v0.11.0

• New features
  • Add Nerves.Runtime.serial_number/0. It will call out to the underlying
    system to return the device's serial number however it's stored.
  • Add a fallback to haveged for systems that don't have hardware random
    number generators or otherwise can't use rngd.

v0.10.3

• Bug fixes
  • Fix potential process accumulation from the kmsg_tailer process ignoring
    stdin being closed on it.
  • Tightened deps to avoid combinations that would be difficult to support

v0.10.2

• Bug fixes

  • Fix off-by-one error when processing uevent messages with device paths
    longer than 16 segments. This also bumps the max number of segments to 32.
  • Fix logger message about rngd failing when it was successful
  • Log errors when required commands aren't available rather than raising. It
    turned out that raising was disabling logging and that was making it hard to
    figure out the root cause.

• Enhancements

  • Switch from parsing /proc/kmsg to /dev/kmsg for kernel messages. The
    device interface supplies a little more information and is unaffected by
    other programs reading from it. This change refactored syslog/kmsg parsing
    to improve test coverage. This is considered to be an internal API. If you
    were using it, you will need to update your code.

v0.10.1

• Bug fixes
  • Fix exception on init when mounting the application data partition.
    This addresses an issue where :nerves_runtime would exit due to an
    unmatched call to System.cmd/3.

v0.10.0

• New feature
  • Added Nerves.Runtime.KV.put and Nerves.Runtime.KV.put_active to support
    setting environment. This removes the need to run fw_setenv and also
    updates the cached key/value pairs. Thanks to Troels Brødsgaard for
    implementing.

v0.9.5

• Bug fixes
  • Fix C compiler error (PATH_MAX undeclared) on x86_64/muslc systems

v0.9.4

• Enhancements
  • Reduced number of syscalls needed for transferring device detection events
    (uevents) to Elixir by batching them
  • Improved start up performance by moving initial device enumeration to C
  • Handle uevent overloads by dropping messages rather than crashing. Uevent
    bursts are handled better by the new batching, but can be more severe due to
    faster enumeration. Both dropping and crashing have drawbacks, but dropping
    made it possible to recover on a 32-processor machine with many peripherals.

v0.9.3

• Enhancements
  • Move C build products to under _build

v0.9.2

• Enhancements
  • Support disabling SystemRegistry integration. This is not a recommended
    setting and should only be used on slow devices that don't need device
    insertion/removal notifications.

v0.9.1

• Enhancements
  • Filter out synth_uuid from uevent reports since it's not supported by the
    current uevent handling code.
  • Further reduce garbage produced by processing uevent reports