view release on metacpan or  search on metacpan

Changes  view on Meta::CPAN

          stop/reap, cleanup-driven stop, forgotten-stop exit reaping) for
          leak-checking under valgrind
        - Added lib/RPi/WiringPi/WORKERS.pod (perldoc RPi::WiringPi::WORKERS),
          the OO concurrency/background-worker guide leading with $pi->worker()
          and demoting raw fork/threads to "under the hood"; cross-linked from
          the worker() method POD and the FAQ
        - Added docs/threads-examples.md, the markdown companion to
          RPi::WiringPi::WORKERS; cross-linked with docs/interrupt-examples.md
        - Implicit cleanup of any object the user never explicitly cleanup()'d
          now runs from an END block, while the IPC::Shareable segment is still
          alive, instead of from DESTROY at global destruction. This makes the
          implicit teardown actually reset the pins and unregister the object,
          and silences the "during global destruction" warnings that were
          emitted when DESTROY operated on a half-torn-down shared segment
        - Added t/03-meta.t coverage asserting the entire meta blob (including
          deeply nested user data) round-trips and stays in a single shared
          memory segment, guarding against IPC::Shareable per-nested-ref
          segment fan-out
        - Retired the broken build_testing/shared_data.pl scratch script and
          updated build_testing/benchmark/sharelite_vs_memfile.pl (renamed to
          shareable_vs_memfile.pl) onto IPC::Shareable, removing the last
          references to the dropped IPC::ShareLite backend
        - Removed t/154-sig_die_multi.t and t/155-sig_die.t: both asserted that a
          caught eval { die } resets pins to INPUT, which the module does not
          provide (cleanup restores each pin's captured original mode, not
          INPUT). Real process-death cleanup remains covered by t/112/113/114
        - Added meta_remove() to RPi::WiringPi::Meta, which frees the underlying
          shared memory segment (and semaphore) entirely rather than only
          emptying it like meta_erase(); a later meta_* call transparently
          recreates a fresh segment. bin/pimetaerase gained an optional fourth
          'remove' argument that removes the segment after erasing
        - Fixed pin cleanup on the Raspberry Pi 5: wiringPi's pinModeAlt() cannot
          set alt mode 31 ("no function", the pristine default for many RP1
          GPIO pins), so cleanup() left those pins in their last-used mode. We
          now fall back to pinctrl for the alt-31 case (gated on alt == 31 and
          pi_rp1_model(), so legacy Pi 3/4 behaviour is unchanged). RPiTest's
          rpi_reset() got the same fallback. Requires gpio-group membership
          (documented in the FAQ)
        - Bumped the RPi::SysInfo prerequisite from 1.00 to 1.01, which fixes
          gpio_info() (now uses pinctrl; raspi-gpio was removed from current Pi
          OS) and raspi_config() (now reads /boot/firmware/config.txt). Updated
          the t/403/404/406/407 sysinfo tests for the corrected output and made
          them board/OS-agnostic, with additional assertions
        - Reworked the crash/signal handling. We no longer trap $SIG{__DIE__}
          (which fired on every die, including a caught eval { die }, silently
          resetting your pins); hardware cleanup on a crash or normal exit is
          handled by the END/DESTROY path instead. $SIG{INT}/$SIG{TERM} are now
          installed correctly (the old form raised "Not a subroutine reference"
          when a signal fired) and chain to any handler you installed first
        - fatal_exit now actually works (it was a no-op): on a trapped INT/TERM
          we reset the hardware, then either re-raise to terminate (default) or,
          with fatal_exit => 0, continue running. POD updated to match
        - Stopped pin cleanup's pinctrl shell-out from clobbering the program's
          exit status (local $?/$! in _restore_pin_alt)
        - Completed the Pi 5/RP1 pin-restore fix: on the RP1, get_alt() returns a
          mode enum, not a classic alt number, so pinModeAlt() could not restore
          INPUT (0) or OUTPUT (1) — only alt 31 was handled before. cleanup() and
          RPiTest's rpi_reset() now restore INPUT/OUTPUT via pinMode() (and "no
          function" via pinctrl), so a pin left in OUTPUT (eg. by a worker) is
          properly returned to its default. t/213-worker.t now asserts BCM18 is
          restored; clears the cascade that failed t/400-407 in a full sweep
        - Made t/107-alt_modes.t board-aware: its "default mode is INPUT" check
          now reads the per-board default (INPUT on Pi 3/4, "no function"/alt 31
          on Pi 5), and the classic ALT0-5 round-trip block is skipped on the
          Pi 5 / RP1, whose funcsel scheme doesn't match it (alt 1/2 don't take).
