v1.09.49

OpenADS v1.09.49

Mutex leak fix: dead sessions release their names

Field (v1.09.48, loopback and WAN): OAds_MutexCreate(cMutexName)
failed after the creating process died — only a server restart
cleared it, blocking extensive multi-user testing. Root cause:
Session::cleanup released indexes, tables, files and ABI handles
but never touched the distributed mutex service
(MutexManager::release_all existed but was dead code — and it
only unlocked without destroying, so the name wedged forever).

Changes

• Creator tracking (src/network/mutex_manager.*):
  create records the creating session; new release_session
  unlocks everything the dying session held and destroys every
  name it created — unless a different live session currently
  holds the lock, in which case creation transfers to that
  holder instead of destroying under it. Waiters are woken.
• Teardown hook (src/network/session.cpp): Session::cleanup
  runs release_session first, covering Disconnect, peer-close
  (killed process) and exception paths alike — all funnel through
  it. No protocol change.
• Tests: new network_mutex_release_test — creator dies
  holding the lock (recreate+lock must succeed), and creator dies
  while a peer holds it (peer keeps working; name dies with the
  peer). Verified the test fails without the fix exactly as the
  field reported (MutexCreate → error after creator death).

Test Results

• MinGW-x64 (GCC 16.2): mutex + pool + batching suites green;
  full unit suite 1569/1570 (sole failure pre-existing,
  MinGW-only CDX alloc-tail).
• Full matrix (Windows MSVC x64/x86, Linux, macOS) runs in CI on the
  tag push.

Packages

• openads-1.09.49-windows-x64.zip
• openads-1.09.49-windows-x86.zip
• openads-1.09.49-linux-x64.tar.gz
• openads-1.09.49-macos-universal.tar.gz

Field triage, release engineering and validation: Pritpal Bedi.