[API Compatibility No.218] Update select_scatter to ChangePrefixMatcher

paconvert/api_mapping.json

@@ -2271,18 +2271,7 @@
     ]
   },
   "torch.Tensor.select_scatter": {
-    "Matcher": "GenericMatcher",
-    "paddle_api": "paddle.Tensor.select_scatter",
-    "min_input_args": 3,
-    "args_list": [
-      "src",
-      "dim",
-      "index"
-    ],
-    "kwargs_change": {
-      "src": "values",
-      "dim": "axis"
-    }
+    "Matcher": "ChangePrefixMatcher"
   },
   "torch.Tensor.set_": {
     "Matcher": "TensorSetMatcher",
@@ -11127,20 +11116,7 @@
     ]
   },
   "torch.select_scatter": {
-    "Matcher": "GenericMatcher",
-    "paddle_api": "paddle.select_scatter",
-    "min_input_args": 4,
-    "args_list": [
-      "input",
-      "src",
-      "dim",
-      "index"
-    ],
-    "kwargs_change": {
-      "input": "x",
-      "src": "values",
-      "dim": "axis"
-    }
+    "Matcher": "ChangePrefixMatcher"
   },
   "torch.selu": {
     "Matcher": "GenericMatcher",

tests/test_Tensor_select_scatter.py

@@ -49,7 +49,7 @@ def test_case_3():
         import torch
         input = torch.zeros(2, 2)
         src = torch.ones(2)
-        result = input.select_scatter(src, dim=1, index=1)
+        result = input.select_scatter(src=src, dim=1, index=1)
         """
     )
     obj.run(pytorch_code, ["result"])
@@ -61,7 +61,7 @@ def test_case_4():
         import torch
         input = torch.zeros(2, 2)
         src = torch.ones(2)
-        result = input.select_scatter(src=src, dim=1, index=1)
+        result = input.select_scatter(src, 1, 1)
         """
     )
     obj.run(pytorch_code, ["result"])

tests/test_select_scatter.py

@@ -49,7 +49,7 @@ def test_case_3():
         import torch
         x = torch.zeros((2,3,4)).type(torch.float32)
         values = torch.ones((2,4)).type(torch.float32)
-        result = torch.select_scatter(input=x, dim=1, src=values, index=1)
+        result = torch.select_scatter(input=x, src=values, dim=1, index=1)
         """
     )
     obj.run(pytorch_code, ["result"])
@@ -61,7 +61,7 @@ def test_case_4():
         import torch
         x = torch.zeros((2,3,4)).type(torch.float32)
         values = torch.ones((2,4)).type(torch.float32)
-        result = torch.select_scatter(input=x, src=values, dim=1, index=1)
+        result = torch.select_scatter(x, values, 1, 1)
         """
     )
     obj.run(pytorch_code, ["result"])
@@ -73,7 +73,7 @@ def test_case_5():
         import torch
         x = torch.zeros((2,3,4)).type(torch.float32)
         values = torch.ones((2,4)).type(torch.float32)
-        result = torch.select_scatter(x, values, 1, 1)
+        result = torch.select_scatter(index=1, dim=1, src=values, input=x)
         """
     )
     obj.run(pytorch_code, ["result"])

tools/validate_unittest/validate_unittest.py

@@ -14,6 +14,8 @@
 #
 
 import argparse
+import importlib
+import inspect
 import json
 import os
 import re
@@ -145,6 +147,149 @@
 }
 
