- Fix more directories not owned by the package (boo#1187546)
- Changes for 24.0.3:
  * erts: A call to erlang:cancel_timer(_, [{info, false}]) could
    cause the calling process to block forever in the call. Note
    that only the synchronous version of the call (that is, the
    async option is false) in combination with the info option set
    to false was effected by this bug.
  * erts: Microstate accounting (msacc) and os:perf_counter()
    unintentionally used system time instead of monotonic time for
    time measurements on a lot of systems. These systems were all
    non x86/x86_64 systems or x86/x86_64 systems without a reliable
    and constant rdtsc instruction. The lock counting (lcnt) built
    runtime system also unintentionally used system time instead of
    monotonic time for time measurements on all systems.
  * erts: Simultaneous calls to
    erlang:system_flag(schedulers_online, _) could cause callers to
    end up in a suspended state forever.
  * dialyzer: Do not expose line number 0 in messages if there are
    other locations to use.
  * dialyzer: In rare circumstances, Dialyzer could crash analyzing
    code with a list comprehension whose value was ignored. (Thanks
    to Ulf Wiger for reporting this bug.)
  * ssh: Filter out sensitive data (passwords etc) from progress
    reports and supervisor reports.
  * inets: Improved user input handling in inets/mod_esi preventing
    unnecessary atom creation.
  * compiler: A compiler optimization pass could crash when given
    odd but legal code using throw/1.
- Changes for 24.0.2:
  * kernel: Fix a race condition in Global.
  * kernel: After a node restart with init:restart/0,1, the module
    socket was not usable because supporting tables had been
    cleared and not re-initialized. This has now been fixed.
    Handling of the "." domain as a search domain was incorrect and
    caused a crash in the DNS resolver inet_res, which has now been
    fixed.
  * kernel: Handling of combinations of the fd option and binding
    to an address has been corrected, especially for the local
    address family.
  * kernel: Bug fixes and code cleanup for the new socket
    implementation, such as: Assertions on the result of
    demonitoring has been added in the NIF code, where appropriate.
    Internal state handling for socket close in the NIF code has
    been reviewed. Looping over close() for EINTR in the NIF code
    has been removed, since it is strongly discouraged on Linux and
    Posix is not clear about if it is allowed. The inet_backend
    temporary socket option for legacy gen_tcp sockets has been
    documented. The return value from net:getaddrinfo/2 has been
    corrected: the protocol field is now an atom(), instead of,
    incorrectly, list(atom()). The documentation has also been
    corrected about this return type. Deferred close of a
    socket:sendfile/* file was broken and has been corrected. Some
    debug code, not enabled by default, in the socket NIF has been
    corrected to not accidentally core dump for debug printouts of
    more or less innocent events.
  * stdlib: Fix a bug that could cause a loop when formatting terms
    using the control sequences p or P and limiting the output with
    the option chars_limit.
  * erts: Not yet handled alias-message signals in the signal queue
    at the time when a garbage collection was performed could cause
    a memory corruption which in turn could result in a crash of
    the runtime system. This bug was introduced in OTP 24.0.
  * erts: Fixed bug when using external pids/ports in keys of big
    maps (> 32). Could cause runtime crash. Bug exists since OTP
    24.0.
  * erts: After a node restart with init:restart/0,1, the module
    socket was not usable because supporting tables had been
    cleared and not re-initialized. This has now been fixed.
    Handling of the "." domain as a search domain was incorrect and
    caused a crash in the DNS resolver inet_res, which has now been
    fixed.
  * erts: A call to port_command() could cause a scheduler to end
    up in an eternal loop if the port was busy and the calling
    process had incoming signals at the time of the call. This bug
    was introduced in OTP 23.3.2 (ERTS version 11.2.1), OTP
    22.3.4.18 (ERTS version 10.7.2.10), and OTP 21.3.8.23 (ERTS
    version 10.3.5.18).
  * erts: Bug fixes and code cleanup for the new socket
    implementation, such as: Assertions on the result of
    demonitoring has been added in the NIF code, where appropriate.
    Internal state handling for socket close in the NIF code has
    been reviewed. Looping over close() for EINTR in the NIF code
    has been removed, since it is strongly discouraged on Linux and
    Posix is not clear about if it is allowed. The inet_backend
    temporary socket option for legacy gen_tcp sockets has been
    documented. The return value from net:getaddrinfo/2 has been
    corrected: the protocol field is now an atom(), instead of,
    incorrectly, list(atom()). The documentation has also been
    corrected about this return type. Deferred close of a
    socket:sendfile/* file was broken and has been corrected. Some
    debug code, not enabled by default, in the socket NIF has been
    corrected to not accidentally core dump for debug printouts of
    more or less innocent events.
  * erts: Dirty execution of a process in combination with an
    unlink signal from a port to the process could cause the signal
    queue of the process to enter into an inconsistent state. The
    result of the inconsistency typically caused a crash of the
    runtime system. This bug was introduced in OTP 23.3 (ERTS
    version 11.2).
  * erl_docgen: Fix links generated in specs to types in other
    applications to point to the correct place. This bug was
    introduced in Erlang/OTP 24.0.
  * ssl: Fix cache invalidation problem for CA certs provided by
    the cacertfile option.
  * ssh: Avoid an extra blank line in the ssh known_hosts file
  * compiler: Fixed a bug that could cause after blocks to be
    ignored when erlang:raise/3 was used in a catch block.
  * compiler: Fixed a bug in the validation pass that could cause
    it to reject valid code.
  * crypto: EC keys are now zero-padded to the expected length if
    needed.

• Files Changed
• Browse Source