| 1 | Revision history for IO::AIO |
1 | Revision history for IO::AIO |
| 2 | |
2 | |
| 3 | TODO: better autoconf.pm that can return LIBS etc. |
3 | TODO: better autoconf.pm that can return LIBS etc. |
|
|
4 | TODO: aio_cptree/mvtree |
|
|
5 | TODO: reduce condvar fairness: schedule hot-cache-threads first? |
|
|
6 | TODO: splice/tee/vmsplice? (http://kerneltrap.org/node/6505 http://lwn.net/Articles/178199/) |
|
|
7 | TODO: aio_fcntl, at least for file-locking |
|
|
8 | TODO: fallocate, if it wouldn't be so useless. |
|
|
9 | TODO: aio_mincore? |
|
|
10 | TODO: add? also linux-specific stuff? |
|
|
11 | O_EXEC |
|
|
12 | O_SEARCH |
|
|
13 | O_CLOEXEC |
|
|
14 | O_DIRECTORY |
|
|
15 | O_DSYNC |
|
|
16 | O_NOCTTY |
|
|
17 | O_NOFOLLOW |
|
|
18 | O_NONBLOCK |
|
|
19 | O_RSYNC |
|
|
20 | O_SYNC |
|
|
21 | O_TTY_INIT |
| 4 | |
22 | |
| 5 | 2.0 |
23 | O_DIRECT |
|
|
24 | O_LARGEFILE |
|
|
25 | O_NOATIME |
|
|
26 | O_ASYNC |
|
|
27 | TODO: document/test madvise, mprotect |
|
|
28 | TODO: aio_mlock/munlock |
|
|
29 | - do not leak memory in IO::AIO::mmap when the scalar already |
|
|
30 | had string-data. |
|
|
31 | - add O_RDWR, O_APPEND and O_EXCL symbols and semi-documented |
|
|
32 | them. |
|
|
33 | - cache the result IV, for a minor speedup in the common case. |
|
|
34 | - croak when an mmapped-scalar changes location, to detect |
|
|
35 | user-errors better. |
|
|
36 | - add madvise, mmap, munmap, mlockall, munlockall, msync and mtouch |
|
|
37 | to @EXPORT_OK. |
|
|
38 | |
|
|
39 | 3.65 Wed Mar 31 02:45:05 CEST 2010 |
|
|
40 | - actually use PATH_MAX instead of NAME_MAX for readlink, as to not |
|
|
41 | cut off long pathnames in aio_readlink (based on patch by |
|
|
42 | Rasmus Andersson). |
|
|
43 | - a double fork partially killed the event pipe (great testcase |
|
|
44 | by dormando). affects IO::AIO, BDB and Async::Interrupt. |
|
|
45 | - suck steve's dick and rename our symbols to avoid colliding |
|
|
46 | with nonstandard identifier spam on the broken os x pseudo-os. |
|
|
47 | affects IO::AIO and BDB. |
|
|
48 | |
|
|
49 | 3.6 Mon Jan 11 00:43:39 CET 2010 |
|
|
50 | - (libeio) more fixes for the freebsd/apple sendfile - broken by |
|
|
51 | manpage standards, but apparently correct accoridng to actual |
|
|
52 | kernel sources. |
|
|
53 | - add IO::AIO::mmap/munmap support. |
|
|
54 | - add IO::AIO::m(un)lockall support. |
|
|
55 | - clean up manpage. |
|
|
56 | |
|
|
57 | 3.5 Thu Jan 7 21:25:04 CET 2010 |
|
|
58 | - (liebio) fix freebsd sendfile (Vanilla Hsu). |
|
|
59 | - (libeio) also fix darwin sendfile that suffered from a similar bug. |
|
|
60 | - add aio_statvfs as interface to statvfs/fstatvfs. |
|
|
61 | - work around buggy symlink() on cygwin in the testsuite. |
|
|
62 | - wtf. freebsd-8.0 actually passes the testsuite! I knew |
|
|
63 | they could implement fork and semaphores one day! cheers! |
|
|
64 | |
|
|
65 | 3.4 Sat Jan 2 15:13:04 CET 2010 |
|
|
66 | - (libeio) max_poll_time was not properly converted to ticks. |
|
|
67 | - clarify cancel_subs description. |
|
|
68 | - IO::AIO::sendfile did not actually return the return value |
|
|
69 | from the sendfile call. |
|
|
70 | - implement aio_msync, aio_mtouch. |
|
|
71 | - (libeio) tentatively support darwin in sendfile. |
|
|
72 | |
|
|
73 | 3.31 Thu Nov 12 02:14:29 CET 2009 |
|
|
74 | - fix result status documentation of aio_copy, aio_move. |
|
|
75 | - speed up object creation considerably by avoiding |
|
|
76 | hash lookups. |
|
|
77 | |
|
|
78 | 3.3 Wed Aug 5 13:52:58 CEST 2009 |
|
|
79 | - use common::sense. |
|
|
80 | - use common schmorp.h header. |
|
|
81 | - allow integers (file descriptors) in addition to file handles. |
|
|
82 | - take advantage of linux' eventfd (but it seems slower on smp?). |
|
|
83 | - use poll not select on posix platforms. |
|
|
84 | |
|
|
85 | 3.261 Wed Jul 1 10:11:51 CEST 2009 |
|
|
86 | - more 0S X workarounds (patch by Tokuhiro Matsuno). |
|
|
87 | |
|
|
88 | 3.26 Tue Jun 30 09:33:26 CEST 2009 |
|
|
89 | - 0S X of course claims to be posix 2008 but lacks posix_fadvise. |
|
|
90 | try to work around this horribly broken OS in a somewhat hackish |
|
|
91 | way. might help other os's too. |
|
|
92 | |
|
|
93 | 3.25 Sat Jun 27 05:18:26 CEST 2009 |
|
|
94 | - added IO::AIO::fadvise and IO::AIO::sendfile. |
|
|
95 | - (libeio) replaced quicksort+insertion sort by a tuned radix |
|
|
96 | sort + insertion sort, resulting in comparable runtime (usually |
|
|
97 | faster) to the old version, but without any ill side effects on |
|
|
98 | degenerated (for quicksort) data. |
|
|
99 | - (libeio) correctly sort dirs by inodes if we have perfect type |
|
|
100 | knowledge and DIRS_FIRST is used. |
|
|
101 | - (libeio) internally the wrong DT_* constants were used, but no |
|
|
102 | known system has differing EIO_DT_*/DT_* values, so not an issue. |
|
|
103 | - removed a (harmless) assert that was left over on the code |
|
|
104 | but should not have been. |
|
|
105 | - use more correct types (change IO length from IV to VAL64, |
|
|
106 | which makes a difference on systems without 64 bit perls). |
|
|
107 | |
|
|
108 | 3.23 Sat Jun 13 16:57:58 CEST 2009 |
|
|
109 | - fix off-by-one bug in aio_readdir that was introduced in the |
|
|
110 | rushed 3.22 release. |
|
|
111 | |
|
|
112 | 3.22 Sat Jun 13 15:32:40 CEST 2009 |
|
|
113 | - speed up readdirx sort algorithm slightly. |
|
|
114 | - bin/treescan was missing from distro tarball. |
|
|
115 | |
|
|
116 | 3.21 Fri Jun 12 18:45:53 CEST 2009 |
|
|
117 | - new options --dirs and --files for treescan. |
|
|
118 | - install bin/treescan by default. |
|
|
119 | - (libeio) aio_readdir can now be cancelled while executing. |
|
|
120 | - fix a printf format string for 64 bit systems (could lead |
|
|
121 | to problems on big endian 64 bit systems). |
|
|
122 | - do not use qsort() but our own algorithm: glibc initialises |
|
|
123 | the whole locale and I/O subsystem inside qsort, causing |
|
|
124 | a stack overflow on 32 bit machines. The new sort uses much less |
|
|
125 | stack and is more than twice as fast in typical situations. |
|
|
126 | |
|
|
127 | 3.2 Sun Jun 7 20:30:05 CEST 2009 |
|
|
128 | - (libeio) pwrite emulation was even more flawed than intended and did |
|
|
129 | not restore the file offset. |
|
|
130 | - add aio_readdirx, which can return inode and filetype and sort |
|
|
131 | the names in various ways. |
|
|
132 | - unfortunately, utime, chmod, chown on an open file that has just |
|
|
133 | been written can easily block, which caused aio_copy to block the |
|
|
134 | process. no more! |
|
|
135 | - no longer rely on dst path in aio_copy when futime is available. |
|
|
136 | |
|
|
137 | 3.19 Tue Apr 21 22:05:21 CEST 2009 |
|
|
138 | - more perl 5.10 workarounds for aio_read and write. |
|
|
139 | - aio_write no longer modifies the sv (if possible). |
|
|
140 | - aio_read now works correctly with magic values. |
|
|
141 | |
|
|
142 | 3.18 Sun Apr 19 21:17:32 CEST 2009 |
|
|
143 | - better diagnostics when some aio_* functions get passed an illegal fh. |
|
|
144 | - try to avoid crashes due to incompatible 5.10 API changes. grmbl. |
|
|
145 | |
|
|
146 | 3.17 Thu Nov 20 08:45:36 CET 2008 |
|
|
147 | - (libeio) added aio_sync_file_range (untested). |
|
|
148 | - add aio_busy to @IO::AIO::AIO_REQ. |
|
|
149 | |
|
|
150 | 3.16 Wed Oct 22 18:28:01 CEST 2008 |
|
|
151 | - use SvREFCNT_inc instead of SvREFCNT_inc_NN in a non-speed critical |
|
|
152 | part to improve portability to perl 5.8 (reported by szymon). |
|
|
153 | |
|
|
154 | 3.15 Mon Oct 13 00:39:55 CEST 2008 |
|
|
155 | - automatic removal of feeders was broken. |
|
|
156 | - (libeio) use a more robust method to detect whether a feeder |
|
|
157 | has added something to the group or not. |
|
|
158 | |
|
|
159 | 3.1 Thu Oct 2 13:34:40 CEST 2008 |
|
|
160 | - pre-deref the passed callback object, for increased speed |
|
|
161 | and decreased memory usage. |
|
|
162 | - call on_next_submit callback even for aio_group - impact unknown, |
|
|
163 | but seems more correct. |
|
|
164 | - $req->cb now returns the original callback. |
|
|
165 | - (libeio) pass EIO_GROUP/aio_group requests through the queue without |
|
|
166 | requiring a thread switch. |
|
|
167 | - (libeio) status code of aio_futime and aio_utime was always passed 0. |
|
|
168 | - do some other ยต-optimisations. |
|
|
169 | |
|
|
170 | 3.07 Sat Aug 2 16:06:13 CEST 2008 |
|
|
171 | - do not include LIBS in autoconf tests. |
|
|
172 | |
|
|
173 | 3.06 Tue Jul 15 12:41:32 CEST 2008 |
|
|
174 | - move preadwritelock definition before it's first reference, |
|
|
175 | to make it compile again on broken platforms (they still exist) |
|
|
176 | or badly cofngiured perls (redhat...), reported by Rob Bloodgood. |
|
|
177 | |
|
|
178 | 3.05 Thu Jun 19 23:23:52 CEST 2008 |
|
|
179 | - work around perl overriding readdir etc. with thread-unsafe |
|
|
180 | (sic!) versions. |
|
|
181 | |
|
|
182 | 3.04 Wed Jun 18 01:35:38 CEST 2008 |
|
|
183 | - (libeio) fix eio_mknod, which confused it's arguments. |
|
|
184 | - (libeio) do not use readdir_r, as Ulrich Drepper pointed out that this |
|
|
185 | is stupid. |
|
|
186 | - (libeio) fix eio__readahead prototype, patch by Jost Krieger. |
|
|
187 | - (libeio) fix a bug that could potentially cause IO::AIO |
|
|
188 | not to get initialised properly. |
|
|
189 | |
|
|
190 | 3.03 Thu May 29 05:33:30 CEST 2008 |
|
|
191 | - (libeio) correctly call pthread_attr_destroy (). |
|
|
192 | - (libeio) work around broken bsd headers once more. |
|
|
193 | - reduce shared library size again by not including wrappers. |
|
|
194 | - max_outstanding could cause poll to enter a busy-waiting loop. |
|
|
195 | - document the new IO::AIO::poll_cb result value. |
|
|
196 | |
|
|
197 | 3.02 Mon May 12 02:32:02 CEST 2008 |
|
|
198 | - fix a memory leak on aio_readlink. |
|
|
199 | - bring back working fchmod. |
|
|
200 | - nop and busy now set result to 0. |
|
|
201 | - set errno to ENOMEM when allocation fails. |
|
|
202 | |
|
|
203 | 3.01 Sun May 11 03:07:03 CEST 2008 |
|
|
204 | - (libeio) make it compile on systems without readahead or readdir_r. |
|
|
205 | - (libeio) improve configure check for readahead availability. |
|
|
206 | - do not try to link against -lrt. |
|
|
207 | - use a separate configure script for IO::AIO (not the libeio one). |
|
|
208 | |
|
|
209 | 3.0 Sun May 11 00:57:14 CEST 2008 |
|
|
210 | - added bin/treescan to the distribution. |
|
|
211 | - switched to using libeio. |
|
|
212 | - LOTS OF INCOMPATIBLE CHANGES: |
|
|
213 | - remove signal functionality, it is not worth the effort. |
|
|
214 | - max_outstanding no longer returns the previous number of requests. |
|
|
215 | - poll_cb no longer returns number of requests. |
|
|
216 | |
|
|
217 | 2.62 Sat Apr 26 13:59:33 CEST 2008 |
|
|
218 | - port to solaris perls configured for old posix |
|
|
219 | (analysed by Jost Krieger). |
|
|
220 | - keep a reference to the perl filehandle object in aio_close, |
|
|
221 | so it doesn't get closed prematurely, leading to ugly races. |
|
|
222 | |
|
|
223 | 2.61 Wed Apr 16 18:45:02 CEST 2008 |
|
|
224 | - fix treescan output duplication and improve output ordering. |
|
|
225 | also display files given on the commandline. |
|
|
226 | - use a different algorithm for aio_close that is faster |
|
|
227 | and probably has even lower chances of blocking. |
|
|
228 | - do our own stack memory management for threads - linux |
|
|
229 | allocates outrageous amounts of VM (not so bad unless you |
|
|
230 | use mlockall...), which severely limits the number of threads |
|
|
231 | on 32-bit arches: stack size is the larger of PTHREAD_STACK_MIN |
|
|
232 | and 4096 * sizeof (long) (usually this is 16..64k). |
|
|
233 | |
|
|
234 | 2.6 Sun Mar 30 08:28:11 CEST 2008 |
|
|
235 | - added aio_sync. |
|
|
236 | - added aio_pathsync. |
|
|
237 | - fix prototypes of (void) functions. |
|
|
238 | |
|
|
239 | 2.51 Sat Oct 6 16:04:54 CEST 2007 |
|
|
240 | - perlio isn't generally threadsafe, so aio_close could not work reliably. |
|
|
241 | aio_close now tries a safe hack that might not be that asynchronous |
|
|
242 | (see the manpage for details). |
|
|
243 | - discard callback return values as to not grow the stack endlessly |
|
|
244 | when poll_cb handles many requests without returning. |
|
|
245 | - minor code cleanups. |
|
|
246 | |
|
|
247 | 2.5 Thu Oct 4 14:49:08 CEST 2007 |
|
|
248 | - replaced _fd2fh with faster xs code. |
|
|
249 | - aio_close will now try to do "the right thing" and thus might |
|
|
250 | work sensibly for the very first time. |
|
|
251 | |
|
|
252 | 2.41 Mon Sep 24 21:28:21 CEST 2007 |
|
|
253 | - after fork, invest some work to make sure that the poll_fileno |
|
|
254 | stays stable (by dup2'ing the new fd over the old one), to make |
|
|
255 | it easier for programs/libs that don't control their forking |
|
|
256 | behaviour and cannot recreate their watchers. |
|
|
257 | |
|
|
258 | 2.4 Sun Aug 5 18:44:22 CEST 2007 |
|
|
259 | - add aio_truncate, aio_chmod, aio_chown, aio_utime (linux |
|
|
260 | successfully demonstrated that you can block on futimes...). |
|
|
261 | - allow undef as fileoffset for aio_read/write and use read/write(2) |
|
|
262 | internally (useful for sockets or O_APPEND handles). |
|
|
263 | - allow undef for length in aio_write. |
|
|
264 | - negative dataoffsets work as expected now in aio_read/aio_write. |
|
|
265 | - use NV instead of UV for 32 bit perls and file offsets, as NVs |
|
|
266 | have a larger range then. |
|
|
267 | - shared code between BDB and IO::AIO. |
|
|
268 | - aio_busy was completely broken. now fixed. |
|
|
269 | - readahead emulation now returns something. |
|
|
270 | - correctly set errno to ENOSYS on unimplemented functions |
|
|
271 | (should never happen, though). |
|
|
272 | - large changes to make it partially compile and run on win32, |
|
|
273 | but win32 lacks too much functionality, and perl overrides way |
|
|
274 | too many functions with crashing versions. |
|
|
275 | |
|
|
276 | 2.33 Tue Jan 23 23:55:41 CET 2007 |
|
|
277 | - fix install path (Andreas J. Koenig). |
|
|
278 | |
|
|
279 | 2.32 Mon Jan 22 16:56:23 CET 2007 |
|
|
280 | - added aio_rmtree. |
|
|
281 | - wow, aio_mkdir was missing. |
|
|
282 | - aio_load did return undef on error, not -1. |
|
|
283 | - use prefixexp not prefix in autoconf.pm |
|
|
284 | (suggested by ... rt.cpan.org unreachable). |
|
|
285 | - avoid installing autoconf.pm. |
|
|
286 | |
|
|
287 | 2.31 Sat Jan 6 03:46:02 CET 2007 |
|
|
288 | - added aio_load. |
|
|
289 | |
|
|
290 | 2.3 Sat Dec 23 05:48:07 CET 2006 |
|
|
291 | - fix off-by-one bug in aio_read, causing memory corruption |
|
|
292 | or crashes. |
|
|
293 | |
|
|
294 | 2.21 Fri Dec 22 05:03:38 CET 2006 |
|
|
295 | - minor doc updates. |
|
|
296 | - minor cleanups. |
|
|
297 | - IO::AIO::poll returns request count. |
|
|
298 | - undocumented and unfinished async-signal-reporting. |
|
|
299 | |
|
|
300 | 2.2 Tue Oct 31 00:31:00 CET 2006 |
|
|
301 | - minor doc updates. |
|
|
302 | - added aio_readlink. |
|
|
303 | - properly zero-terminate aio_read buffer. |
|
|
304 | - fix aio_read/write with nonzero dataoffset. |
|
|
305 | - reduced size of aio request structure. |
|
|
306 | - better diagnostics when pathnames etc. are not byte-encoded; |
|
|
307 | do not leak request in that case, either. |
|
|
308 | |
|
|
309 | 2.1 Sun Oct 29 02:01:57 CET 2006 |
|
|
310 | - INCOMPATIBLE CHANGE: replace IO::AIO::poll_some by |
|
|
311 | IO::AIO::max_poll_reqs and IO::AIO::max_poll_time. |
|
|
312 | - fix a bug in aio_move where it would fail everytime |
|
|
313 | it would have to copy files manually. |
|
|
314 | - doc updates. |
|
|
315 | - start threads only on (very low) demand. |
|
|
316 | - end superfluous threads automatically after 10s idling. |
|
|
317 | - add IO::AIO::max_idle to control this. |
|
|
318 | - some tuning applied to random places. |
|
|
319 | - add aio_mknod. |
|
|
320 | |
|
|
321 | 2.0 Thu Oct 26 18:27:58 CEST 2006 |
|
|
322 | - minor incompatibility: max_outstanding semantics have changed. |
| 6 | - major rewrite, consider all functionality EXPERIMENTAL. |
323 | - major rewrite, consider all functionality EXPERIMENTAL. |
| 7 | - aio requests can be cancelled. |
324 | - aio requests can be cancelled. |
| 8 | - callbacks can be (re-)set later. |
325 | - callbacks can be (re-)set later. |
| 9 | - aio requests can have choose 9 priorities. |
326 | - aio requests can choose between 9 priorities. |
| 10 | - add aio_group: aio requests can be grouped |
327 | - add aio_group: aio requests can be grouped |
| 11 | into composite requests. |
328 | into composite requests. |
| 12 | - generator interface for groups. |
329 | - generator interface for groups. |
|
|
330 | - add IO::AIO::poll_some. |
| 13 | - major documentation improvements. |
331 | - major documentation improvements. |
| 14 | - very minor bugfixes. |
332 | - very minor bugfixes. |
| 15 | - add aio_nop for dummy requests. |
333 | - add aio_nop for dummy requests. |
| 16 | - add aio_busy mainly for benchmarking and debugging. |
334 | - add aio_busy mainly for benchmarking and debugging. |
| 17 | - deprecated max_outstanding. |
|
|
| 18 | - use "fast" mutexes on linux. |
335 | - use "fast" mutexes on linux. |
| 19 | - use dynamic (and larger) buffers instead of |
336 | - use dynamic (and larger) buffers instead of |
| 20 | putting them onto the stack. |
337 | putting them onto the stack. |
| 21 | - optimise x86 and amd64 a bit. |
338 | - optimise x86 and amd64 a bit. |
| 22 | - better error checking when ENOMEM. |
339 | - better error checking when ENOMEM. |
| 23 | - hopefully fix mem and dirp leaks on fork. |
340 | - hopefully fix mem and dirp leaks on fork. |
|
|
341 | - warn about broken perl malloc. |
|
|
342 | - compiles on cygwin (but pread is broken on cygwin |
|
|
343 | which is not detected by autoconf, so you have to |
|
|
344 | edit autoconf/config.h yourself to enable emulation). |
| 24 | |
345 | |
| 25 | 1.8 Sat Jun 24 22:20:44 CEST 2006 |
346 | 1.8 Sat Jun 24 22:20:44 CEST 2006 |
| 26 | - add and/or document aio_rename, aio_link, aio_symlink and aio_move. |
347 | - add and/or document aio_rename, aio_link, aio_symlink and aio_move. |
| 27 | |
348 | |
| 28 | 1.73 Wed Mar 1 22:49:32 CET 2006 |
349 | 1.73 Wed Mar 1 22:49:32 CET 2006 |
| … | |
… | |
| 62 | - somebody nagged about C89 compatibility, so this release should |
383 | - somebody nagged about C89 compatibility, so this release should |
| 63 | be C89 compatible, but no guarantees for future ones. |
384 | be C89 compatible, but no guarantees for future ones. |
| 64 | - abort when the configure script fails. |
385 | - abort when the configure script fails. |
| 65 | |
386 | |
| 66 | 1.4 Wed Aug 17 08:07:27 CEST 2005 |
387 | 1.4 Wed Aug 17 08:07:27 CEST 2005 |
| 67 | - forget to recreate the pipe after forking, this could cause |
388 | - forgot to recreate the pipe after forking, this could cause |
| 68 | deadlocks. |
389 | deadlocks. |
| 69 | |
390 | |
| 70 | 1.3 Wed Aug 17 07:25:54 CEST 2005 |
391 | 1.3 Wed Aug 17 07:25:54 CEST 2005 |
| 71 | - properly propagate exceptions from callbacks, instead of silently |
392 | - properly propagate exceptions from callbacks, instead of silently |
| 72 | eating them. |
393 | eating them. |
