diff options
Diffstat (limited to 'erts/doc/src/notes.xml')
| -rw-r--r-- | erts/doc/src/notes.xml | 936 |
1 files changed, 936 insertions, 0 deletions
diff --git a/erts/doc/src/notes.xml b/erts/doc/src/notes.xml index aaab52e424..ed0946d6ba 100644 --- a/erts/doc/src/notes.xml +++ b/erts/doc/src/notes.xml @@ -31,6 +31,942 @@ </header> <p>This document describes the changes made to the ERTS application.</p> +<section><title>Erts 10.0</title> + + <section><title>Fixed Bugs and Malfunctions</title> + <list> + <item> + <p> + The type specifications for <c>file:posix/0</c> and + <c>inet:posix/0</c> have been updated according to which + errors file and socket operations should be able to + return.</p> + <p> + Own Id: OTP-14019 Aux Id: ERL-550 </p> + </item> + <item> + <p> + Fix error printout from run_erl and a bug that could + cause unintended fds to be leaked into the started + program.</p> + <p> + Own Id: OTP-14537 Aux Id: PR1529 </p> + </item> + <item> + <p> File operations used to accept <seealso + marker="kernel:file#type-name_all">filenames</seealso> + containing null characters (integer value zero). This + caused the name to be truncated and in some cases + arguments to primitive operations to be mixed up. + Filenames containing null characters inside the filename + are now <em>rejected</em> and will cause primitive file + operations to fail. </p> <p> Also environment variable + operations used to accept <seealso + marker="kernel:os#type-env_var_name">names</seealso> and + <seealso + marker="kernel:os#type-env_var_value">values</seealso> of + environment variables containing null characters (integer + value zero). This caused operations to silently produce + erroneous results. Environment variable names and values + containing null characters inside the name or value are + now <em>rejected</em> and will cause environment variable + operations to fail. </p> <p>Primitive environment + variable operations also used to accept the <c>$=</c> + character in environment variable names causing various + problems. <c>$=</c> characters in environment variable + names are now also <em>rejected</em>. </p> <p>Also + <seealso + marker="kernel:os#cmd/1"><c>os:cmd/1</c></seealso> now + reject null characters inside its <seealso + marker="kernel:os#type-os_command">command</seealso>. + </p> <p><seealso + marker="erts:erlang#open_port/2"><c>erlang:open_port/2</c></seealso> + will also reject null characters inside the port name + from now on.</p> + <p> + *** POTENTIAL INCOMPATIBILITY ***</p> + <p> + Own Id: OTP-14543 Aux Id: ERL-370 </p> + </item> + <item> + <p> + Fix bugs related to the bookkeeping of microstate + accounting states.</p> + <p> + Own Id: OTP-14652</p> + </item> + <item> + <p><c>os:putenv</c> and <c>os:getenv</c> no longer access + the process environment directly and instead work on a + thread-safe emulation. The only observable difference is + that it's <em>not</em> kept in sync with libc + <c>getenv(3)</c> / <c>putenv(3)</c>, so those who relied + on that behavior in drivers or NIFs will need to add + manual synchronization.</p> <p>On Windows this means that + you can no longer resolve DLL dependencies by modifying + the <c>PATH</c> just before loading the driver/NIF. To + make this less of a problem, the emulator now adds the + target DLL's folder to the DLL search path.</p> + <p> + *** POTENTIAL INCOMPATIBILITY ***</p> + <p> + Own Id: OTP-14666</p> + </item> + <item> + <p>Corrected <c>erlang:is_builtin(erlang, M, F)</c> to + return <c>true</c> for <c>apply/2</c> and + <c>yield/0</c>.</p> + <p> + Own Id: OTP-14713 Aux Id: ERL-500 </p> + </item> + <item> + <p>Fixed a bug where the PATH environment variable wasn't + updated correctly on a release downgrade, effectively + keeping the PATH of the new release.