| 1 | Revision history for Perl extension EV and the embedded libev library |
1 | Revision history for Perl extension EV |
|
|
2 | |
|
|
3 | Changes marked with (libev) are changes in libev, and have more |
|
|
4 | documentation in the libev Changes file. |
|
|
5 | |
|
|
6 | 4.28 Tue Nov 19 13:55:39 CET 2019 |
|
|
7 | - (libev) fix ev_port backend, thanks to David H. Gutteridge for |
|
|
8 | reminding me to actually release the fix. |
|
|
9 | - try to name ev.h more explicitly, to hopefully improve portability. |
|
|
10 | - opportunistically round up wait times for poll and epoll backend, |
|
|
11 | to avoid unnecessary loop iterations. |
|
|
12 | - add build dependency on ev_linuxaio.c. |
|
|
13 | - quickly (re)-ported to minix 3.3 before minix crashed again. |
|
|
14 | |
|
|
15 | 4.27 Thu Jun 27 09:39:58 CEST 2019 |
|
|
16 | - (libev) completely rewritten linuxaio backend, maybe |
|
|
17 | usable as a general-use backend. |
|
|
18 | - (libev) use more aggressive assertions to catch |
|
|
19 | more usage errors. |
|
|
20 | - allow users to re-enable assert() in case it is disabled |
|
|
21 | by perl (which is typically the case). |
|
|
22 | |
|
|
23 | 4.26 Mon Jun 24 23:39:40 CEST 2019 |
|
|
24 | - (libev) included experimental linux aio backend. |
|
|
25 | - allow the linux aio backend to be used by default |
|
|
26 | only when explicitly configured during Makefile.PL time. |
|
|
27 | |
|
|
28 | 4.25 Fri Dec 21 08:04:26 CET 2018 |
|
|
29 | - (libev) updated to libev 4.25, minor fixes and enhancements. |
|
|
30 | - document the requirement to "use EV" when using EV::MakeMaker. |
|
|
31 | - only use __register_atfork with glibc and uclibc, as musl |
|
|
32 | defines __linux__, but doesn't implement the linux standard |
|
|
33 | base ABI, nor makes itself detectable via a macro, both apparently |
|
|
34 | by design, winning the "broken by design 2016 award" - well done. |
|
|
35 | - correct EV::periodic example 24h after midnight -> one day after |
|
|
36 | midnight (reported by Felix Ostmann). |
|
|
37 | |
|
|
38 | 4.22 Sun Dec 20 02:34:39 CET 2015 |
|
|
39 | - (libev) when epoll detects unremovable fds in the fd set, rebuild |
|
|
40 | only the epoll descriptor, not the signal pipe, to avoid |
|
|
41 | SIGPIPE in ev_async_send. This doesn't solve it on fork, |
|
|
42 | so document what needs to be done in ev_loop_fork |
|
|
43 | (analyzed by Benjamin Mahler). |
|
|
44 | - (libev) remove superfluous sys/timeb.h include on win32 |
|
|
45 | (analyzed by Jason Madden). |
|
|
46 | - updated libecb. |
|
|
47 | |
|
|
48 | 4.21 Mon Jul 13 21:47:33 CEST 2015 |
|
|
49 | - allow argument in timer->again. |
|
|
50 | - document timer->remaining. |
|
|
51 | - document default repeat value for timer->set. |
|
|
52 | |
|
|
53 | 4.20 Sat Jun 20 13:07:34 CEST 2015 |
|
|
54 | - added stability canary support. |
|
|
55 | - truly rely on automatic configuration for clock_gettime and others. |
|
|
56 | unfortunately, this doesn't help with activeperl and similar perls, |
|
|
57 | which define _POSIX_TIMERS without actually implementing any of it. |
|
|
58 | - (libev, ecb) make it compile as C++ again. |
|
|
59 | - (libev) fix a potential aliasing issue when accessing watcher cbs. |
|
|
60 | |
|
|
61 | 4.18 Sat Sep 6 20:37:23 CEST 2014 |
|
|
62 | - use slightly better weay to find includes, to support multiarch |
|
|
63 | on newer perls. |
|
|
64 | |
|
|
65 | 4.17 Fri Apr 11 06:22:38 CEST 2014 |
|
|
66 | - perl5porters broke Async::Interrupt, BDB, EV, IO::AIO, OpenCL |
|
|
67 | without warning by switching the meaning of USE_SOCKETS_AS_HANDLES |
|
|
68 | in 5.18. What's so attractive about giving a shit about backwards |
|
|
69 | compatibility - I will never understand. |
|
|
70 | |
|
|
71 | 4.16 Sat Mar 8 16:49:05 CET 2014 |
|
|
72 | - make sure EV::CHECK is available at all times. |
|
|
73 | - avoid a spurious warning when perl -w overrides "no warnings". |
|
|
74 | - use perl's signal implementation on WIN32. |
|
|
75 | - ensure extern "C" function pointers are used for externally-visible |
|
|
76 | pointers. |
|
|
77 | - (libev) mark event pipe fd as cloexec after a fork (analyzed by Sami Farin). |
|
|
78 | - (ecb) support m68k, m88k and sh (patch by Miod Vallat). |
|
|
79 | - (libev) in the absence of autoconf, do not use the clock syscall |
|
|
80 | on glibc >= 2.17 (avoids the syscall AND -lrt on systems |
|
|
81 | doing clock_gettime in userspace). |
|
|
82 | - (ecb) work around memory barriers and volatile apparently both being |
|
|
83 | broken in visual studio 2008 and later (analysed and patch by Nicolas Noble). |
|
|
84 | |
|
|
85 | 4.15 Fri Mar 1 12:15:53 CET 2013 |
|
|
86 | - (libev) upgrade to 4.15 - too many changes to list. |
|
|
87 | - EV::run now returns a boolean. |
|
|
88 | - API version 5:1. |
|
|
89 | - document that cleanup watchers are not available via perl. |
|
|
90 | - cast I32 to int in error message printf. |
|
|
91 | - remove dependencies on librt and libpthreads on GNU/Linux. |
|
|
92 | |
|
|
93 | 4.11 Sat Feb 4 19:56:26 CET 2012 |
|
|
94 | - (libev) implement memory fences for (obsolete) llvm-gcc. |
|
|
95 | |
|
|
96 | 4.10 Thu Jan 19 18:54:23 CET 2012 |
|
|
97 | - (libev) fix a race where the workaround against the epoll fork |
|
|
98 | bugs caused signals to not be handled anymore. |
|
|
99 | - (libev) correct backend_fudge for most backends, and implement |
|
|
100 | a windows specific workaround to avoid looping because we call |
|
|
101 | both select and Sleep, both with different time resolutions. |
|
|
102 | - e_new wasn't declared static (causing very minor .so bloat). |
|
|
103 | - replace more old api names by new ones. |
|
|
104 | |
|
|
105 | 4.03 Tue Jan 11 14:51:05 CET 2011 |
|
|
106 | - do not avoid the clock_gettime call on GNU/Linux anymore, as EV |
|
|
107 | links against -lpthread anyways - as a result, EV might now take |
|
|
108 | advantage of fast userspace clock_gettime implementations, |
|
|
109 | but also links against -lrt. |
|
|
110 | - (libev) lots of event port bug workarounds. |
|
|
111 | - (libev) officially support files in I/O watchers. |
|
|
112 | - (libev) new function ev_feed_signal. |
|
|
113 | - fix documentation parts still refering to the 3.x API. |
|
|
114 | |
|
|
115 | 4.02 Thu Dec 30 08:27:41 CET 2010 |
|
|
116 | - the revents argument did not stringify correctly, as only the |
|
|
117 | numeric value was updated, while the string value was kept from |
|
|
118 | previous invocations. |
|
|
119 | |
|
|
120 | 4.01 Sun Dec 5 12:42:13 CET 2010 |
|
|
121 | - fully support EV_COMPAT3=0. |
|
|
122 | - default_fork was stupidly defined as inline. |
|
|
123 | - ask cpan to upgrade AnyEvent if < 5.29. |
|
|
124 | - support EV_EXTRA_DEFS during configuration. |
|
|
125 | - support -DEV_NO_LOOPS for snakker build. |
|
|
126 | |
|
|
127 | 4.00 Mon Oct 25 13:30:09 CEST 2010 |
|
|
128 | - many API changes, see the manual. |
|
|
129 | - (libev) lots and lots of bugfixes, see the ev documentation. |
|
|
130 | - fix a bug where inotify usage would parse the same event |
|
|
131 | multiple times, causing various forms of breakage. |
|
|
132 | - greatly reduce stack usage for inotify (8kb to <0.5kb). |
|
|
133 | - expose ev_depth and ev_verify via the XS API. |
|
|
134 | - implement ev_cleanup watchers. |
|
|
135 | - (libev) ev_embed_stop did not correctly stop the watcher. |
|
|
136 | - (libev) disable poll backend on AIX. |
|
|
137 | - (libev) rename EV_TIMEOUT to EV_TIMER. |
|
|
138 | - (libev) add section on accept() problems to the manpage. |
|
|
139 | - (libev) no child watchers on win32. |
|
|
140 | - make code more aliasing compliant, in case perl is ever |
|
|
141 | translated to C. |
|
|
142 | - document the EV::CHECK runtime unavailability. |
|
|
143 | - ported to minix 3.1.7. |
|
|
144 | |
|
|
145 | 3.9 Thu Dec 31 07:59:59 CET 2009 |
|
|
146 | - disable t/07* under automatic testing. |
|
|
147 | - increase t/09* timeouts as netbsd has *horribly* broken |
|
|
148 | select/kevent that *usually* sleep >>0.6s instead of 0.3. |
|
|
149 | - add constants EV::FLAG_NOSIGFD, EV::FLAG_NOINOTIFY and |
|
|
150 | EV::BACKEND_ALL. |
|
|
151 | - (libev) signalfd is no longer used by default. |
|
|
152 | - (libev) backport inotify code to C89. |
|
|
153 | - (libev) inotify file descriptors could leak into child processes. |
|
|
154 | - (libev) ev_stat watchers could keep an errornous extra ref on the loop. |
|
|
155 | - (libev) take advantage of inotify_init1, if available. |
|
|
156 | - (libev) the signal handling pipe wasn't always initialised under windows. |
|
|
157 | - changed minimum glibc requirement from glibc 2.9 to 2.7, for signalfd. |
|
|
158 | - (libev) only replace ev_stat.prev when we detect an actual difference. |
|
|
159 | |
|
|
160 | 3.8 Sun Aug 9 15:30:10 CEST 2009 |
|
|
161 | - implement $loop->signal/signal_ns. |
|
|
162 | - (libev) incompatible change: do not necessarily reset signal |
|
|
163 | handler to SIG_DFL when a sighandler is stopped. |
|
|
164 | - (libev) ev_default_destroy did not properly free or zero some |
|
|
165 | members, potentially causing crashes and memory corruption on |
|
|
166 | repated ev_default_destroy/ev_default_loop calls. |
|
|
167 | - (libev) take advantage of signalfd on GNU/Linux systems. |
|
|
168 | - (libev) document that the signal mask might be in an unspecified |
|
|
169 | state when using libev's signal handling. |
|
|
170 | - (libev) take advantage of some GNU/Linux calls to set |
|
|
171 | cloexec/nonblock on fd creation, to avoid race conditions. |
|
|
172 | - implement internal glue code to interface more efficiently |
|
|
173 | with AnyEvent. |
|
|
174 | |
|
|
175 | 3.7 Fri Jul 17 16:49:16 CEST 2009 |
|
|
176 | - add EV::loop_verify and EV::loop_depth. |
|
|
177 | - use output filehandle for i/o watchers waiting |
|
|
178 | for EV_WRITE, otherwise input filehandle. |
|
|
179 | - use common schmorp.h header. |
|
|
180 | - add EV::Timer->remaining. |
|
|
181 | - allow for subclassing of EV::Loop (see the exciting |
|
|
182 | EV::Loop::Async module for an example). |
|
|
183 | - added EV::invoke_pending and EV::pending_count. |
|
|
184 | |
|
|
185 | - (libev) ev_unloop and ev_loop wrongly used a global variable to |
|
|
186 | exit loops, instead of using a per-loop variable. |
|
|
187 | - (libev) the ev_set_io_collect_interval interpretation has changed. |
|
|
188 | - add new functionality: EV::invoke_pending, EV::pending_count. |
|
|
189 | - add $timer->remaining. |
|
|
190 | - add EV::loop_depth. |
|
|
191 | - (libev) calling unloop in fork/prepare watchers will no longer poll |
|
|
192 | for new events. |
|
|
193 | - (libev) use GetSystemTimeAsFileTime instead of _timeb on windows, |
|
|
194 | for slightly higher accuracy. |
|
|
195 | - (libev) actually 0-initialise struct sigaction when installing |
|
|
196 | signals. |
|
|
197 | |
|
|
198 | 3.6 Tue Apr 28 02:50:37 CEST 2009 |
|
|
199 | - keepalive(0) could decrease the refcount of the loop |
|
|
200 | permanently. |
|
|
201 | - add ev_suspend/ev_resume, also make ev_now_update |
|
|
202 | accessible via the XS API. |
|
|
203 | - most EV::<eventtype> constants were missing and have been |
|
|
204 | added :/. |
|
|
205 | - add EV::VERSION_MAJOR/VERSION_MINOR constants. |
|
|
206 | - (libev) multiple timers becoming ready within an event loop |
|
|
207 | iteration will be invoked in the "correct" order now. |
|
|
208 | - (libev) do not leave the event loop early just because we have |
|
|
209 | no active watchers, fixing a problem when embedding a kqueue |
|
|
210 | loop that has active kernel events but no registered watchers |
|
|
211 | (reported by blacksand blacksand). |
|
|
212 | - (libev) correctly zero the idx values for arrays, so destroying |
|
|
213 | and reinitialising the default loop actually works (patch by |
|
|
214 | Malek Hadj-Ali). |
|
|
215 | - (libev) new EV::CUSTOM revents flag for use by applications. |
|
|
216 | - (libev) add documentation section about priorites. |
|
|
217 | - (libev) add a glossary to the dcoumentation. |
|
|
218 | - (libev) extend the ev_fork description slightly. |
|
|
219 | - (libev) optimize a jump out of call_pending. |
|
|
220 | - t/03_keepalive could fail when there was no actual error. |
|
|
221 | |
|
|
222 | 3.53 Sun Feb 15 02:38:20 CET 2009 |
|
|
223 | - (libev) on win32, the event loop creation could randomly fail |
|
|
224 | due to an initialised variable having the wrong value. |
|
|
225 | - (libev) probe for CLOCK_REALTIME support at runtime as well and |
|
|
226 | fall back to gettimeofday if there is an error, to support older |
|
|
227 | operating systems with newer header files/libraries. |
|
|
228 | - prefer gettimeofday over clock_gettime by default. |
|
|
229 | |
|
|
230 | 3.52 Wed Jan 7 21:46:14 CET 2009 |
|
|
231 | - (libev) fix some issues in the select backend when in fd_set mode. |
|
|
232 | - (libev) due to a thinko, instead of disabling everything but |
|
|
233 | select on the borked OS X platform, everything but select was |
|
|
234 | allowed (reported by Emanuele Giaquinta). |
|
|
235 | - (libev) actually verify that local and remote port are matching in |
|
|
236 | libev's socketpair emulation, which makes denial-of-service |
|
|
237 | attacks harder (but not impossible - it's windows). Make sure |
|
|
238 | it even works under vista, which thinks that getpeer/sockname |
|
|
239 | should return fantasy port numbers. |
|
|
240 | |
|
|
241 | 3.51 Wed Dec 24 23:01:59 CET 2008 |
|
|
242 | - do not cache the arguments passed to callbacks if the refcount |
|
|
243 | indicates that the callback has stolen them. |
|
|
244 | - (libev) try to avoid librt on GNU/Linux. |
|
|
245 | - (libev) check port_getn return value dfferently, might potentially |
|
|
246 | avoid problems. |
|
|
247 | - (libev) fix a bug with stat watchers possibly causing freezes. |
|
|
248 | - (libev) work around OS X 10.5 breaking poll, now select is the |
|
|
249 | only viable choice left on that pile of garbage. |
|
|
250 | - play tester whore: disable some tests that typically fail only |
|
|
251 | on cpan tester machines. |
|
|
252 | |
|
|
253 | 3.49 Wed Nov 19 11:26:53 CET 2008 |
|
|
254 | - fix typos in manpage (Alex Efros). |
|
|
255 | - increase timing even further, for the ever-overloaded cpan-tester |
|
|
256 | machines, and to the detriment of everybody else who wants a fast |
|
|
257 | make test. |
|
|
258 | - possible 5.6 compatibility. |
|
|
259 | - (libev) use inotify even on buggy kernels, but don't rely on it. |
|
|
260 | - (libev) use inotify only as an added hint on network filesystems. |
|
|
261 | |
|
|
262 | 3.48 Thu Oct 30 09:09:48 CET 2008 |
|
|
263 | - (libev) use a generation counter to detect spurious epoll events |
|
|
264 | and recreate the kernel in such a case (sorry, it's slow, but |
|
|
265 | I didn't design epoll...). |
|
|
266 | - (libev) optimise away an EPOLL_CTL_ADD/MOD combo in the epoll |
|
|
267 | backend in some cases. |
|
|
268 | - (libev) use memset to initialise most arrays now and do away |
|
|
269 | with the init functions. |
|
|
270 | - speed up event callback invocation further (~15%), by |
|
|
271 | not creating the object reference each time. |
|
|
272 | - EV::sleep was documented, but not implemented. fun. |
|
|
273 | |
|
|
274 | 3.45 Tue Oct 21 22:20:39 CEST 2008 |
|
|
275 | - (libev) disable inotify usage on linux <2.6.25 (kernel bug). |
|
|
276 | - (libev) ev_embed will now automatically follow fork. |
|
|
277 | - (libev) ev_once will now pass both io and timeout events to |
|
|
278 | the callback when both happen concurrently, instead of giving |
|
|
279 | one precedence. |
|
|
280 | - install C library documentation as EV::libev manpage (sorry, oesi). |
|
|
281 | - (libev) fix a minor performance bug in ev_stat handling. |
|
|
282 | |
|
|
283 | 3.44 Mon Sep 29 05:16:31 CEST 2008 |
|
|
284 | - (libev) add EV::now_update and $loop->now_update. |
| 2 | |
285 | |
| 3 | 3.431 Sun Jul 13 00:19:02 CEST 2008 |
286 | 3.431 Sun Jul 13 00:19:02 CEST 2008 |
| 4 | - made the EV::embed callback optional again. reported by |
287 | - made the EV::embed callback optional again. reported by |
| 5 | Vladimir Timofeev. |
288 | Vladimir Timofeev. |
| 6 | |
289 | |
