[Dan Cristoloveanu (dcristoloveanu)]

Adds robust_semaphore (win32), a named cross-process counting semaphore whose units are given back when the holding process dies.

A Windows semaphore is not owned by anyone, so a unit taken by a process that crashes is lost until the semaphore is recreated. robust_semaphore uses N named mutexes <name>_<i> instead: a mutex is owned by a thread and is released (abandoned) by the OS when that thread or process dies.

API

• robust_semaphore_create(name, max_count, security_descriptor): opens the N mutexes (CreateMutexExA, optional SDDL) and starts one worker thread for the object. max_count is 1..63.
• robust_semaphore_acquire_async(h, timeout_ms, max_hold_ms, on_acquire_complete, context): one acquisition at a time per object. The callback always runs on the worker thread (never inline) with OK + unit index, TIMEOUT, CANCELLED or ERROR.
• robust_semaphore_cancel(h): stops a waiting acquisition.
• robust_semaphore_release(h): ends the hold (or stops a waiting acquisition).
• robust_semaphore_destroy(h): releases, stops and joins the worker. It can be called from on_acquire_complete, in which case the worker frees the object.
• max_hold_ms is counted from acquisition and includes the callback, so a unit is never held longer than that.

Tests

• 45 unit tests (VLD clean).
• 14 integration tests:
  • at most max_count holders; separate names don't share units;
  • SDDL; cancel; timeout; max hold, including an overrunning callback;
  • abandoned mutex, and a thread that exits while holding a unit;
  • a holder process (robust_semaphore_tester) that is killed;
  • destroy while waiting, while holding, and from the callback.
• Design and requirements: win32/devdoc/robust_semaphore_requirements.md.

[Dan Cristoloveanu (dcristoloveanu)]

/azp run

[azure-pipelines]

Azure Pipelines:
There may be pipelines that require an authorized user to comment /azp run to run.

[azure-pipelines]

Azure Pipelines:
Successfully started running 1 pipeline(s).

[Dan Cristoloveanu (dcristoloveanu)]

/azp run

[azure-pipelines]

Azure Pipelines:
Successfully started running 1 pipeline(s).