</p> + <p> + Own Id: OTP-14719</p> + </item> + <item> + <p>A receive optimization that avoids scanning the entire + message queue when receiving a message containing a + freshly created reference could in rare circumstances + (involving recursive calls to the functions that does the + receive) cause the receive to hang. This has been + corrected.</p> + <p> + Own Id: OTP-14782 Aux Id: ERL-511 </p> + </item> + <item> + <p> + Fix building of Erlang/OTP on platforms which have small + data area with short addressing. For example the + PowerPC/RTEMS platform.</p> + <p> + Own Id: OTP-14909 Aux Id: PR-1692 </p> + </item> + <item> + <p>Fixed a crash when <c>enif_make_binary</c> is called + with a binary produced by <c>enif_inspect_binary</c> in a + different environment.</p> + <p> + Own Id: OTP-14931</p> + </item> + <item> + <p>Fixed a crash when <c>enif_make_binary</c> is called + more than once with a binary that had previously been + added to an <c>enif_ioq</c>.</p> + <p> + Own Id: OTP-14932</p> + </item> + <item> + <p> + The erl_child_setup program now ignores SIGTERM signals.</p> + <p> + Own Id: OTP-14943 Aux Id: ERL-576 </p> + </item> + <item> + <p> + Force 64-bit alignment on pre-allocators on architectures + which needs it.</p> + <p> + Own Id: OTP-14977</p> + </item> + <item> + <p> + Fixed a bug where dirty scheduler picked up non-dirty + work.</p> + <p> + Own Id: OTP-14978</p> + </item> + <item> + <p> + Calls to <c>gen_tcp:send/2</c> on closed sockets now + returns <c>{error, closed}</c> instead of + <c>{error,enotconn}</c>.</p> + <p> + Own Id: OTP-15001</p> + </item> + <item> + <p> + <c>erlang:monotonic_time/1</c> failed with <c>badarg</c> + when passing the <c>perf_counter</c> time unit as + argument.</p> + <p> + Own Id: OTP-15008</p> + </item> + <item> + <p> + Fix bug where rapid <c>init:restart()</c> calls would + sometimes crash because a code load request leaked in + between the restarts.</p> + <p> + Own Id: OTP-15013</p> + </item> + <item> + <p> + Improve <c>float_to_list(F, [{decimals,D}])</c> to closer + conform with <c>io_lib:format("~.*f", [D,F])</c>.</p> + <p> + There are however, still cases when <c>float_to_list</c> + does not produce the exact same result as + <c>io_lib:format</c>, especially for large values + <c>F</c> and/or many decimals <c>D</c>.</p> + <p> + Own Id: OTP-15015 Aux Id: OTP-14890 </p> + </item> + <item> + <p>Fixed a deadlock that would occur on certain + allocators when a reallocation failed with <c>+ramv</c> + enabled.</p> + <p> + Own Id: OTP-15024</p> + </item> + <item> + <p> + Fix bug that made it impossible to use an erl_tracer as + the seq_trace trace receiver.</p> + <p> + Own Id: OTP-15029</p> + </item> + <item> + <p> + Fix bug where a large (> 1 GB) emulator generated error + logger message would cause the emulator to crash.</p> + <p> + Own Id: OTP-15032</p> + </item> + <item> + <p>The emulator will no longer crash when reading the + file information of an ordinary file that has an NTFS + reparse point, such as files stored in a OneDrive-mapped + folder.</p> + <p> + Own Id: OTP-15062 Aux Id: ERL-615 </p> + </item> + <item> + <p> + Fixed bug in <c>enif_binary_to_term</c> which could cause + memory corruption for immediate terms (atoms, small + integers, pids, ports, empty lists).</p> + <p> + Own Id: OTP-15080</p> + </item> + <item> + <p> + Fixed bug in <c>erlang:system_profile/2</c> that could + cause superfluous <c>{profile,_,active,_,_}</c> messages + for terminating processes.</p> + <p> + Own Id: OTP-15085</p> + </item> + <item> + <p> + On OSs with per thread CPU time support, change + <c>cpu_timestamp</c> in <seealso + marker="erlang#trace/3">erlang:trace/3</seealso> to use + it instead of per process CPU time. This makes this + option useable on such OSs when running multiple + schedulers.</p> + <p> + Own Id: OTP-15090</p> + </item> + <item> + <p> + Fix segfault in abort_signal_task which could happen if a + port terminated while there were outstanding port tasks + that were not signals, for example a + ready_input/ready_output event.