| 1 | Revision history for IO::AIO |
1 | Revision history for IO::AIO |
| 2 | |
2 | |
| 3 | TODO: better autoconf.pm that can return LIBS etc. |
|
|
| 4 | TODO: aio_cptree/mvtree |
3 | 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? |
4 | TODO: reduce condvar fairness: schedule hot-cache-threads first? |
| 7 | TODO: fix setsig |
5 | TODO: vmsplice? (http://kerneltrap.org/node/6505 http://lwn.net/Articles/178199/) |
|
|
6 | TODO: aio_fcntl, at least for file-locking |
|
|
7 | TODO: aio_mincore? |
|
|
8 | TODO: getxattr etc.? |
|
|
9 | TODO: F_DUPFD_CLOEXEC |
|
|
10 | TODO: emulation for splice? |
|
|
11 | TODO: eio_mmap|mlock|munmap|splice... |
|
|
12 | |
|
|
13 | - make fh argument of IO::AIO::mmap optional, also add mmap/munmap |
|
|
14 | to quick overview. |
|
|
15 | - splice can corrupt data. |
|
|
16 | - (libeio) remove pread/pwrite emulation. |
|
|
17 | - do not grow mmaped scalars in aio_read. |
|
|
18 | - add FALLOC_FL_PUNCH_HOLE and document fallocate. |
|
|
19 | - fix a crash in aio_fiemap, when used on files without any |
|
|
20 | extents. |
|
|
21 | |
|
|
22 | 4.15 Tue Apr 10 06:59:00 CEST 2012 |
|
|
23 | - always include linux/types.h for fiemap, for compatibility to |
|
|
24 | ancient systems (Paul Howarth). |
|
|
25 | - experimental support for IO::AIO::splice and ::tee (no aio_...). |
|
|
26 | - provide SEEK_HOLE and SEEK_DATA, if available. |
|
|
27 | - work around (again!) an immensely stupid bug in RHEL, defining |
|
|
28 | autoconf macros in linux system headers (analysed by Paul Howarth). |
|
|
29 | |
|
|
30 | 4.14 Sat Apr 7 02:45:18 CEST 2012 |
|
|
31 | - fix stat structure usage on windows, which caused bogus stat results. |
|
|
32 | - (libeio) make readahead emulation behave more like actual readahead by never failing. |
|
|
33 | - new request aio_seek. |
|
|
34 | - new request aio_fiemap. |
|
|
35 | - autogenerate the #ifdef/#define 0 blocks for symbols we export. |
|
|
36 | |
|
|
37 | 4.12 Fri Dec 30 08:51:25 CET 2011 |
|
|
38 | - realpath would return a random value if the name was too long, |
|
|
39 | instead of -1. |
|
|
40 | - port to c89. |
|
|
41 | |
|
|
42 | 4.11 Mon Oct 10 00:24:11 CEST 2011 |
|
|
43 | - libeio didn't compile on !linux (Paul Howarth). |
|
|
44 | |
|
|
45 | 4.1 Sun Oct 9 10:24:11 CEST 2011 |
|
|
46 | - IO::AIO did access uninitialised memory on unsuccessful stats. |
|
|
47 | - (libeio) added syncfs syscall wrapper. |
|
|
48 | - (libeio) set thread name on linux (ps -L/Hcx, top, gdb). |
|
|
49 | - (libeio) support multiple indepenent working directories. |
|
|
50 | - applied speling corrections by Nicholas Bamber. |
|
|
51 | - tune treescan #threads and #outstanding requests a bit. |
|
|
52 | - reduce the number of file descriptors in use at any one time |
|
|
53 | in treescan, to avoid running out of them (and using too much ram). |
|
|
54 | - take advantage of working directoriy abstractions |
|
|
55 | in aio_scandir and treescan. |
|
|
56 | - reduce compiled size by ~10% by not inlining some key functionality. |
|
|
57 | - added --progress switch to treescan. |
|
|
58 | |
|
|
59 | 4.0 Mon Jul 18 05:01:10 CEST 2011 |
|
|
60 | - INCOMPATIBLE CHANGE: fork is no longer supported (indeed, it never was), |
|
|
61 | see FORK BEHAVIOUR in manpage for details. |
|
|
62 | - passes testsuite on win32 now and may actually work |
|
|
63 | (activestate 5.10.1 / MSVC6, YMMV). |
|
|
64 | - (libeio) fix a deadlock where a wakeup signal could be missed when |
|
|
65 | a timeout occured at the same time. |
|
|
66 | - (libeio) added realpath. |
|
|
67 | - (libeio) added fallocate. |
|
|
68 | - (libeio) disabling sendfile on darwin unfortunately broke the emulation. |
|
|
69 | - (libeio) do not acquire any locks when forking. |
|
|
70 | - (libeio) use fewer time() syscalls when waiting for new requests. |
|
|
71 | |
|
|
72 | 3.93 Wed Jun 29 23:44:18 CEST 2011 |
|
|
73 | - ECB.H WAS MISSING. |
|
|
74 | |
|
|
75 | 3.92 Wed Jun 29 14:45:41 CEST 2011 |
|
|
76 | - ecb.h was missing. |
|
|
77 | |
|
|
78 | 3.91 Wed Jun 29 13:24:42 CEST 2011 |
|
|
79 | - (libeio) work around a Linux (and likely FreeBSD and other |
|
|
80 | kernels) bug where sendfile would not transfer all the requested |
|
|
81 | bytes on large transfers, using a heuristic. |
|
|
82 | - buggy sendfile caused aio_move/copy to sometimes fail for big |
|
|
83 | files (fortunately it checks that the whole file has been |
|
|
84 | transferred...) |
|
|
85 | - use libecb for higher performance and higher portability. |
|
|
86 | - (libeio) disable sendfile on darwin, it's too broken. |
|
|
87 | - disable fork tests on !linux, as only linux supports mixing |
|
|
88 | pthread and fork in perl. |
|
|
89 | - document the fact that fork doesn't work anymore when using |
|
|
90 | this module, on anything !GNU/Linux. |
|
|
91 | - increase timeout in t/04_fork.t, as too many CPAN-tester setups |
|
|
92 | run on a 0.1MHz cpu. Or so. |
|
|
93 | |
|
|
94 | 3.9 Fri May 27 02:43:47 CEST 2011 |
|
|
95 | - (libeio) fix memory corruption in aio_readdirx for the flags |
|
|
96 | combination READDIR_STAT_ORDER | READDIR_DIRS_FIRST. |
|
|
97 | - add lots of newer POSIX and GNU/Linux-specific open |
|
|
98 | flags. |
|
|
99 | |
|
|
100 | 3.8 Sun Mar 27 12:25:33 CEST 2011 |
|
|
101 | - use nonstandard but maybe-working-on-bsd fork technique. |
|
|
102 | - support a max_idle value of 0. |
|
|
103 | - support setting of idle timeout value (IO::AIO::idle_timeout). |
|
|
104 | |
|
|
105 | 3.72 Fri Feb 11 04:25:38 CET 2011 |
|
|
106 | - use _POSIX_MEMLOCK_RANGE to detect mlock/munlock. |
|
|
107 | - aio_mknod always used a dev_t value of 0. |
|
|
108 | - new treescan option: --grep. |
|
|
109 | - add more S_IF macros, and major/minor/makedev "macros". |
|
|
110 | |
|
|
111 | 3.71 Thu Dec 30 08:18:46 CET 2010 |
|
|
112 | - the numerical result value passed to callbacks did not stringify |
|
|
113 | correctly, due to internal reuse and failure to reset the sv flags. |
|
|
114 | - actually test for posix_[mf]advise, as at least uClibc defines |
|
|
115 | _POSIX_ADVISORY_INFO without actually having any of the required |
|
|
116 | funcitonality. ugh. |
|
|
117 | |
|
|
118 | 3.7 Mon Nov 1 23:00:34 CET 2010 |
|
|
119 | - implement/add madvise, mmap, munmap, aio_mlockall, munlockall, |
|
|
120 | aio_mlock, munlock, msync and mtouch to @EXPORT_OK. |
|
|
121 | - document the sad state of affairs w.r.t. pthread on many bsds. |
|
|
122 | - do not enable mmap on systems without _POSIX_MAPPED_FILES |
|
|
123 | (openbsd 4.8). |
|
|
124 | - do not leak memory in IO::AIO::mmap when the scalar already |
|
|
125 | had string-data. |
|
|
126 | - add O_RDWR, O_APPEND and O_EXCL symbols and semi-document |
|
|
127 | them. |
|
|
128 | - cache the result IV, for a minor speedup in the common case. |
|
|
129 | - croak when an mmapped-scalar changes location, to detect |
|
|
130 | user-errors better. |
|
|
131 | - fix aio_readlink prototype. |
|
|
132 | |
|
|
133 | 3.65 Wed Mar 31 02:45:05 CEST 2010 |
|
|
134 | - actually use PATH_MAX instead of NAME_MAX for readlink, as to not |
|
|
135 | cut off long pathnames in aio_readlink (based on patch by |
|
|
136 | Rasmus Andersson). |
|
|
137 | - a double fork partially killed the event pipe (great testcase |
|
|
138 | by dormando). affects IO::AIO, BDB and Async::Interrupt. |
|
|
139 | - suck steve's dick and rename our symbols to avoid colliding |
|
|
140 | with nonstandard identifier spam on the broken os x pseudo-os. |
|
|
141 | affects IO::AIO and BDB. |
|
|
142 | |
|
|
143 | 3.6 Mon Jan 11 00:43:39 CET 2010 |
|
|
144 | - (libeio) more fixes for the freebsd/apple sendfile - broken by |
|
|
145 | manpage standards, but apparently correct accoridng to actual |
|
|
146 | kernel sources. |
|
|
147 | - add IO::AIO::mmap/munmap support. |
|
|
148 | - add IO::AIO::m(un)lockall support. |
|
|
149 | - clean up manpage. |
|
|
150 | |
|
|
151 | 3.5 Thu Jan 7 21:25:04 CET 2010 |
|
|
152 | - (liebio) fix freebsd sendfile (Vanilla Hsu). |
|
|
153 | - (libeio) also fix darwin sendfile that suffered from a similar bug. |
|
|
154 | - add aio_statvfs as interface to statvfs/fstatvfs. |
|
|
155 | - work around buggy symlink() on cygwin in the testsuite. |
|
|
156 | - wtf. freebsd-8.0 actually passes the testsuite! I knew |
|
|
157 | they could implement fork and semaphores one day! cheers! |
|
|
158 | |
|
|
159 | 3.4 Sat Jan 2 15:13:04 CET 2010 |
|
|
160 | - (libeio) max_poll_time was not properly converted to ticks. |
|
|
161 | - clarify cancel_subs description. |
|
|
162 | - IO::AIO::sendfile did not actually return the return value |
|
|
163 | from the sendfile call. |
|
|
164 | - implement aio_msync, aio_mtouch. |
|
|
165 | - (libeio) tentatively support darwin in sendfile. |
|
|
166 | |
|
|
167 | 3.31 Thu Nov 12 02:14:29 CET 2009 |
|
|
168 | - fix result status documentation of aio_copy, aio_move. |
|
|
169 | - speed up object creation considerably by avoiding |
|
|
170 | hash lookups. |
|
|
171 | |
|
|
172 | 3.3 Wed Aug 5 13:52:58 CEST 2009 |
|
|
173 | - use common::sense. |
|
|
174 | - use common schmorp.h header. |
|
|
175 | - allow integers (file descriptors) in addition to file handles. |
|
|
176 | - take advantage of linux' eventfd (but it seems slower on smp?). |
|
|
177 | - use poll not select on posix platforms. |
|
|
178 | |
|
|
179 | 3.261 Wed Jul 1 10:11:51 CEST 2009 |
|
|
180 | - more 0S X workarounds (patch by Tokuhiro Matsuno). |
|
|
181 | |
|
|
182 | 3.26 Tue Jun 30 09:33:26 CEST 2009 |
|
|
183 | - 0S X of course claims to be posix 2008 but lacks posix_fadvise. |
|
|
184 | try to work around this horribly broken OS in a somewhat hackish |
|
|
185 | way. might help other os's too. |
|
|
186 | |
|
|
187 | 3.25 Sat Jun 27 05:18:26 CEST 2009 |
|
|
188 | - added IO::AIO::fadvise and IO::AIO::sendfile. |
|
|
189 | - (libeio) replaced quicksort+insertion sort by a tuned radix |
|
|
190 | sort + insertion sort, resulting in comparable runtime (usually |
|
|
191 | faster) to the old version, but without any ill side effects on |
|
|
192 | degenerated (for quicksort) data. |
|
|
193 | - (libeio) correctly sort dirs by inodes if we have perfect type |
|
|
194 | knowledge and DIRS_FIRST is used. |
|
|
195 | - (libeio) internally the wrong DT_* constants were used, but no |
|
|
196 | known system has differing EIO_DT_*/DT_* values, so not an issue. |
|
|
197 | - removed a (harmless) assert that was left over on the code |
|
|
198 | but should not have been. |
|
|
199 | - use more correct types (change IO length from IV to VAL64, |
|
|
200 | which makes a difference on systems without 64 bit perls). |
|
|
201 | |
|
|
202 | 3.23 Sat Jun 13 16:57:58 CEST 2009 |
|
|
203 | - fix off-by-one bug in aio_readdir that was introduced in the |
|
|
204 | rushed 3.22 release. |
|
|
205 | |
|
|
206 | 3.22 Sat Jun 13 15:32:40 CEST 2009 |
|
|
207 | - speed up readdirx sort algorithm slightly. |
|
|
208 | - bin/treescan was missing from distro tarball. |
|
|
209 | |
|
|
210 | 3.21 Fri Jun 12 18:45:53 CEST 2009 |
|
|
211 | - new options --dirs and --files for treescan. |
|
|
212 | - install bin/treescan by default. |
|
|
213 | - (libeio) aio_readdir can now be cancelled while executing. |
|
|
214 | - fix a printf format string for 64 bit systems (could lead |
|
|
215 | to problems on big endian 64 bit systems). |
|
|
216 | - do not use qsort() but our own algorithm: glibc initialises |
|
|
217 | the whole locale and I/O subsystem inside qsort, causing |
|
|
218 | a stack overflow on 32 bit machines. The new sort uses much less |
|
|
219 | stack and is more than twice as fast in typical situations. |
|
|
220 | |
|
|
221 | 3.2 Sun Jun 7 20:30:05 CEST 2009 |
|
|
222 | - (libeio) pwrite emulation was even more flawed than intended and did |
|
|
223 | not restore the file offset. |
|
|
224 | - add aio_readdirx, which can return inode and filetype and sort |
|
|
225 | the names in various ways. |
|
|
226 | - unfortunately, utime, chmod, chown on an open file that has just |
|
|
227 | been written can easily block, which caused aio_copy to block the |
|
|
228 | process. no more! |
|
|
229 | - no longer rely on dst path in aio_copy when futime is available. |
|
|
230 | |
|
|
231 | 3.19 Tue Apr 21 22:05:21 CEST 2009 |
|
|
232 | - more perl 5.10 workarounds for aio_read and write. |
|
|
233 | - aio_write no longer modifies the sv (if possible). |
|
|
234 | - aio_read now works correctly with magic values. |
|
|
235 | |
|
|
236 | 3.18 Sun Apr 19 21:17:32 CEST 2009 |
|
|
237 | - better diagnostics when some aio_* functions get passed an illegal fh. |
|
|
238 | - try to avoid crashes due to incompatible 5.10 API changes. grmbl. |
|
|
239 | |
|
|
240 | 3.17 Thu Nov 20 08:45:36 CET 2008 |
|
|
241 | - (libeio) added aio_sync_file_range (untested). |
|
|
242 | - add aio_busy to @IO::AIO::AIO_REQ. |
|
|
243 | |
|
|
244 | 3.16 Wed Oct 22 18:28:01 CEST 2008 |
|
|
245 | - use SvREFCNT_inc instead of SvREFCNT_inc_NN in a non-speed critical |
|
|
246 | part to improve portability to perl 5.8 (reported by szymon). |
|
|
247 | |
|
|
248 | 3.15 Mon Oct 13 00:39:55 CEST 2008 |
|
|
249 | - automatic removal of feeders was broken. |
|
|
250 | - (libeio) use a more robust method to detect whether a feeder |
|
|
251 | has added something to the group or not. |
|
|
252 | |
|
|
253 | 3.1 Thu Oct 2 13:34:40 CEST 2008 |
|
|
254 | - pre-deref the passed callback object, for increased speed |
|
|
255 | and decreased memory usage. |
|
|
256 | - call on_next_submit callback even for aio_group - impact unknown, |
|
|
257 | but seems more correct. |
|
|
258 | - $req->cb now returns the original callback. |
|
|
259 | - (libeio) pass EIO_GROUP/aio_group requests through the queue without |
|
|
260 | requiring a thread switch. |
|
|
261 | - (libeio) status code of aio_futime and aio_utime was always passed 0. |
|
|
262 | - do some other ยต-optimisations. |
|
|
263 | |
|
|
264 | 3.07 Sat Aug 2 16:06:13 CEST 2008 |
|
|
265 | - do not include LIBS in autoconf tests. |
|
|
266 | |
|
|
267 | 3.06 Tue Jul 15 12:41:32 CEST 2008 |
|
|
268 | - move preadwritelock definition before it's first reference, |
|
|
269 | to make it compile again on broken platforms (they still exist) |
|
|
270 | or badly cofngiured perls (redhat...), reported by Rob Bloodgood. |
|
|
271 | |
|
|
272 | 3.05 Thu Jun 19 23:23:52 CEST 2008 |
|
|
273 | - work around perl overriding readdir etc. with thread-unsafe |
|
|
274 | (sic!) versions. |
|
|
275 | |
|
|
276 | 3.04 Wed Jun 18 01:35:38 CEST 2008 |
|
|
277 | - (libeio) fix eio_mknod, which confused it's arguments. |
|
|
278 | - (libeio) do not use readdir_r, as Ulrich Drepper pointed out that this |
|
|
279 | is stupid. |
|
|
280 | - (libeio) fix eio__readahead prototype, patch by Jost Krieger. |
|
|
281 | - (libeio) fix a bug that could potentially cause IO::AIO |
|
|
282 | not to get initialised properly. |
|
|
283 | |
|
|
284 | 3.03 Thu May 29 05:33:30 CEST 2008 |
|
|
285 | - (libeio) correctly call pthread_attr_destroy (). |
|
|
286 | - (libeio) work around broken bsd headers once more. |
|
|
287 | - reduce shared library size again by not including wrappers. |
|
|
288 | - max_outstanding could cause poll to enter a busy-waiting loop. |
|
|
289 | - document the new IO::AIO::poll_cb result value. |
|
|
290 | |
|
|
291 | 3.02 Mon May 12 02:32:02 CEST 2008 |
|
|
292 | - fix a memory leak on aio_readlink. |
|
|
293 | - bring back working fchmod. |
|
|
294 | - nop and busy now set result to 0. |
|
|
295 | - set errno to ENOMEM when allocation fails. |
|
|
296 | |
|
|
297 | 3.01 Sun May 11 03:07:03 CEST 2008 |
|
|
298 | - (libeio) make it compile on systems without readahead or readdir_r. |
|
|
299 | - (libeio) improve configure check for readahead availability. |
|
|
300 | - do not try to link against -lrt. |
|
|
301 | - use a separate configure script for IO::AIO (not the libeio one). |
|
|
302 | |
|
|
303 | 3.0 Sun May 11 00:57:14 CEST 2008 |
|
|
304 | - added bin/treescan to the distribution. |
|
|
305 | - switched to using libeio. |
|
|
306 | - LOTS OF INCOMPATIBLE CHANGES: |
|
|
307 | - remove signal functionality, it is not worth the effort. |
|
|
308 | - max_outstanding no longer returns the previous number of requests. |
|
|
309 | - poll_cb no longer returns number of requests. |
|
|
310 | |
|
|
311 | 2.62 Sat Apr 26 13:59:33 CEST 2008 |
|
|
312 | - port to solaris perls configured for old posix |
|
|
313 | (analysed by Jost Krieger). |
|
|
314 | - keep a reference to the perl filehandle object in aio_close, |
|
|
315 | so it doesn't get closed prematurely, leading to ugly races. |
|
|
316 | |
|
|
317 | 2.61 Wed Apr 16 18:45:02 CEST 2008 |
|
|
318 | - fix treescan output duplication and improve output ordering. |
|
|
319 | also display files given on the commandline. |
|
|
320 | - use a different algorithm for aio_close that is faster |
|
|
321 | and probably has even lower chances of blocking. |
|
|
322 | - do our own stack memory management for threads - linux |
|
|
323 | allocates outrageous amounts of VM (not so bad unless you |
|
|
324 | use mlockall...), which severely limits the number of threads |
|
|
325 | on 32-bit arches: stack size is the larger of PTHREAD_STACK_MIN |
|
|
326 | and 4096 * sizeof (long) (usually this is 16..64k). |
|
|
327 | |
|
|
328 | 2.6 Sun Mar 30 08:28:11 CEST 2008 |
|
|
329 | - added aio_sync. |
|
|
330 | - added aio_pathsync. |
|
|
331 | - fix prototypes of (void) functions. |
|
|
332 | |
|
|
333 | 2.51 Sat Oct 6 16:04:54 CEST 2007 |
|
|
334 | - perlio isn't generally threadsafe, so aio_close could not work reliably. |
|
|
335 | aio_close now tries a safe hack that might not be that asynchronous |
|
|
336 | (see the manpage for details). |
|
|
337 | - discard callback return values as to not grow the stack endlessly |
|
|
338 | when poll_cb handles many requests without returning. |
|
|
339 | - minor code cleanups. |
|
|
340 | |
|
|
341 | 2.5 Thu Oct 4 14:49:08 CEST 2007 |
|
|
342 | - replaced _fd2fh with faster xs code. |
|
|
343 | - aio_close will now try to do "the right thing" and thus might |
|
|
344 | work sensibly for the very first time. |
|
|
345 | |
|
|
346 | 2.41 Mon Sep 24 21:28:21 CEST 2007 |
|
|
347 | - after fork, invest some work to make sure that the poll_fileno |
|
|
348 | stays stable (by dup2'ing the new fd over the old one), to make |
|
|
349 | it easier for programs/libs that don't control their forking |
|
|
350 | behaviour and cannot recreate their watchers. |
|
|
351 | |
|
|
352 | 2.4 Sun Aug 5 18:44:22 CEST 2007 |
|
|
353 | - add aio_truncate, aio_chmod, aio_chown, aio_utime (linux |
|
|
354 | successfully demonstrated that you can block on futimes...). |
|
|
355 | - allow undef as fileoffset for aio_read/write and use read/write(2) |
|
|
356 | internally (useful for sockets or O_APPEND handles). |
|
|
357 | - allow undef for length in aio_write. |
|
|
358 | - negative dataoffsets work as expected now in aio_read/aio_write. |
|
|
359 | - use NV instead of UV for 32 bit perls and file offsets, as NVs |
|
|
360 | have a larger range then. |
|
|
361 | - shared code between BDB and IO::AIO. |
|
|
362 | - aio_busy was completely broken. now fixed. |
|
|
363 | - readahead emulation now returns something. |
|
|
364 | - correctly set errno to ENOSYS on unimplemented functions |
|
|
365 | (should never happen, though). |
|
|
366 | - large changes to make it partially compile and run on win32, |
|
|
367 | but win32 lacks too much functionality, and perl overrides way |
|
|
368 | too many functions with crashing versions. |
| 8 | |
369 | |
| 9 | 2.33 Tue Jan 23 23:55:41 CET 2007 |
370 | 2.33 Tue Jan 23 23:55:41 CET 2007 |
| 10 | - fix install path (Andreas J. Koenig). |
371 | - fix install path (Andreas J. Koenig). |
| 11 | |
372 | |
| 12 | 2.32 Mon Jan 22 16:56:23 CET 2007 |
373 | 2.32 Mon Jan 22 16:56:23 CET 2007 |