          Pi 3/4 behaviour is unchanged
        - Made io_led() and pwr_led() board-aware: Pi 5 / RP1 names its LEDs
          ACT/PWR rather than led0/led1, which produced "tee: .../led0: No such
          file or directory" noise and left the LEDs undriven. Resolved via
          pi_rp1_model(); restore triggers are now board-correct (PWR default
          is "none" on Pi 5, "input" on earlier boards)
        - Bumped the RPi::SysInfo prereq to 1.02, which adds the inherited
          pi_model() method (normalized board name, Pi 0-5) plus the camera
          (libcamera), core_temp (thermal-zone) and network_info (ip addr)
          fallbacks for current Raspberry Pi OS
        - Added t/408-sysinfo_pi_model.t covering the inherited pi_model()
        - Documented pi_model() in the SYNOPSIS, method list and FAQ
        - Corrected the FAQ/POD to show config.txt at both /boot/firmware (Bookworm
          and later) and /boot (older), and network_info using ifconfig or ip addr
        - Updated RPiTest's Pi 5 default pin-config for the now-enabled SPI, I2C
          and UART buses: GPIO2/3 (I2C) report funcsel 7, GPIO9/10/11 (SPI)
          report 4, and GPIO14/15 (UART) report 3 and idle high. Previously these
          encoded the bus-disabled values (0/0/31), so rpi_check_pin_status()
          mis-validated once the interfaces were turned on
        - Tightened the GPIO14/15 check in rpi_check_pin_status() and
          rpi_verify_pin_status(): dropped the "alt == 2" escape hatch (a
          serial-disabled workaround) now that the enabled UART reports a
          deterministic funcsel, folding 14/15 into the standard per-board check
        - Fixed rpi_check_pin_status() to emit the "I2C locked ... skipping pins
          2 and 3" note only when the OLED lock file is actually present; it
          previously printed unconditionally
        - Re-enabled t/315-serial.t by removing its blanket skip_all, and removed
          two meta_lock(name => ..., state => ...) calls that used a non-existent
          signature (meta_lock takes a single lock flag) and crashed the test;
          the RPI_SERIAL gate (and TX/RX loopback requirement) still govern it
        - Reconciled the Pi 5 default-pin table and serial/Bluetooth note in
          docs/test-platform/test-pinout-doc.md, the serial POD note in
          RPi::WiringPi, and fixed a "ram=i2c_arm" typo in the FAQ, to match the
          enabled-bus pin values
        - Fixed gpio_layout() in Core infinitely recursing into itself instead
          of dispatching to WiringPi::API's implementation
        - Fixed pwm_mode() getter form croaking; with no param it now returns
          the current mode instead of dying, and only the setter form marks
          PWM as in-use
        - Fixed _pwm_in_use() registration guard using && where every sibling
          guard uses ||; with rpi_register_pins disabled it wrote pwm.users
          meta that cleanup() could never clear
        - cleanup() now removes only the calling object's uuid from a shared
          pin's users and resets/deletes the pin only when no users remain,
          mirroring the PWM teardown (it previously dropped the pin for all
          co-owners)
        - Fixed new() re-running GPIO setup when RPI_PIN_MODE held the falsy
          WPI scheme (0); the env check is now defined-based like Core's
          pin_scheme(), so a second object no longer restamps WPI mode to GPIO
        - pwm_range()/pwm_mode() default return values now use the
          PWM_DEFAULT_RANGE/PWM_DEFAULT_MODE constants instead of the magic
          numbers 1023/1 (resolves the long-standing FIXME)
        - All shared-memory critical sections now run under an internal
          _meta_txn() wrapper that guarantees meta_unlock() runs even if the
          section dies (a die between lock and unlock previously held the
          exclusive lock indefinitely across processes)