reextract: reclaim RSS with malloc_trim + gc.collect every 50 activities

CPython's allocator holds freed memory in arenas and doesn't return it to
the OS, causing RSS to grow throughout the 2015-activity loop even when
each iteration's objects are freed. Call gc.collect() + malloc_trim(0)
every 50 activities to return freed pages to the kernel and keep RSS flat.

bincio/reextract_cmd.py

@@ -2,10 +2,11 @@
 
 from __future__ import annotations
 
+import ctypes
+import gc
 import json
 import sys
 from pathlib import Path
-from typing import Optional
 
 import click
 
@@ -15,6 +16,19 @@ def _emit(obj: dict) -> None:
     print(json.dumps(obj), flush=True)
 
 
+# On Linux, malloc_trim(0) returns freed arenas to the OS, keeping RSS low.
+# CPython's allocator otherwise holds onto freed memory indefinitely.
+try:
+    _libc = ctypes.CDLL("libc.so.6")
+    def _trim_heap() -> None:
+        _libc.malloc_trim(0)
+except Exception:
+    def _trim_heap() -> None:  # type: ignore[misc]
+        pass
+
+_GC_EVERY = 50  # call gc.collect() + malloc_trim every N activities
+
+
 @click.command("reextract-originals")
 @click.option("--data-dir", required=True, type=click.Path(), help="BAS data directory")
 @click.option("--handle", required=True, help="User handle to re-extract for")
@@ -83,13 +97,27 @@ def reextract_originals(data_dir: str, handle: str, force: bool) -> None:
                 imported += 1
                 _emit({"type": "progress", "n": n, "total": total, "name": name, "status": "imported"})
 
-                # Explicitly free large objects to keep memory low
+            # Explicitly free large objects; also free the raw JSON dict and streams
+            raw = meta = streams = None  # type: ignore[assignment]
+            try:
                 del parsed, metrics
+            except NameError:
+                pass
+
         except Exception as exc:
             errors += 1
             _emit({"type": "progress", "n": n, "total": total, "name": orig_path.stem,
                    "status": "error", "detail": str(exc)})
 
+        # Periodically reclaim freed memory from CPython's allocator arena
+        if n % _GC_EVERY == 0:
+            gc.collect()
+            _trim_heap()
+
+    # Final cleanup before the index write (which loads all summaries at once)
+    gc.collect()
+    _trim_heap()
+
     if imported > 0:
         _emit({"type": "status", "message": "Writing index…"})
         try: