| 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 |
4 | TODO: aio_cptree/mvtree |
| 5 | TODO: aio_utime, aio_chmod, aio_chown? should simply stat/open first? |
|
|
| 6 | TODO: reduce condvar fairness: schedule hot-cache-threads first? |
5 | TODO: reduce condvar fairness: schedule hot-cache-threads first? |
| 7 | TODO: fix setsig |
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 | |
|
|
11 | TODO: posix_madvise. |
|
|
12 | TODO: OS X spams thread_t into the global namespace, and doesn't even document it. |
|
|
13 | |
|
|
14 | - a double fork partially killed the event pipe (great testcase |
|
|
15 | by dormando). affects IO::AIO, BDB and Async::Interrupt. |
|
|
16 | - actually use PATH_MAX instead of NAME_MAX for readlink, as to not |
|
|
17 | cut off long pathnames in aio_readlink (based on patch by |
|
|
18 | Rasmus Andersson). |
|
|
19 | |
|
|
20 | 3.6 Mon Jan 11 00:43:39 CET 2010 |
|
|
21 | - (libeio) more fixes for the freebsd/apple sendfile - broken by |
|
|
22 | manpage standards, but apparently correct accoridng to actual |
|
|
23 | kernel sources. |
|
|
24 | - add IO::AIO::mmap/munmap support. |
|
|
25 | - add IO::AIO::m(un)lockall support. |
|
|
26 | - clean up manpage. |
|
|
27 | |
|
|
28 | 3.5 Thu Jan 7 21:25:04 CET 2010 |
|
|
29 | - (liebio) fix freebsd sendfile (Vanilla Hsu). |
|
|
30 | - (libeio) also fix darwin sendfile that suffered from a similar bug. |
|
|
31 | - add aio_statvfs as interface to statvfs/fstatvfs. |
|
|
32 | - work around buggy symlink() on cygwin in the testsuite. |
|
|
33 | - wtf. freebsd-8.0 actually passes the testsuite! I knew |
|
|
34 | they could implement fork and semaphores one day! cheers! |
|
|
35 | |
|
|
36 | 3.4 Sat Jan 2 15:13:04 CET 2010 |
|
|
37 | - (libeio) max_poll_time was not properly converted to ticks. |
|
|
38 | - clarify cancel_subs description. |
|
|
39 | - IO::AIO::sendfile did not actually return the return value |
|
|
40 | from the sendfile call. |
|
|
41 | - implement aio_msync, aio_mtouch. |
|
|
42 | - (libeio) tentatively support darwin in sendfile. |
|
|
43 | |
|
|
44 | 3.31 Thu Nov 12 02:14:29 CET 2009 |
|
|
45 | - fix result status documentation of aio_copy, aio_move. |
|
|
46 | - speed up object creation considerably by avoiding |
|
|
47 | hash lookups. |
|
|
48 | |
|
|
49 | 3.3 Wed Aug 5 13:52:58 CEST 2009 |
|
|
50 | - use common::sense. |
|
|
51 | - use common schmorp.h header. |
|
|
52 | - allow integers (file descriptors) in addition to file handles. |
|
|
53 | - take advantage of linux' eventfd (but it seems slower on smp?). |
|
|
54 | - use poll not select on posix platforms. |
|
|
55 | |
|
|
56 | 3.261 Wed Jul 1 10:11:51 CEST 2009 |
|
|
57 | - more 0S X workarounds (patch by Tokuhiro Matsuno). |
|
|
58 | |
|
|
59 | 3.26 Tue Jun 30 09:33:26 CEST 2009 |
|
|
60 | - 0S X of course claims to be posix 2008 but lacks posix_fadvise. |
|
|
61 | try to work around this horribly broken OS in a somewhat hackish |
|
|
62 | way. might help other os's too. |
|
|
63 | |
|
|
64 | 3.25 Sat Jun 27 05:18:26 CEST 2009 |
|
|
65 | - added IO::AIO::fadvise and IO::AIO::sendfile. |
|
|
66 | - (libeio) replaced quicksort+insertion sort by a tuned radix |
|
|
67 | sort + insertion sort, resulting in comparable runtime (usually |
|
|
68 | faster) to the old version, but without any ill side effects on |
|
|
69 | degenerated (for quicksort) data. |
|
|
70 | - (libeio) correctly sort dirs by inodes if we have perfect type |
|
|
71 | knowledge and DIRS_FIRST is used. |
|
|
72 | - (libeio) internally the wrong DT_* constants were used, but no |
|
|
73 | known system has differing EIO_DT_*/DT_* values, so not an issue. |
|
|
74 | - removed a (harmless) assert that was left over on the code |
|
|
75 | but should not have been. |
|
|
76 | - use more correct types (change IO length from IV to VAL64, |
|
|
77 | which makes a difference on systems without 64 bit perls). |
|
|
78 | |
|
|
79 | 3.23 Sat Jun 13 16:57:58 CEST 2009 |
|
|
80 | - fix off-by-one bug in aio_readdir that was introduced in the |
|
|
81 | rushed 3.22 release. |
|
|
82 | |
|
|
83 | 3.22 Sat Jun 13 15:32:40 CEST 2009 |
|
|
84 | - speed up readdirx sort algorithm slightly. |
|
|
85 | - bin/treescan was missing from distro tarball. |
|
|
86 | |
|
|
87 | 3.21 Fri Jun 12 18:45:53 CEST 2009 |
|
|
88 | - new options --dirs and --files for treescan. |
|
|
89 | - install bin/treescan by default. |
|
|
90 | - (libeio) aio_readdir can now be cancelled while executing. |
|
|
91 | - fix a printf format string for 64 bit systems (could lead |
|
|
92 | to problems on big endian 64 bit systems). |
|
|
93 | - do not use qsort() but our own algorithm: glibc initialises |
|
|
94 | the whole locale and I/O subsystem inside qsort, causing |
|
|
95 | a stack overflow on 32 bit machines. The new sort uses much less |
|
|
96 | stack and is more than twice as fast in typical situations. |
|
|
97 | |
|
|
98 | 3.2 Sun Jun 7 20:30:05 CEST 2009 |
|
|
99 | - (libeio) pwrite emulation was even more flawed than intended and did |
|
|
100 | not restore the file offset. |
|
|
101 | - add aio_readdirx, which can return inode and filetype and sort |
|
|
102 | the names in various ways. |
|
|
103 | - unfortunately, utime, chmod, chown on an open file that has just |
|
|
104 | been written can easily block, which caused aio_copy to block the |
|
|
105 | process. no more! |
|
|
106 | - no longer rely on dst path in aio_copy when futime is available. |
|
|
107 | |
|
|
108 | 3.19 Tue Apr 21 22:05:21 CEST 2009 |
|
|
109 | - more perl 5.10 workarounds for aio_read and write. |
|
|
110 | - aio_write no longer modifies the sv (if possible). |
|
|
111 | - aio_read now works correctly with magic values. |
|
|
112 | |
|
|
113 | 3.18 Sun Apr 19 21:17:32 CEST 2009 |
|
|
114 | - better diagnostics when some aio_* functions get passed an illegal fh. |
|
|
115 | - try to avoid crashes due to incompatible 5.10 API changes. grmbl. |
|
|
116 | |
|
|
117 | 3.17 Thu Nov 20 08:45:36 CET 2008 |
|
|
118 | - (libeio) added aio_sync_file_range (untested). |
|
|
119 | - add aio_busy to @IO::AIO::AIO_REQ. |
|
|
120 | |
|
|
121 | 3.16 Wed Oct 22 18:28:01 CEST 2008 |
|
|
122 | - use SvREFCNT_inc instead of SvREFCNT_inc_NN in a non-speed critical |
|
|
123 | part to improve portability to perl 5.8 (reported by szymon). |
|
|
124 | |
|
|
125 | 3.15 Mon Oct 13 00:39:55 CEST 2008 |
|
|
126 | - automatic removal of feeders was broken. |
|
|
127 | - (libeio) use a more robust method to detect whether a feeder |
|
|
128 | has added something to the group or not. |
|
|
129 | |
|
|
130 | 3.1 Thu Oct 2 13:34:40 CEST 2008 |
|
|
131 | - pre-deref the passed callback object, for increased speed |
|
|
132 | and decreased memory usage. |
|
|
133 | - call on_next_submit callback even for aio_group - impact unknown, |
|
|
134 | but seems more correct. |
|
|
135 | - $req->cb now returns the original callback. |
|
|
136 | - (libeio) pass EIO_GROUP/aio_group requests through the queue without |
|
|
137 | requiring a thread switch. |
|
|
138 | - (libeio) status code of aio_futime and aio_utime was always passed 0. |
|
|
139 | - do some other ยต-optimisations. |
|
|
140 | |
|
|
141 | 3.07 Sat Aug 2 16:06:13 CEST 2008 |
|
|
142 | - do not include LIBS in autoconf tests. |
|
|
143 | |
|
|
144 | 3.06 Tue Jul 15 12:41:32 CEST 2008 |
|
|
145 | - move preadwritelock definition before it's first reference, |
|
|
146 | to make it compile again on broken platforms (they still exist) |
|
|
147 | or badly cofngiured perls (redhat...), reported by Rob Bloodgood. |
|
|
148 | |
|
|
149 | 3.05 Thu Jun 19 23:23:52 CEST 2008 |
|
|
150 | - work around perl overriding readdir etc. with thread-unsafe |
|
|
151 | (sic!) versions. |
|
|
152 | |
|
|
153 | 3.04 Wed Jun 18 01:35:38 CEST 2008 |
|
|
154 | - (libeio) fix eio_mknod, which confused it's arguments. |
|
|
155 | - (libeio) do not use readdir_r, as Ulrich Drepper pointed out that this |
|
|
156 | is stupid. |
|
|
157 | - (libeio) fix eio__readahead prototype, patch by Jost Krieger. |
|
|
158 | - (libeio) fix a bug that could potentially cause IO::AIO |
|
|
159 | not to get initialised properly. |
|
|
160 | |
|
|
161 | 3.03 Thu May 29 05:33:30 CEST 2008 |
|
|
162 | - (libeio) correctly call pthread_attr_destroy (). |
|
|
163 | - (libeio) work around broken bsd headers once more. |
|
|
164 | - reduce shared library size again by not including wrappers. |
|
|
165 | - max_outstanding could cause poll to enter a busy-waiting loop. |
|
|
166 | - document the new IO::AIO::poll_cb result value. |
|
|
167 | |
|
|
168 | 3.02 Mon May 12 02:32:02 CEST 2008 |
|
|
169 | - fix a memory leak on aio_readlink. |
|
|
170 | - bring back working fchmod. |
|
|
171 | - nop and busy now set result to 0. |
|
|
172 | - set errno to ENOMEM when allocation fails. |
|
|
173 | |
|
|
174 | 3.01 Sun May 11 03:07:03 CEST 2008 |
|
|
175 | - (libeio) make it compile on systems without readahead or readdir_r. |
|
|
176 | - (libeio) improve configure check for readahead availability. |
|
|
177 | - do not try to link against -lrt. |
|
|
178 | - use a separate configure script for IO::AIO (not the libeio one). |
|
|
179 | |
|
|
180 | 3.0 Sun May 11 00:57:14 CEST 2008 |
|
|
181 | - added bin/treescan to the distribution. |
|
|
182 | - switched to using libeio. |
|
|
183 | - LOTS OF INCOMPATIBLE CHANGES: |
|
|
184 | - remove signal functionality, it is not worth the effort. |
|
|
185 | - max_outstanding no longer returns the previous number of requests. |
|
|
186 | - poll_cb no longer returns number of requests. |
|
|
187 | |
|
|
188 | 2.62 Sat Apr 26 13:59:33 CEST 2008 |
|
|
189 | - port to solaris perls configured for old posix |
|
|
190 | (analysed by Jost Krieger). |
|
|
191 | - keep a reference to the perl filehandle object in aio_close, |
|
|
192 | so it doesn't get closed prematurely, leading to ugly races. |
|
|
193 | |
|
|
194 | 2.61 Wed Apr 16 18:45:02 CEST 2008 |
|
|
195 | - fix treescan output duplication and improve output ordering. |
|
|
196 | also display files given on the commandline. |
|
|
197 | - use a different algorithm for aio_close that is faster |
|
|
198 | and probably has even lower chances of blocking. |
|
|
199 | - do our own stack memory management for threads - linux |
|
|
200 | allocates outrageous amounts of VM (not so bad unless you |
|
|
201 | use mlockall...), which severely limits the number of threads |
|
|
202 | on 32-bit arches: stack size is the larger of PTHREAD_STACK_MIN |
|
|
203 | and 4096 * sizeof (long) (usually this is 16..64k). |
|
|
204 | |
|
|
205 | 2.6 Sun Mar 30 08:28:11 CEST 2008 |
|
|
206 | - added aio_sync. |
|
|
207 | - added aio_pathsync. |
|
|
208 | - fix prototypes of (void) functions. |
|
|
209 | |
|
|
210 | 2.51 Sat Oct 6 16:04:54 CEST 2007 |
|
|
211 | - perlio isn't generally threadsafe, so aio_close could not work reliably. |
|
|
212 | aio_close now tries a safe hack that might not be that asynchronous |
|
|
213 | (see the manpage for details). |
|
|
214 | - discard callback return values as to not grow the stack endlessly |
|
|
215 | when poll_cb handles many requests without returning. |
|
|
216 | - minor code cleanups. |
|
|
217 | |
|
|
218 | 2.5 Thu Oct 4 14:49:08 CEST 2007 |
|
|
219 | - replaced _fd2fh with faster xs code. |
|
|
220 | - aio_close will now try to do "the right thing" and thus might |
|
|
221 | work sensibly for the very first time. |
|
|
222 | |
|
|
223 | 2.41 Mon Sep 24 21:28:21 CEST 2007 |
|
|
224 | - after fork, invest some work to make sure that the poll_fileno |
|
|
225 | stays stable (by dup2'ing the new fd over the old one), to make |
|
|
226 | it easier for programs/libs that don't control their forking |
|
|
227 | behaviour and cannot recreate their watchers. |
|
|
228 | |
|
|
229 | 2.4 Sun Aug 5 18:44:22 CEST 2007 |
|
|
230 | - add aio_truncate, aio_chmod, aio_chown, aio_utime (linux |
|
|
231 | successfully demonstrated that you can block on futimes...). |
|
|
232 | - allow undef as fileoffset for aio_read/write and use read/write(2) |
|
|
233 | internally (useful for sockets or O_APPEND handles). |
|
|
234 | - allow undef for length in aio_write. |
|
|
235 | - negative dataoffsets work as expected now in aio_read/aio_write. |
|
|
236 | - use NV instead of UV for 32 bit perls and file offsets, as NVs |
|
|
237 | have a larger range then. |
|
|
238 | - shared code between BDB and IO::AIO. |
|
|
239 | - aio_busy was completely broken. now fixed. |
|
|
240 | - readahead emulation now returns something. |
|
|
241 | - correctly set errno to ENOSYS on unimplemented functions |
|
|
242 | (should never happen, though). |
|
|
243 | - large changes to make it partially compile and run on win32, |
|
|
244 | but win32 lacks too much functionality, and perl overrides way |
|
|
245 | too many functions with crashing versions. |
| 8 | |
246 | |
| 9 | 2.33 Tue Jan 23 23:55:41 CET 2007 |
247 | 2.33 Tue Jan 23 23:55:41 CET 2007 |
| 10 | - fix install path (Andreas J. Koenig). |
248 | - fix install path (Andreas J. Koenig). |
| 11 | |
249 | |
| 12 | 2.32 Mon Jan 22 16:56:23 CET 2007 |
250 | 2.32 Mon Jan 22 16:56:23 CET 2007 |
