stdlib: don't clear LogRecord.args before ProcessorFormatter processors run

CHANGELOG.md

@@ -32,6 +32,9 @@ You can find our backwards-compatibility policy [here](https://github.com/hynek/
 - `structlog.BytesLogger`, `structlog.PrintLogger`, and `structlog.WriteLogger` now hold *weak* references to the files they use for output.
   This prevents their leakage in long-running processes that open many logfiles, such as task executors that create a per-task `BytesLogger` or `WriteLogger`.
   [#807](https://github.com/hynek/structlog/pull/807)
+- `structlog.stdlib.ProcessorFormatter` no longer clears `LogRecord.args` before running its processor chain, so processors that inspect `event_dict["_record"].args` can see the original positional arguments.
+  `args` is still cleared before the record is handed to the underlying `logging.Formatter`.
+  [#771](https://github.com/hynek/structlog/issues/771)
 
 
 ### Changed

src/structlog/stdlib.py

@@ -1141,8 +1141,6 @@ def format(self, record: logging.LogRecord) -> str:
             if self.pass_foreign_args:
                 ed["positional_args"] = record.args
 
-            record.args = ()
-
             # Add stack-related attributes to the event dict
             if record.exc_info:
                 ed["exc_info"] = record.exc_info
@@ -1176,6 +1174,12 @@ def format(self, record: logging.LogRecord) -> str:
             )
             ed = cast(str, ed)
 
+        # Clear args after processors have run so user processors can access
+        # ``record.args`` via ``_record``. We render our log records before
+        # sending them back to logging, so ``LogRecord.args`` must be cleared,
+        # otherwise stdlib's formatter would raise ``TypeError: not all
+        # arguments converted during string formatting``.
+        record.args = ()
         record.msg = ed
 
         return super().format(record)

tests/test_stdlib.py

@@ -1105,6 +1105,23 @@ def test_pass_foreign_args_true_sets_positional_args_key(self):
         assert "positional_args" in event_dict
         assert positional_args == event_dict["positional_args"]
 
+    def test_record_args_accessible_in_processor(self):
+        """
+        Processors can access ``record.args`` via ``_record``: args are not
+        cleared on the record before the formatter's processors run.
+        """
+        seen = []
+
+        def capture_args(_, __, event_dict):
+            seen.append(event_dict["_record"].args)
+            return event_dict
+
+        configure_logging((capture_args,))
+
+        logging.getLogger().info("test a=%s b=%s", "a", "b")
+
+        assert [("a", "b")] == seen
+
     def test_log_dict(self, capsys):
         """
         dicts can be logged with std library loggers.