</p> + <p> + Own Id: OTP-15108 Aux Id: ERL-621 </p> + </item> + <item> + <p> + Fixed bug in <c>ets</c> that could cause VM crash if + process A terminates after fixating a table and process B + deletes the table at "the same time". The table fixation + could be done with <c>ets:safe_fixtable</c> or if process + A terminates in the middle of a long running + <c>select</c> or <c>match</c> call.</p> + <p> + Own Id: OTP-15109</p> + </item> + <item> + <p>Owner and group changes through + <c>file:write_file_info</c>, <c>file:change_owner</c>, + and <c>file:change_group</c> will no longer report + success on permission errors.</p> + <p> + *** POTENTIAL INCOMPATIBILITY ***</p> + <p> + Own Id: OTP-15118</p> + </item> + <item> + <p> + Fix a bug error reporting from escripts on windows where + the error message would get garbled.</p> + <p> + Own Id: OTP-15119 Aux Id: PR-1826 </p> + </item> + <item> + <p> + Fix segfault when a process is interally re-scheduled + while being traced for in out events. This bug was + introduced in erts-8.0 (OTP-19.0).</p> + <p> + Own Id: OTP-15125</p> + </item> + </list> + </section> + + + <section><title>Improvements and New Features</title> + <list> + <item> + <p>It is now possible to open device files and FIFOs with + <c>file:open/2</c>.</p> + <p> + Own Id: OTP-11462</p> + </item> + <item> + <p> + The <c>erlang:system_flag(scheduler_wall_time,Bool)</c> + call is now reference counted and will be turned off if + the (last) process that started the performance + statistics dies. Thus it is no longer possible to start + the statistics with <c>rpc:call(Node, erlang, + system_flag, [scheduler_wall_time, true])</c> since it + will be turned off directly afterwards when the rpc + process dies.</p> + <p> + *** POTENTIAL INCOMPATIBILITY ***</p> + <p> + Own Id: OTP-11694</p> + </item> + <item> + <p>A new logging API is added to Erlang/OTP, see the + <seealso + marker="kernel:logger"><c>logger(3)</c></seealso> manual + page, and section <seealso + marker="kernel:logger_chapter">Logging</seealso> in the + Kernel User's Guide.</p> + <p>Calls to <c>error_logger</c> are automatically + redirected to the new API, and legacy error logger event + handlers can still be used. It is, however, recommended + to use the Logger API directly when writing new code.</p> + <p>Notice the following potential incompatibilities:</p> + <list> <item><p>Kernel configuration parameters + <c>error_logger</c> still works, but is overruled if the + default handler's output destination is configured with + Kernel configuration parameter <c>logger</c>.</p> <p>In + general, parameters for configuring error logger are + overwritten by new parameters for configuring + Logger.</p></item> <item><p>The concept of SASL error + logging is deprecated, meaning that by default the SASL + application does not affect which log events are + logged.</p> <p>By default, supervisor reports and crash + reports are logged by the default Logger handler started + by Kernel, and end up at the same destination (terminal + or file) as other standard log event from Erlang/OTP.</p> + <p>Progress reports are not logged by default, but can be + enabled by setting the primary log level to info, for + example with the Kernel configuration parameter + <c>logger_level</c>.</p> <p>To obtain backwards + compatibility with the SASL error logging functionality + from earlier releases, set Kernel configuration parameter + <c>logger_sasl_compatible</c> to <c>true</c>. This + prevents the default Logger handler from logging any + supervisor-, crash-, or progress reports. Instead, SASL + adds a separate Logger handler during application start, + which takes care of these log events. The SASL + configuration parameters <c>sasl_error_logger</c> and + <c>sasl_errlog_type</c> specify the destination (terminal + or file) and severity level to log for these + events.