 
+# ChangePrefixMatcher removes explicit args_list metadata, so builtin APIs that
+# are not introspectable still need a validator-side fallback.
+change_prefix_validator_metadata_fallback = {
+    "torch.select_scatter": {
+        "args_list": ["input", "src", "dim", "index"],
+        "min_input_args": 4,
+    },
+    "torch.Tensor.select_scatter": {
+        "args_list": ["src", "dim", "index"],
+        "min_input_args": 3,
+    },
+}
+
+
+def isclassname(name, module_parts):
+    if name and name[0].isupper():
+        return True
+    elif (
+        name == "profile" and len(module_parts) >= 1 and module_parts[-1] == "profiler"
+    ):
+        return True
+    return False
+
+
+def resolve_torch_callable(function_string):
+    try:
+        parts = function_string.split(".")
+        function_name = parts.pop()
+        classname = None
+        if len(parts) >= 1 and isclassname(parts[-1], parts[:-1]):
+            classname = parts.pop()
+
+        if len(parts) > 0:
+            module_name = ".".join(parts)
+            module = importlib.import_module(module_name)
+        else:
+            module = globals()
+
+        if classname is not None:
+            module = getattr(module, classname)
+
+        if not hasattr(module, function_name):
+            return None
+
+        return getattr(module, function_name)
+    except Exception:
+        return None
+
+
+def build_args_list_from_signature(api_target, signature):
+    args_list = []
+    need_positional_separator = False
+    need_keyword_separator = True
+
+    for param in signature.parameters.values():
+        if param.name == "self" and api_target.startswith("torch.Tensor."):
+            continue
+
+        if param.kind == inspect.Parameter.POSITIONAL_ONLY:
+            args_list.append(param.name)
+            need_positional_separator = True
+        elif param.kind == inspect.Parameter.POSITIONAL_OR_KEYWORD:
+            if need_positional_separator:
+                args_list.append("/")
+                need_positional_separator = False
+            args_list.append(param.name)
+        elif param.kind == inspect.Parameter.VAR_POSITIONAL:
+            if need_positional_separator:
+                args_list.append("/")
+                need_positional_separator = False
+            args_list.append("*" + param.name)
+            need_keyword_separator = False
+        elif param.kind == inspect.Parameter.KEYWORD_ONLY:
+            if need_positional_separator:
+                args_list.append("/")
+                need_positional_separator = False
+            if need_keyword_separator:
+                args_list.append("*")
+                need_keyword_separator = False
+            args_list.append(param.name)
+        elif param.kind == inspect.Parameter.VAR_KEYWORD:
+            if need_positional_separator:
+                args_list.append("/")
+                need_positional_separator = False
+            args_list.append("**" + param.name)
+
+    if need_positional_separator:
+        args_list.append("/")
+
+    return args_list
+
+
+def infer_min_input_args(api_target, signature):
+    min_input_args = 0
+    for param in signature.parameters.values():
+        if param.name == "self" and api_target.startswith("torch.Tensor."):
+            continue
+        if (
+            param.kind
+            in (
+                inspect.Parameter.POSITIONAL_ONLY,
+                inspect.Parameter.POSITIONAL_OR_KEYWORD,
+            )
+            and param.default == inspect.Parameter.empty
+        ):
+            min_input_args += 1
+    return min_input_args
+
+
+def resolve_validator_metadata(api_target, mapping_data):
+    min_input_args = mapping_data.get("min_input_args")
+    args_list_full = mapping_data.get("args_list")
+
+    if (
+        mapping_data.get("Matcher") != "ChangePrefixMatcher"
+        or "args_list" in mapping_data
+    ):
+        if min_input_args is None:
+            min_input_args = -1
+        if args_list_full is None:
+            args_list_full = []
+        return min_input_args, args_list_full
+
+    pytorch_callable = resolve_torch_callable(api_target)
+    if pytorch_callable is not None:
+        try:
+            signature = inspect.signature(pytorch_callable)
+            if args_list_full is None:
+                args_list_full = build_args_list_from_signature(api_target, signature)
+            if min_input_args is None:
+                min_input_args = infer_min_input_args(api_target, signature)
+        except (TypeError, ValueError):
+            pass
+
+    fallback_metadata = change_prefix_validator_metadata_fallback.get(api_target, {})
+    if args_list_full is None:
+        args_list_full = fallback_metadata.get("args_list", [])
+    if min_input_args is None:
+        min_input_args = fallback_metadata.get("min_input_args", -1)
+
+    return min_input_args, args_list_full
+
+
 def get_test_cases(discovery_paths=["tests"]):
     # Collect the test cases
     monkeypatch = pytest.MonkeyPatch()
@@ -364,11 +509,11 @@ def check_call_variety(test_data, api_mapping, *, api_alias={}, verbose=True):
 
         position_args_checkable = True
 
-        min_input_args = mapping_data.get("min_input_args", -1)
+        min_input_args, args_list_full = resolve_validator_metadata(
+            api_target, mapping_data
+        )
         aux_detailed_data_api["min_input_args"] = min_input_args
 
-        args_list_full = mapping_data.get("args_list", [])
-
         var_arg_name = None
         var_kwarg_name = None
 
@@ -444,7 +589,7 @@ def check_call_variety(test_data, api_mapping, *, api_alias={}, verbose=True):
         all_args = False
         all_kwargs = False
         not_subsequence = False
-        all_default = False if "min_input_args" in mapping_data else None
+        all_default = False if min_input_args >= 0 else None
 
         for case_name, code in unittest_data["code"].items():
             try: