diff options
Diffstat (limited to 'erts/doc/src/notes.xml')
| -rw-r--r-- | erts/doc/src/notes.xml | 703 |
1 files changed, 703 insertions, 0 deletions
diff --git a/erts/doc/src/notes.xml b/erts/doc/src/notes.xml index 2574d45184..2d96ed6105 100644 --- a/erts/doc/src/notes.xml +++ b/erts/doc/src/notes.xml @@ -31,6 +31,709 @@ </header> <p>This document describes the changes made to the ERTS application.</p> +<section><title>Erts 7.0</title> + + <section><title>Fixed Bugs and Malfunctions</title> + <list> + <item> + <p> + Fix issuing with spaces and quoting in the arguments when + using erlang:open_port spawn_executable on windows. The + behavior now mimics how unix works. This change implies a + backwards incompatibility for how spawn_executable works + on windows.</p> + <p> + *** POTENTIAL INCOMPATIBILITY ***</p> + <p> + Own Id: OTP-11905</p> + </item> + <item> + <p> + Fix global call trace when hipe compiled code call beam + compiled functions. Tracing of beam functions should now + alway work regardless who the caller is.</p> + <p> + Own Id: OTP-11939</p> + </item> + <item> + <p> + Correct cache alignment for ETS <c>write_concurrency</c> + locks to improve performance by reduced false sharing. + May increase memory footprint for tables with + <c>write_concurrency</c>.</p> + <p> + Own Id: OTP-11974</p> + </item> + <item> + <p> + All possibly blocking operations in the fd/spawn and + terminal driver have been converted to non-blocking + operations. Before this fix it was possible for the VM to + be blocked for a long time if the entity consuming + stdout/stderr did not consume it fast enough.</p> + <p> + Own Id: OTP-12239</p> + </item> + <item> + <p> + Add missing overhead for offheap binaries created from + external format. This fix can improve the garbage + collection of large binaries originating from + <c>binary_to_term</c> or messages from remote nodes.</p> + <p> + Own Id: OTP-12554</p> + </item> + <item> + <p> + Ensure hashing of zero is consistent</p> + <p> Erlang treats positive and negative zero as + equal:</p> + <p> + <c>true = 0.0 =:= 0.0/-1</c></p> + <p>However, Erlangs hash functions: hash, phash and + phash2 did not reflect this behaviour. The hash values + produced by the different hash functions would not be + identical for positive and negative zero.</p> <p>This + change ensures that hash value of positive zero is always + produced regardless of the signedness of the zero float, + i.e.,</p> + <p> + <c>true = erlang:phash2(0.0) =:= + erlang:phash2(0.0/-1)</c></p> + <p> + Own Id: OTP-12641</p> + </item> + <item> + <p> + Ensure NIF term creation disallows illegal floating point + values and too long atoms. Such values will cause a NIF + to throw badarg exception when it returns.</p> + <p> + Own Id: OTP-12655</p> + </item> + <item> + <p> + Fixed building of Map results from match_specs</p> + <p> + A faulty "box-value" entered into the heap which could + cause a segmentation fault in the garbage collector if it + was written on a heap fragment.</p> + <p> + Own Id: OTP-12656</p> + </item> + <item> + <p> + Fix hipe bug when matching a "writable" binary. The bug + has been seen to sometimes cause a failed binary matching + of a correct utf8 character, but other symptoms are also + possible.</p> + <p> + Own Id: OTP-12667</p> + </item> + <item> + <p> + Keep dirty schedulers from waking other schedulers.</p> + <p> + Own Id: OTP-12685</p> + </item> + <item> + <p> + Disable floating point exceptions if the VM is compiled + by clang/llvm. This is a known long-standing problem in + clang/llvm.</p> + <p> + Own Id: OTP-12717</p> + </item> + <item> + <p> + Fix bug in <c>file:sendfile</c> for FreeBSD causing not + the entire file to be sent.</p> + <p> + Own Id: OTP-12720</p> + </item> + <item> + <p> + Fix the broken Android support in erl_child_setup.c</p> + <p> + Own Id: OTP-12751</p> + </item> + <item> + <p> + Faulty statistics reported by the <c>fix_alloc</c> + allocator.</p> + <p> + Own Id: OTP-12766</p> + </item> + <item> + <p> + Fix two erts_snprintf() calls to correct sizes.</p> + <p> + - run_erl.c (ose): Use the size of the signal type, not + its pointer. - erl_node_tables.c: Use the size of the + _BUFFER in erts_snprintf() to make sure we can use the + full space.</p> + <p> + Own Id: OTP-12771</p> + </item> + <item> + <p> + Delayed memory allocations could be delayed an + unnecessarily long time.</p> + <p> + Own Id: OTP-12812</p> + </item> + <item> + <p> + Make sure that timeouts on a pool of acceptors are + released in the correct order.</p> + <p> + Own Id: OTP-12817</p> + </item> + <item> + <p> + Fix segmentation fault in module_info for deleted modules</p> + <p> + Own Id: OTP-12820</p> + </item> + <item> + <p>Fix garbage collection of literals in code purge</p> + <p>During code purging and check_process_code, the + checking of the binary reference embedded in the match + binary state was omitted for the tracing tests. This + would cause the binary match state to reference + deallocated memory.</p> + <p> + Own Id: OTP-12821</p> + </item> + <item> + <p> + A bug has been corrected for gen_tcp:close so when + {linger,{true,0}} is in effect it does not wait for data + in the driver queue to transfer out before closing the + port. Bug fix by Rory Byrne.</p> + <p> + Own Id: OTP-12840</p> + </item> + <item> + <p> + The documentation of the driver callback <seealso + marker="driver_entry#start"><c>start()</c></seealso> + erroneously stated that a return value of + <c>ERL_DRV_ERROR_ERRNO</c> caused the error value to be + passed via <c>erl_errno</c> when it should have been + <c>errno</c>.</p> + <p> + Own Id: OTP-12855</p> + </item> + </list> + </section> + + + <section><title>Improvements and New Features</title> + <list> + <item> + <p> + Add <c>md5</c> and <c>module</c> entries to + <c>?MODULE:module_info/0/1</c> and remove obsolete entry + 'import'.</p> + <p> + *** POTENTIAL INCOMPATIBILITY ***</p> + <p> + Own Id: OTP-11940</p> + </item> + <item> + <p> + Debug function <c>erlang:display/1</c> shows content of + binaries and bitstrings, not only the length.</p> + <p> + Own Id: OTP-11941</p> + </item> + <item> + <p>The time functionality of Erlang has been extended. + This both includes a <seealso + marker="time_correction#The_New_Time_API">new + API</seealso> for time, as well as <seealso + marker="time_correction#Time_Warp_Modes">time warp + modes</seealso> which alters the behavior of the system + when system time changes. <em>You are strongly encouraged + to use the new API</em> instead of the old API based on + <seealso + marker="erlang#now/0"><c>erlang:now/0</c></seealso>. + <c>erlang:now/0</c> has been deprecated since it is and + forever will be a scalability bottleneck. For more + information see the <seealso + marker="time_correction">Time and Time + Correction</seealso> chapter of the ERTS User's + Guide.</p> + <p>Besides the API changes and time warp modes a lot of + scalability and performance improvements regarding time + management has been made internally in the runtime + system. Examples of such improvements are scheduler + specific timer wheels, scheduler specific BIF timer + management, parallel retrieval of monotonic time and + system time on systems with primitives that are not + buggy.</p> + <p> + Own Id: OTP-11997</p> + </item> + <item> + <p><c>erlang:function_exported(M, F, A)</c> will now + return <c>true</c> if <c>M:F/A</c> refers to a BIF.</p> + <p> + *** POTENTIAL INCOMPATIBILITY ***</p> + <p> + Own Id: OTP-12099</p> + </item> + <item> + <p> + New BIF: <c>erlang:get_keys/0</c>, lists all keys + associated with the process dictionary. Note: + <c>erlang:get_keys/0</c> is auto-imported.</p> + <p> + *** POTENTIAL INCOMPATIBILITY ***</p> + <p> + Own Id: OTP-12151 Aux Id: seq12521 </p> + </item> + <item> + <p> + Make distributed send of large messages yield to improve + real-time characteristics.</p> + <p> + Own Id: OTP-12232</p> + </item> + <item> + <p> + Use high accuracy poll timeouts</p> + <p> + Where available, use poll/select API's that can handle + time resolutions less than 1ms. In the cases where such + API's are not available the timeout is rounded up to the + nearest ms.</p> + <p> + Own Id: OTP-12236</p> + </item> + <item> + <p> + The internal group to user_drv protocol has been changed + to be synchronous in order to guarantee that output sent + to a process implementing the user_drv protocol is + printed before replying. This protocol is used by the + standard_output device and the ssh application when + acting as a client. </p> + <p> + This change changes the previous unlimited buffer when + printing to standard_io and other devices that end up in + user_drv to 1KB.</p> + <p> + *** POTENTIAL INCOMPATIBILITY ***</p> + <p> + Own Id: OTP-12240</p> + </item> + <item> + <p>The previously introduced "eager check I/O" feature is + now enabled by default.</p> + <p>Eager check I/O can be disabled using the <c>erl</c> + command line argument: <seealso + marker="erl#+secio"><c>+secio false</c></seealso></p> + <p>Characteristics impact compared to previous + default:</p> <list> <item>Lower latency and smoother + management of externally triggered I/O operations.</item> + <item>A slightly reduced priority of externally triggered + I/O operations.</item> </list> + <p> + Own Id: OTP-12254 Aux Id: OTP-12117 </p> + </item> + <item> + <p> + Properly support maps in match_specs</p> + <p> + Own Id: OTP-12270</p> + </item> + <item> + <p> + The notice that a crashdump has been written has been + moved to be printed before the crashdump is generated + instead of afterwords. The wording of the notice has also + been changed.</p> + <p> + *** POTENTIAL INCOMPATIBILITY ***</p> + <p> + Own Id: OTP-12292</p> + </item> + <item> + <p> + New function <c>ets:take/2</c>. Works the same as + <c>ets:delete/2</c> but also returns the deleted + object(s).</p> + <p> + Own Id: OTP-12309</p> + </item> + <item> + <p> + Tracing with cpu_timestamp option has been enabled on + Linux.</p> + <p> + Own Id: OTP-12366</p> + </item> + <item> + <p> + ets:info/1,2 now contains information about whether + write_concurrency or read_concurrency is enabled.</p> + <p> + Own Id: OTP-12376</p> + </item> + <item> + <p> + Improved usage of <c>gcc</c>'s builtins for atomic memory + access. These are used when no other implementation of + atomic memory operations is available. For example, when + compiling for ARM when <c>libatomic_ops</c> is not + available.</p> + <p> + The largest improvement will be seen when compiling with + a <c>gcc</c> with support for the <c>__atomic_*</c> + builtins (using a <c>gcc</c> of at least version 4.7), + but also when only the legacy <c>__sync_*</c> builtins + are available (using a <c>gcc</c> of at least version + 4.1) an improvement can be seen.</p> + <p> + For more information see the "<seealso + marker="doc/installation_guide:INSTALL#Advanced-configuration-and-build-of-ErlangOTP_Configuring_Atomic-Memory-Operations-and-the-VM">Atomic + Memory Operations and the VM</seealso>" section of + <c>$ERL_TOP/HOWTO/INSTALL.md</c>.</p> + <p> + Own Id: OTP-12383</p> + </item> + <item> + <p> + Introduce <c>math:log2/1</c> function to math module.</p> + <p> + Own Id: OTP-12411</p> + </item> + <item> + <p>The documentation of the Abstract Format (in the ERTS + User's Guide) has been updated with types and + specification. (Thanks to Anthony Ramine.) </p> <p> The + explicit representation of parentheses used in types of + the abstract format has been removed. Instead the new + functions <c>erl_parse:type_inop_prec()</c> and + <c>erl_parse:type_preop_prec()</c> can be used for + inserting parentheses where needed. </p> + <p> + Own Id: OTP-12492</p> + </item> + <item> + <p> + Remove perfctr support</p> + <p> + Development of perfctr in the linux kernel ceased in + 2010. The perfctr support code in the Erlang VM is thus + effectively dead code and therefor removed.</p> + <p> + Own Id: OTP-12508</p> + </item> + <item> + <p><c>zlib:inflateChunk/2</c> has been added. It works + like <c>zlib:inflate/2</c>, but decompresses no more data + than will fit in the buffer configured by + <c>zlib:setBufSize/2</c>.</p> + <p> + Own Id: OTP-12548</p> + </item> + <item> + <p> + Use linear search for small select_val arrays</p> + <p> + Own Id: OTP-12555</p> + </item> + <item> + <p> + New BIF ets:update_counter/4 with a default object as + argument, which will be inserted in the table if the key + was not found.</p> + <p> + Own Id: OTP-12563</p> + </item> + <item> + <p> + Export missing types from zlib module</p> + <p> + Own Id: OTP-12584</p> + </item> + <item> + <p> + Use persistent hashmaps for large Maps <p>Maps will use a + persistent hashmap implementation when the number of + pairs in a Map becomes sufficiently large. The change + will occur when a Map reaches 33 pairs in size but this + limit might change in the future.</p></p> + <p>The most significant impact for the user by this + change is speed, and to a lesser degree memory + consumption and introspection of Maps. Memory consumption + size is probalistic but lesser than <c>gb_trees</c> or + <c>dict</c> for instance. Any other impacts will be + transparent for the user except for the following + changes.</p> + <p>Semantics of Maps have changed in two incompatible + ways compared to the experimental implementation in OTP + 17:</p> <list> <item>Hashing of maps is done different by + <c>erlang:phash2/1,2</c>, <c>erlang:phash/1</c> and + <c>erlang:hash/2</c>.</item> <item>Comparing two maps + with ==, /=, =<, <, >= and >, is done + different if the keys contain floating point + numbers.</item> </list> + <p> + *** POTENTIAL INCOMPATIBILITY ***</p> + <p> + Own Id: OTP-12585</p> + </item> + <item> + <p> + Scalability improvement for <seealso + marker="erlang#make_ref/0">erlang:make_ref/0</seealso>, + and other functionality that create references. Each + scheduler now manage its own set of references. By this + no communication at all is needed when creating + references.</p> + <p> + Previous implementation generated a strictly + monotonically increasing sequence of references + corresponding to creation time on the runtime system + instance. This is <em>not</em> the case with current + implementation. You can only expect reference to be + unique. The Erlang/OTP documentation has never mentioned + anything else but the uniqueness property, so this change + <em>is</em> fully compatible. The only reason we've + marked this as a potential incompatibility is since an + early draft for an Erlang specification mentions strict + monotonicity as a property.</p> + <p> + If you need to create data with a strict monotonicity + property use <seealso + marker="erlang#unique_integer/1">erlang:unique_integer([monotonic])</seealso>. + Do <em>not</em> use the deprecated <seealso + marker="erlang#now/0">erlang:now()</seealso>.</p> + <p> + *** POTENTIAL INCOMPATIBILITY ***</p> + <p> + Own Id: OTP-12610</p> + </item> + <item> + <p> + Enable different abort signal from heart</p> + <p>By using environment variable HEART_KILL_SIGNAL, heart + can now use a different signal to kill the old running + Erlang.</p> + <p>By default the signal is SIGKILL but SIGABRT may also + be used by setting environment variable: + HEART_KILL_SIGNAL=SIGABRT</p> + <p> + Own Id: OTP-12613 Aux Id: seq12826 </p> + </item> + <item> + <p> + Update autconf to latest version 2015-03-04</p> + <p> + Own Id: OTP-12646</p> + </item> + <item> + <p> + Optimization of timers internally in the VM. This include + process timers (<c>receive ... after</c>), port timers + (<c>driver_set_timer()</c>) as well as BIF timers + (<c>erlang:send_after()</c>/<c>erlang:start_timer()</c>).</p> + <p> + Each scheduler thread now has its own lock-free timer + service instead of one locked central service. This + dramatically improves performance of timer management on + systems with a large amount of schedulers and timers.</p> + <p> + The timer service internal data structure has also been + optimized to be able to handle more timers than before. + That is, each timer service is by its self able to handle + more timers without dramatic performance loss than the + old centralized timer service.</p> + <p> + The API of BIF timers has also been extended. Timeout + values are for example no longer limited to 32-bit + integers. For more information see the documentation of + <seealso + marker="erlang#start_timer/4"><c>erlang:start_timer/4</c></seealso>, + <seealso + marker="erlang#send_after/4"><c>erlang:send_after/4</c></seealso>, + <seealso + marker="erlang#cancel_timer/2"><c>erlang:cancel_timer/2</c></seealso>, + and <seealso + marker="erlang#read_timer/2"><c>erlang:read_timer/2</c></seealso>.</p> + <p> + Characteristics impact: Calls to the synchronous versions + of <c>erlang:cancel_timer()</c>, and + <c>erlang:read_timer()</c> may take substantially longer + time to complete than before. This occur when the timer + that is accessed is managed by a remote scheduler. You + typically want to use the new asynchronous option in + order to avoid blocking the calling process.</p> + <p> + Own Id: OTP-12650 Aux Id: OTP-11997 </p> + </item> + <item> + <p> + Specialize instructions from common assembler patterns</p> + <p>Specialize common instructions of <c>rem</c>, + <c>band</c>, <c>minus</c> and <c>plus</c> in the beam + loader. This will reduce the number of fetches and thus + lessen the instruction dispatch pressure during runtime + and speed up those operations in some common cases.</p> + <p>Specialize move patterns from x-registers to the stack + with a new <c>move_window</c> instruction. This change + will reduce instruction dispatch pressure.</p> + <p> + Own Id: OTP-12690</p> + </item> + <item> + <p> + Fix cross compilation for Android.</p> + <p> + Own Id: OTP-12693</p> + </item> + <item> + <p> + Fix incorrect use of autoconf macro AC_EGREP_CPP, which + could cause faulty configuration if run from a path + containing the string 'yes'.</p> + <p> + Own Id: OTP-12706</p> + </item> + <item> + <p> + Minimal Java version is now 1.6</p> + <p> + Own Id: OTP-12715</p> + </item> + <item> + <p> + Send format and args on process exit to error_logger</p> + <p> + Previously, the emulator would generate a whole string + with values and call the error_logger passing + <c>"~s~n"</c>. This changes it to a format string + containing <c>~p</c> with the respective values as + arguments.</p> + <p> + Own Id: OTP-12735</p> + </item> + <item> + <p> + Map error logger warnings to warning messages by default.</p> + <p> + Own Id: OTP-12755</p> + </item> + <item> + <p> + Configure architecture ppc64le architecture as a ppc64</p> + <p> + Own Id: OTP-12761</p> + </item> + <item> + <p> + Add function <c>enif_raise_exception</c> to allow a NIF + to raise an error exception with any type of reason.</p> + <p> + Own Id: OTP-12770</p> + </item> + <item> + <p> + Optimized node table statistics retrieval.</p> + <p> + Own Id: OTP-12777</p> + </item> + <item> + <p> + Map beam error logger warnings to warning messages by + default. Previously these messages were mapped to the + error channel by default.</p> + <p> + Own Id: OTP-12781</p> + </item> + <item> + <p> + gen_tcp:shutdown/2 is now asynchronous</p> + <p> + This solves the following problems with the old + implementation:</p> + <p> + It doesn't block when the TCP peer is idle or slow. This + is the expected behaviour when shutdown() is called: the + caller needs to be able to continue reading from the + socket, not be prevented from doing so.</p> + <p> + It doesn't truncate the output. The current version of + gen_tcp:shutdown/2 will truncate any outbound data in the + driver queue after about 10 seconds if the TCP peer is + idle of slow. Worse yet, it doesn't even inform anyone + that the data has been truncated: 'ok' is returned to the + caller; and a FIN rather than an RST is sent to the TCP + peer.</p> + <p> + *** POTENTIAL INCOMPATIBILITY ***</p> + <p> + Own Id: OTP-12797</p> + </item> + <item> + <p> + Introduced delayed node table GC. This in order to avoid + oscillation of entries in and out of the tables. The + oscillation caused unnecessary lock contention on the + table locks. The delay length can be set by passing the + <seealso marker="erl#+zdntgc"><c>+zdntgc</c></seealso> + command line argument.</p> + <p> + Characteristics impact: The tables can grow to very large + sizes with unused entries if the node is get huge amounts + of short lived connections from other nodes. This problem + can be alleviated by shortening the length of the delay + using the <c>+zdntgc</c> command line argument.</p> + <p> + Own Id: OTP-12802</p> + </item> + <item> + <p>Improved implementation of <seealso + marker="erlang#statistics/1"><c>erlang:statistics</c></seealso><c>(io)</c> + in order to reduce contention between schedulers.</p> + <p>Characteristics impact: The actual call to + <c>erlang:statistics(io)</c> takes longer time to + complete, but the overall impact on the system is + improved.</p> + <p> + Own Id: OTP-12842</p> + </item> + <item> + <p> + There are many cases where user code needs to be able to + distinguish between a socket that was closed normally and + one that was aborted. Setting the option + {show_econnreset, true} enables the user to receive + ECONNRESET errors on both active and passive sockets.</p> + <p> + Own Id: OTP-12843</p> + </item> + <item> + <p> + Do not preallocate too large event pool</p> + <p> + A default pool size of 4000 is too excessive for the + common case. This corresponds directly to the number of + threads in the system. Change + ERTS_TS_EV_ALLOC_DEFAULT_POOL_SIZE to 2048. Change + ERTS_TS_EV_ALLOC_POOL_SIZE to 32.</p> + <p> + Own Id: OTP-12849</p> + </item> + </list> + </section> + +</section> + <section><title>Erts 6.4.1</title> <section><title>Fixed Bugs and Malfunctions</title> |