</p></item></list> + <p> + Since Logger is new in Erlang/OTP 21.0, we do reserve the + right to introduce changes to the Logger API and + functionality in patches following this release. These + changes might or might not be backwards compatible with + the initial version.</p> + <p> + *** POTENTIAL INCOMPATIBILITY ***</p> + <p> + Own Id: OTP-13295</p> + </item> + <item> + <p> + <c>gen_sctp:connect_init/4</c> or rather connect in + <c>inet_drv.c</c> for SCTP has been fixed to not check + the write file descriptor for writeability after a + connect, since for SCTP (SOCK_SEQPACKET) that property + does not seem to be any kind of indicator for when a + connect has finished. This fixes connects that the OS + returned as "in progress" that was misinterpreted by + <c>gen_sctp:connect_init</c> as failed.</p> + <p> + Own Id: OTP-13760 Aux Id: PR-1592 </p> + </item> + <item> + <p>The file driver has been rewritten as a NIF, + decreasing the latency of file operations. Notable + incompatibilities are:</p> <list> <item><p>The + <c>use_threads</c> option for <c>file:sendfile/5</c> no + longer has any effect; we either use non-blocking + <c>sendfile(2)</c> or fall back to <c>file:read</c> + + <c>gen_tcp:send</c>. </p></item> <item><p>The + file-specific DTrace probes have been removed. The same + effect can be achieved with normal tracing together with + the <c>nif__entry</c>/<c>nif__return</c> probes to track + scheduling.</p></item> </list> + <p> + *** POTENTIAL INCOMPATIBILITY ***</p> + <p> + Own Id: OTP-14256</p> + </item> + <item> + <p>The I/O polling functionality of erts has been + re-written to better make use of the OSs polling + mechanisms. This change means that erts will now always + prefer to use a kernel-polling mechanism if possible. + Also all of the I/O polling has been moved to dedicated + threads instead of being placed in the scheduler + loops.</p> <p>As a result of this, the <c>erl</c> options + <c>+K</c> and <c>+secio</c> have been removed. It is + still possible to disable kernel-poll, but it has to be + done at compile time through the configure option + <c>--disable-kernel-poll</c>.</p> <p>The new <c>erl</c> + options <seealso marker="erl#+IOt"><c>+IOt</c></seealso> + and <seealso marker="erl#+IOp"><c>+IOp</c></seealso> can + be used to change how many IO poll threads and poll sets + that erts should use. See their respective documentation + for more details.</p> + <p> + Own Id: OTP-14346</p> + </item> + <item> + <p>Truly asynchronous auto-connect. Earlier, when + <c>erlang:send</c> was aimed toward an unconnected node, + the function would not return until the connection setup + had completed (or failed). Now the function returns + directly after the message has been enqueued and the + connection setup started.</p> + <p>The same applies to all distributed operations that + may trigger auto-connect, i.e. <c>'!'</c>, <c>send</c>, + <c>link</c>, <c>monitor</c>, <c>monitor_node</c>, + <c>exit/2</c> and <c>group_leader</c>.</p> + <p>The interface for all these functions are unchanged as + they do not return connection failures. The only + exception is <c>erlang:monitor</c> where a <em>possible + incompatibility</em> is introduced: An attempt to monitor + a process on a primitive node (such as erl_interface or + jinterface), where remote process monitoring is not + implemented, will no longer fail with <c>badarg</c> + exception. Instead a monitor will be created, but it will + only supervise the connection to the node.</p> + <p> + *** POTENTIAL INCOMPATIBILITY ***</p> + <p> + Own Id: OTP-14370</p> + </item> + <item> + <p>Changed the default behaviour of <c>.erlang</c> + loading: <c>.erlang</c> is no longer loaded from the + current directory. <c>c:erlangrc(PathList)</c> can be + used to search and load an <c>.erlang</c> file from user + specified directories.