Deduplicate rows in table reporter with identical displayed fields

news/857.bugfix.rst

@@ -0,0 +1,3 @@
+Deduplicate rows in the table reporter that had identical displayed fields but
+different underlying call stacks. Also add a ``--split-threads`` option to the
+table reporter, matching the flamegraph reporter's existing option.

src/memray/commands/table.py

@@ -15,6 +15,12 @@ def __init__(self) -> None:
 
     def prepare_parser(self, parser: argparse.ArgumentParser) -> None:
         super().prepare_parser(parser)
+        parser.add_argument(
+            "--split-threads",
+            help="Do not merge allocations across threads",
+            action="store_true",
+            default=False,
+        )
         parser.add_argument(
             "--no-web",
             help="Use local assets instead of fetching from CDN",

src/memray/reporters/table.py

@@ -33,7 +33,10 @@ def from_snapshot(
         native_traces: bool,
         **kwargs: Any,
     ) -> "TableReporter":
-        result = []
+        # Aggregate records that share the same displayed fields, since the
+        # table only shows the top stack frame and records with different full
+        # call stacks can map to the same displayed row.
+        aggregated: Dict[tuple, Dict[str, Any]] = {}
         for record in allocations:
             stack_trace = (
                 list(record.hybrid_stack_trace(max_stacks=1))
@@ -46,17 +49,21 @@ def from_snapshot(
                 stack = f"{function} at {file}:{line}"
 
             allocator = AllocatorType(record.allocator)
-            result.append(
-                {
-                    "tid": format_thread_name(record),
+            tid = format_thread_name(record)
+            key = (tid, allocator.name.lower(), stack)
+            if key in aggregated:
+                aggregated[key]["size"] += record.size
+                aggregated[key]["n_allocations"] += record.n_allocations
+            else:
+                aggregated[key] = {
+                    "tid": tid,
                     "size": record.size,
                     "allocator": allocator.name.lower(),
                     "n_allocations": record.n_allocations,
                     "stack_trace": html.escape(stack),
                 }
-            )
 
-        return cls(result, memory_records=memory_records)
+        return cls(list(aggregated.values()), memory_records=memory_records)
 
     def render(
         self,
@@ -67,8 +74,6 @@ def render(
         inverted: bool,
         no_web: bool = False,
     ) -> None:
-        if not merge_threads:
-            raise NotImplementedError("TableReporter only supports merged threads.")
         if inverted:
             raise NotImplementedError(
                 "TableReporter does not support inverted argument"

tests/integration/test_main.py

@@ -1246,23 +1246,32 @@ def test_reads_from_correct_file(self, tmp_path, simple_test_file):
         assert output_file.exists()
         assert str(source_file) in output_file.read_text()
 
-    def test_no_split_threads(self, tmp_path):
-        # GIVEN/WHEN/THEN
-        with pytest.raises(subprocess.CalledProcessError):
-            subprocess.run(
-                [
-                    sys.executable,
-                    "-m",
-                    "memray",
-                    "table",
-                    "--split-threads",
-                    "somefile",
-                ],
-                cwd=str(tmp_path),
-                check=True,
-                capture_output=True,
-                text=True,
-            )
+    def test_split_threads(self, tmp_path, simple_test_file):
+        # GIVEN
+        results_file, source_file = generate_sample_results(
+            tmp_path, simple_test_file, native=True
+        )
+
+        # WHEN
+        subprocess.run(
+            [
+                sys.executable,
+                "-m",
+                "memray",
+                "table",
+                "--split-threads",
+                str(results_file),
+            ],
+            cwd=str(tmp_path),
+            check=True,
+            capture_output=True,
+            text=True,
+        )
+
+        # THEN
+        output_file = tmp_path / "memray-table-result.html"
+        assert output_file.exists()
+        assert str(source_file) in output_file.read_text()
 
 
 class TestReporterSubCommands: