 1)}'"
+                )
+            elif arg.startswith("-O") and arg != "-O":
                 # allow -O flag to be used without space, e.g. -O3 or -Odecode
-                # -O.<...> handled later
                 # also handle -O=<optimization_level> here
                 optimization_level = arg[3:] if arg[2] == "=" else arg[2:]
                 processed_args += ["--optimization-level", optimization_level]
@@ -257,17 +263,6 @@ class FlexibleArgumentParser(ArgumentParser):
             ):
                 # Convert -O <n> to --optimization-level <n>
                 processed_args.append("--optimization-level")
-            elif arg.startswith("-O."):
-                # Handle -O.* dotted syntax - ALL dotted syntax is deprecated
-                logger.warning_once(
-                    "The -O.* dotted syntax for --compilation-config is "
-                    "deprecated and will be removed in v0.13.0 or v1.0.0"
-                    ", whichever is earlier.  Please use -cc.* instead. "
-                    "Example: -cc.backend=eager instead of "
-                    "-O.backend=eager."
-                )
-                converted_arg = arg.replace("-O", "-cc", 1)
-                processed_args.append(converted_arg)
             else:
                 processed_args.append(arg)