</p> <p><c>escript</c>, + <c>erlc</c>, <c>dialyzer</c> and <c>typer</c> no longer + load an <c>.erlang</c> at all.</p> + <p> + *** POTENTIAL INCOMPATIBILITY ***</p> + <p> + Own Id: OTP-14439</p> + </item> + <item> + <p> + New functionality for implementation of alternative + carriers for the Erlang distribution has been introduced. + This mainly consists of support for usage of distribution + controller processes (previously only ports could be used + as distribution controllers). For more information see + <seealso marker="erts:alt_dist#distribution_module">ERTS + User's Guide ➜ How to implement an Alternative Carrier + for the Erlang Distribution ➜ Distribution + Module</seealso>.</p> + <p> + Own Id: OTP-14459</p> + </item> + <item> + <p> + Add support for the lcc compiler and in extension the + Elbrus 2000 platform.</p> + <p> + Own Id: OTP-14492</p> + </item> + <item> + <p>Support for "tuple calls" have been removed from the + run-time system. Tuple calls was an undocumented and + unsupported feature which allowed the module argument for + an apply operation to be a tuple: <c>Var = dict:new(), + Var:size()</c>. This "feature" frequently caused + confusion, especially when such call failed. The + stacktrace would point out functions that don't exist in + the source code.</p> + <p>For legacy code that need to use parameterized modules + or tuple calls for some other reason, there is a new + compiler option called <c>tuple_calls</c>. When this + option is given, the compiler will generate extra code + that emulates the old behavior for calls where the module + is a variable.</p> + <p> + *** POTENTIAL INCOMPATIBILITY ***</p> + <p> + Own Id: OTP-14497</p> + </item> + <item> + <p>Creation of small maps with literal keys has been + optimized to be faster and potentially use less memory. + The keys are combined into a literal key tuple which is + put into the literal pool. The key tuple can be shared + between many instances of maps having the same keys.</p> + <p> + Own Id: OTP-14502</p> + </item> + <item> + <p> + When an exception is thrown, include the arguments of the + call in the stacktrace for BIFs <c>band</c>, <c>bor</c>, + <c>bsl</c>, <c>bsr</c>, <c>bxor</c>, <c>div</c>, + <c>rem</c> and the operators <c>+</c>, <c>-</c>, <c>*</c> + and <c>/</c>.</p> + <p> + *** POTENTIAL INCOMPATIBILITY ***</p> + <p> + Own Id: OTP-14508</p> + </item> + <item> + <p> + The non-smp emulators have been removed. This means that + the configure options <c>--disable-threads</c> and + <c>--enable-plain-emulator</c> have been removed and + configure will now refuse to build Erlang/OTP on + platforms without thread support.</p> + <p> + In order to achieve a similar setup as the non-smp + emulator, it is possible to start Erlang/OTP with the + <c>+S 1</c> option.</p> + <p> + *** POTENTIAL INCOMPATIBILITY ***</p> + <p> + Own Id: OTP-14518</p> + </item> + <item> + <p>Modules that use floating point constants compiled + with R15 or earlier will need to be re-compiled before + they can be loaded.</p> + <p> + Own Id: OTP-14575</p> + </item> + <item> + <p> + Implementation of true asynchronous signaling between + processes in order to improve scalability. Signals + affected include exit, monitor, demonitor, monitor + triggered, link, unlink, and group leader.</p> + <p> + Own Id: OTP-14589</p> + </item> + <item> + <p> + Added a PGO (profile guided optimization) pass to the + build step of erts. This can be disabled by passing + --disable-pgo to configure.</p> + <p> + Own Id: OTP-14604</p> + </item> + <item> + <p> + Improved the performance of <c>binary:split</c> and + <c>binary:match</c>.</p> + <p> + Own Id: OTP-14610 Aux Id: PR-1480 </p> + </item> + <item> + <p> + It is not longer possible to disable dirty schedulers + when building erlang.</p> + <p> + Own Id: OTP-14613</p> + </item> + <item> + <p>Loaded BEAM code in a 64-bit system requires less + memory because of better packing of operands for + instructions.