view release on metacpan or  search on metacpan

Changes  view on Meta::CPAN

    - shared_scalar() no longer stores tied refs inside %events;
      %events now holds an arrayref of hex key strings instead,
      eliminating a same-process FETCH deadlock in IPC::Shareable
    - $SIG{__WARN__} moved to local inside _event() so it no longer
      silently replaces the caller's handler at module load; $SIG{CHLD}
      remains at file scope (children outlive _event(), need auto-reap)
    - Add write-path tests and automated source audit to t/09-locking.t
    - Extract _run_callback() from _event() to deduplicate the eval +
      error-capture + store pattern and isolate $@ preservation
    - Add _end() clean_up_protected coverage test; fix _end() mock-counter
      test that was silently destroying the semaphore set, causing 11
      subsequent tests to be skipped; fix events() deep-copy still treating
      shared_scalars as a hashref instead of the new arrayref
    - Fix fork failure silently falling through to child path and running
      the callback in the parent process; now croaks with "fork() failed"
    - Wrap _run_callback in eval inside _event() and always call
      _pm->finish($@ ? 1 : 0) so ForkManager isn't left with a stale
      child record when the callback dies
    - Guard _end() with a local SIGALRM (END_LOCK_TIMEOUT, default 2s)
      so process exit doesn't block forever on _events_read' LOCK_SH if
      a peer still holds LOCK_EX on the events knot
    - stop() now polls kill(0) at STOP_KILL_POLL_INTERVAL (0.05s) for up
      to STOP_KILL_TIMEOUT (1s) instead of always sleeping a fixed 1s;
      returns as soon as the target process is gone (full test suite
      wallclock roughly halved)
    - Refactor error()/status() to remove their mutual-recursion
      side-effect chain. Crash detection now lives in a private
      _detect_crash helper that both methods call independently
    - Retire the undocumented -99 PID sentinel that marked crashes
    - stop() now sends SIGTERM first (STOP_TERM_TIMEOUT 0.5s) and escalates
      to SIGKILL only if the child is still alive; _signal_and_wait() helper
      encapsulates the signal-and-poll logic
    - Replace mutable module-level $is_child_process flag with
      $$ != $creator_pid check in DESTROY; a forked child inherently has
      a different PID than the process that loaded the module, so no
      mutable state that mock-fork tests can corrupt is needed
    - Replace lexical $id counter with shared _id_counter in %events to
      prevent duplicate IDs across forked processes
    - Closes #15; Add _stop_requested cooperative flag in shared %events so the
      child's interval loop can break cleanly and call finish() instead of being
      killed by SIGTERM; stop() sets it, start() clears it, the loop checks it
      on each iteration
    - Closes #14; Add timeout() accessor so a callback that exceeds the
      specified number of whole seconds self-terminates with an error.
      Accepts a non-negative integer or undef; fractional seconds are
      not supported
    - Closes #9; Add immediate() method. Set a flag to have the event fire
      immediately on start rather than waiting for the first interval to be
      reached
    - Adopt IPC::Shareable's testing_set/clean_up_testing API to brand all
      test-suite segments and purge orphans from prior crashed runs; bump
      IPC::Shareable prereq to 1.17
    - Fixes and enhancements to the local VM test infrastructure
    - Pin Parallel::ForkManager to < 2.00 on Perl < 5.14 to avoid Moo XS
      compile failures on macOS
    - Guard DESTROY's _events_write in eval and clear stale _lock on
      failure to prevent EAGAIN from IPC::Shareable's unlock() leaking
      IPC segments on Linux
    - Ensure counters are properly cleaned up in _end() cleanup
    - Use Time::HiRes wall-clock time in _signal_and_wait so stop() timeout
      is accurate on busy/VM systems where select() jitter accumulates
    - Relax t/15-interval.t exact-count timing checks to >= so slow macOS VMs
      don't fail before the first callback fires
    - _detect_crash no longer marks cleanly-exited one-shot events as
      crashed; child writes _clean_exit flag to shared %events before
      finish(0) so the parent can distinguish normal completion from a crash
    - Fix flaky t/90 on macOS CI by replacing nested-hash shared memory
      writes with flat keys, eliminating child-segment race between forks
    - Fix shared memory leak on SIGINT/SIGTERM: install signal handlers
      that run _end() cleanup before re-raising; _end() now unconditionally
      stops children and removes protected segments instead of skipping
      when _event_count > 0
    - Fix _end() SIGALRM deadlock: clear SA_RESTART via POSIX::sigaction so
      the alarm actually interrupts blocked semop(); split the single 2s alarm
      into per-phase evals so each cleanup step runs even when an earlier one
      times out; add @all_pids fallback so children are killed even when the
      %events read-lock is stuck
    - Add 'error' and 'waiting' fields to events() and info() snapshots
    - Add wait() method that polls until the event is dormant (optional
      poll interval, default 0.01s)

1.13    2024-03-04
    - Added ability to send in per-callback call parameters via the start()
      method (closes #10)
    - Add prereq of Test::SharedFork to keep tests in order in t/46
    - Bumped prereq version of IPC::Shareable to 1.13

1.12    2022-03-31
    - Only remove the Async::Event::Interval protected shared %events hash
      if it doesn't contain keys (ie. event objects). When running in things
      like 'plackup', END{} was being called on each browser session close,
      but there were still objects, so we were trashing the %events
      infrastructure prematurely
    - Each unit test file now does pre-and-post segment counts and displays
      them if PRINT_SEGS env var is true, and the suite itself has a before
      and after count comparison to ensure the suite leaked no segments or
      semaphores
    - Fix issue where we were trying to set a _pid() on an undefined value,
      causing IPC::Shareable to complain that "Can't STORE on undef val"

1.11    2022-03-09
    - Modified tests to run only on valid CI platforms

1.10    2022-03-07
    - Add tests to ensure that all shared memory segments created during the
      unit test suite run get cleaned up ok
    - Bump prereq of IPC::Shareable to 1.11 due to bug fix where child shared
      memory segments created under one process weren't being registered if the
      parent was created in a different process

1.09    2022-03-05
    - Bump prereq IPC::Shareable to 1.08 due to a fix in creating random SHM
      keys in forked environments, major improvements of its _shm_key_rand()
      function, and its ability to set a 'protected' option so our global shared
      %events hash doesn't get cleaned up automatically
    - Modified examples/shared.pl so it doesn't leak segments and semaphores
      after completion
    - We now remove %events hash in END instead of DESTROY. This way we can have
      objects go out of scope in a script without blowing away the global shared
      data
    - Added _end() so that we can test the END block in unit tests. END{} calls