</p> + <p>These memory savings were achieved by major + improvements to the <c>beam_makeops</c> scripts used when + building the run time system and BEAM compiler. There is + also new for documentation for <c>beam_makeops</c> that + describes how new BEAM instructions and loader + transformations can be implemented. The documentation is + found in here in a source directory or git repository: + erts/emulator/internal_doc/beam_makeops.md. An online + version can be found here: + https://github.com/erlang/otp/blob/master/erts/emulator/internal_doc/beam_makeops.md</p> + <p> + Own Id: OTP-14626</p> + </item> + <item> + <p><c>file:read_file</c> has been changed to read the + content of files that report a size of 0 even when data + can be read from them. An example of such a file is + <c>/proc/cpuinfo</c> on Linux.</p> + <p> + Own Id: OTP-14637 Aux Id: ERL-327 PR-1524 </p> + </item> + <item> + <p> + It is no longer possible to disable the <c>temp_alloc</c> + allocator. Disabling it caused serious performance + degradations and was never what was wanted.</p> + <p> + Own Id: OTP-14651</p> + </item> + <item> + <p>The reduction cost of sending messages is now + constant. It will no longer scale according to the length + of the receiving process' message queue.</p> + <p> + Own Id: OTP-14667</p> + </item> + <item> + <p> + Improved loading of modules with <c>-on_load</c> + directive, to no longer block all schedulers when the + load operation is completed.</p> + <p> + Own Id: OTP-14680</p> + </item> + <item> + <p> + On platforms with real-time signals available, SIGRTMIN+1 + is now used as the internal scheduler suspend signal + instead of SIGUSR2.</p> + <p> + Own Id: OTP-14682</p> + </item> + <item> + <p>When the value returned from a '<c>catch</c>' + expression is ignored, no stacktrace will be built if an + exception is caught. That will save time and produce less + garbage. There are also some minor optimizations of + '<c>try</c>/<c>catch</c>' both in the compiler and + run-time system.</p> + <p> + Own Id: OTP-14683</p> + </item> + <item> + <p>The guarantees and non-guarantees of + <c>erlang:get_stacktrace/0</c> are now documented.</p> + <p> + Own Id: OTP-14687</p> + </item> + <item> + <p>There is a new syntax in '<c>try/catch</c>' for + retrieving the stacktrace without calling + '<c>erlang:get_stacktrace/0</c>'. See the reference + manual for a description of the new syntax. The + '<c>erlang:get_stacktrace/0</c>' BIF is now + deprecated.</p> + <p> + Own Id: OTP-14692</p> + </item> + <item> + <p> + New 'used' option for <c>binary_to_term/2</c> that will + also return number of bytes actually read from the + binary. This enables easy access to any extra data in the + binary located directly after the returned term.</p> + <p> + Own Id: OTP-14780</p> + </item> + <item> + <p> + Added more statistics for + <c>erlang:system_info({allocator,A})</c> in the + <c>mbcs_pool</c> section.</p> + <p> + Own Id: OTP-14795 Aux Id: ERL-88 </p> + </item> + <item> + <p>Added <c>enif_ioq_peek_head</c> to retrieve Erlang + terms from NIF IO queues without having to resort to + copying.</p> + <p> + Own Id: OTP-14797</p> + </item> + <item> + <p>There is a new option '<c>makedep_side_effect</c>' for + the compiler and <c>-MMD</c> for '<c>erlc</c>' that + generates dependencies and continues to compile as + normal.</p> + <p> + Own Id: OTP-14830</p> + </item> + <item> + <p>Added <c>ets:whereis/1</c> for retrieving the table + identifier of a named table.</p> + <p> + Own Id: OTP-14884</p> + </item> + <item> + <p><c>seq_trace</c> labels may now be any erlang + term.</p> + <p> + Own Id: OTP-14899</p> + </item> + <item> + <p> + Optimized the common case of <c>monitor</c> followed by + <c>send</c> to the same local process. The monitor signal + is now delayed in order to be piggybacked with the sent + message and thereby only get one lock operation on the + message queue of the receiver. A delayed monitor signal + is flushed if no <c>send</c> has been done at the latest + when the process is scheduled out.</p> + <p> + Own Id: OTP-14901</p> + </item> + <item> + <p> + Make hipe compiled code work on x86_64 (amd64) with OS + security feature PIE, where executable code can be loaded + into a random location. Old behavior, if hipe was + enabled, was to disable PIE build options for the VM.</p> + <p> + Own Id: OTP-14903</p> + </item> + <item> + <p>The number of driver async threads will now default to + 1 as the standard drivers do not use them anymore. Users + that changed this value to tweak the file driver should + replace <c>+A</c> with <c>+SDio</c> since it now uses + dirty IO schedulers instead of async threads.</p> + <p> + *** POTENTIAL INCOMPATIBILITY ***</p> + <p> + Own Id: OTP-14928</p> + </item> + <item> + <p> + Optimize <c>==</c> and <c>/=</c> for binaries with + different sizes to be constant in time instead of + proportional to the size of their common prefix.</p> + <p> + Own Id: OTP-14934 Aux Id: PR-1708 </p> + </item> + <item> + <p> + Refactorings making some internal process flags available + for other usage.</p> + <p> + Own Id: OTP-14948</p> + </item> + <item> + <p> + Removed need for HiPE to allocate native executable + memory in low 2GB address space on x86_64. Command line + option <c>+MXscs</c> is thereby obsolete and ignored.</p> + <p> + Own Id: OTP-14951</p> + </item> + <item> + <p>Added <c>enif_make_map_from_arrays</c> for creating a + populated map, analogous to + <c>enif_make_list_from_array</c>.</p> + <p> + Own Id: OTP-14954</p> + </item> + <item> + <p>Added configuration switches for busy-wait and wake up + thresholds for dirty schedulers, and changing these + settings for normal schedulers will no longer affect + dirty schedulers. </p> <p>Refer to the documentation for + details. The new switches are <seealso + marker="erl#+sbwtdcpu">+sbwtdcpu</seealso>, <seealso + marker="erl#+sbwtdio">+sbwtdio</seealso>, <seealso + marker="erl#+swtdcpu">+swtdcpu</seealso>, and <seealso + marker="erl#+swtdio">+swtdio</seealso>.</p> <p>The + default busy wait threshold for dirty scheduler threads + has also been lowered to <c>short</c>.</p> + <p> + Own Id: OTP-14959</p> + </item> + <item> + <p> + The list of "taints" now also includes dynamic loaded + drivers in addition to NIF libraries. Statically linked + drivers and NIF libraries that are part of erts are not + included. The "taints" are returned by + <c>system_info(taints)</c> and printed in the header of + <c>erl_crash.dump</c> files.</p> + <p> + Own Id: OTP-14960</p> + </item> + <item> + <p>Added <c>instrument:allocations</c> and + <c>instrument:carriers</c> for retrieving information + about memory utilization and fragmentation.</p> + <p>The old <c>instrument</c> interface has been removed, + as have the related options <c>+Mim</c> and + <c>+Mis</c>.</p> + <p> + *** POTENTIAL INCOMPATIBILITY ***</p> + <p> + Own Id: OTP-14961</p> + </item> + <item> + <p>The process suspend functionality used by the <seealso + marker="erlang#suspend_process/2">erlang:suspend_process/2</seealso> + BIF has been reimplemented using the newly introduced + true asynchronous signaling between processes. This + mainly to reduce memory usage in the process control + block of all processes, but also in order to simplify the + implementation.</p> <warning> <p>You can easily create + deadlocks if processes suspends each other (directly or + in circles). In ERTS versions prior to ERTS version 10.0, + the runtime system prevented such deadlocks, but this + prevention has now been removed due to performance + reasons.</p> </warning> <p>Other ERTS internal + functionality that used the previous process suspend + functionality have also been reimplemented to use + asynchronous signaling instead.</p> + <p> + *** POTENTIAL INCOMPATIBILITY ***</p> + <p> + Own Id: OTP-14964 Aux Id: OTP-14589 </p> + </item> + <item> + <p>Added the <c>nifs</c> option to + <c>?MODULE:module_info/1</c> for listing a module's + installed NIF functions.</p> + <p> + Own Id: OTP-14965</p> + </item> + <item> + <p> + New implementation of <c>erlang:process_info/[1,2]</c>.</p> + <p> + In the general case when inspecting another process, the + new implementation sends an asynchronous process-info + request signal to the other process and waits for the + result instead of locking the other process and reading + the result directly. In some special cases where no + conflicts occur, signal order wont be violated, and the + amount of data requested is guaranteed to be small, the + inspected process may be inspected directly.</p> + <p> + Appropriate amount of reductions are now also bumped when + inspecting a process.</p> + <p> + Own Id: OTP-14966</p> + </item> + <item> + <p> + Removed process start time from crash dump in order to + save memory in process control block.</p> + <p> + Own Id: OTP-14975 Aux Id: PR-1597 </p> + </item> + <item> + <p> + Optimize <c>erlang:put/2</c> when updating existing key + with a new immediate value (atom, small integer, pid, + port).</p> + <p> + Own Id: OTP-14976</p> + </item> + <item> + <p> + <c>erlang:process_info/1</c> has been changed to no + longer include <c>messages</c> by default. Instead + <c>erlang:process_info/2</c> should be used.</p> + <p> + *** POTENTIAL INCOMPATIBILITY ***</p> + <p> + Own Id: OTP-14986 Aux Id: PR-1745 </p> + </item> + <item> + <p> + New <c>erlang:system_info(ets_count)</c> to get total + number of ets tables existing at the local node.</p> + <p> + Own Id: OTP-14987</p> + </item> + <item> + <p> + New NIF functions: <c>enif_mutex_name</c>, + <c>enif_cond_name</c>, <c>enif_rwlock_name</c>, + <c>enif_thread_name</c>, <c>enif_vfprintf</c>, + <c>enif_vsnprintf</c>.</p> + <p> + Own Id: OTP-14994</p> + </item> + <item> + <p>When <c>erlang:system_flag(backtrace_depth, 0)</c> has + been called, all exceptions will now contain the entry + for <em>one</em> function (despite the zero). It used to + be that a hand-made stack backtrace passed to + <c>erlang:raise/3</c> would be be truncated to an empty + list.</p> + <p> + Own Id: OTP-15026</p> + </item> + <item> + <p> + Fixed bug for named <c>ets</c> tables which could cause + unexpected results from matchspec iteration functions + (<c>ets:select*</c> and <c>ets:match*</c>) if the table + was deleted and recreated with the same name during the + iteration. The iteration could incorrectly continue + through the recreated table. The expected correct + behavior is now for the iteration call to fail with a + <c>badarg</c> exception if the table is deleted before + the iteration has completed.</p> + <p> + Own Id: OTP-15031</p> + </item> + <item> + <p>Two new guards BIFs operating on maps have been added: + <c>map_get/2</c> and <c>is_map_key/2</c>. They do the + same as <c>maps:get/2</c> and <c>maps:is_key/2</c>, + respectively, except that they are allowed to be used in + guards.</p> + <p> + Own Id: OTP-15037 Aux Id: PR-1784, PR-1802 </p> + </item> + <item> + <p> + Release run-queue lock while cleaning up terminated dirty + process.</p> + <p> + Own Id: OTP-15081</p> + </item> + <item> + <p>The callback module passed as <c>-epmd_module</c> to + erl has been expanded to be able to do name and port + resolving.</p> <p>Documentation has also been added in + the <seealso + marker="kernel:erl_epmd"><c>erl_epmd</c></seealso> + reference manual and ERTS User's Guide <seealso + marker="erts:alt_disco">How to Implement an Alternative + Service Discovery for Erlang Distribution</seealso>.</p> + <p> + Own Id: OTP-15086 Aux Id: PR-1694 </p> + </item> + </list> + </section> + +</section> + <section><title>Erts 9.3.3</title> <section><title>Fixed Bugs and Malfunctions</title> |