| … | |
… | |
| 4 | TODO: aio_cptree/mvtree |
4 | TODO: aio_cptree/mvtree |
| 5 | TODO: reduce condvar fairness: schedule hot-cache-threads first? |
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/) |
6 | TODO: splice/tee/vmsplice? (http://kerneltrap.org/node/6505 http://lwn.net/Articles/178199/) |
| 7 | TODO: aio_fcntl, at least for file-locking |
7 | TODO: aio_fcntl, at least for file-locking |
| 8 | |
8 | |
| 9 | # when apssing a ref as data buffer with dreaded perl 5.10 |
9 | - (libeio) replaced quicksort+insertion sort by a tuned radix |
| 10 | perl: AIO.xs:662: XS_IO__AIO_aio_read: Assertion `!((data)->sv_flags & 0x00000800) || !((data)->sv_u.svu_rv)' failed. |
10 | sort + insertion sort, resulting in comparable runtime (usually |
|
|
11 | faster) to the old version, but without any ill side effects on |
|
|
12 | degenerated (for quicksort) data. |
|
|
13 | - (libeio) correctly sort dirs by inodes if we have perfect type |
|
|
14 | knowledge and DIRS_FIRST is used. |
|
|
15 | - (libeio) internally the wrong DT_* constants were used, but no |
|
|
16 | known system has differing EIO_DT_*/DT_* values, so not an issue. |
|
|
17 | - removed a (harmless) assert that was left over on the code |
|
|
18 | but should not have been. |
|
|
19 | |
|
|
20 | 3.23 Sat Jun 13 16:57:58 CEST 2009 |
|
|
21 | - fix off-by-one bug in aio_readdir that was introduced in the |
|
|
22 | rushed 3.22 release. |
|
|
23 | |
|
|
24 | 3.22 Sat Jun 13 15:32:40 CEST 2009 |
|
|
25 | - speed up readdirx sort algorithm slightly. |
|
|
26 | - bin/treescan was missing from distro tarball. |
|
|
27 | |
|
|
28 | 3.21 Fri Jun 12 18:45:53 CEST 2009 |
|
|
29 | - new options --dirs and --files for treescan. |
|
|
30 | - install bin/treescan by default. |
|
|
31 | - (libeio) aio_readdir can now be cancelled while executing. |
|
|
32 | - fix a printf format string for 64 bit systems (could lead |
|
|
33 | to problems on big endian 64 bit systems). |
|
|
34 | - do not use qsort() but our own algorithm: glibc initialises |
|
|
35 | the whole locale and I/O subsystem inside qsort, causing |
|
|
36 | a stack overflow on 32 bit machines. The new sort uses much less |
|
|
37 | stack and is more than twice as fast in typical situations. |
|
|
38 | |
|
|
39 | 3.2 Sun Jun 7 20:30:05 CEST 2009 |
|
|
40 | - (libeio) pwrite emulation was even more flawed than intended and did |
|
|
41 | not restore the file offset. |
|
|
42 | - add aio_readdirx, which can return inode and filetype and sort |
|
|
43 | the names in various ways. |
|
|
44 | - unfortunately, utime, chmod, chown on an open file that has just |
|
|
45 | been written can easily block, which caused aio_copy to block the |
|
|
46 | process. no more! |
|
|
47 | - no longer rely on dst path in aio_copy when futime is available. |
|
|
48 | |
|
|
49 | 3.19 Tue Apr 21 22:05:21 CEST 2009 |
|
|
50 | - more perl 5.10 workarounds for aio_read and write. |
|
|
51 | - aio_write no longer modifies the sv (if possible). |
|
|
52 | - aio_read now works correctly with magic values. |
|
|
53 | |
|
|
54 | 3.18 Sun Apr 19 21:17:32 CEST 2009 |
|
|
55 | - better diagnostics when some aio_* functions get passed an illegal fh. |
|
|
56 | - try to avoid crashes due to incompatible 5.10 API changes. grmbl. |
| 11 | |
57 | |
| 12 | 3.17 Thu Nov 20 08:45:36 CET 2008 |
58 | 3.17 Thu Nov 20 08:45:36 CET 2008 |
| 13 | - (libeio) added aio_sync_file_range (untested). |
59 | - (libeio) added aio_sync_file_range (untested). |
| 14 | - add aio_busy to @IO::AIO::AIO_REQ. |
60 | - add aio_busy to @IO::AIO::AIO_REQ. |
| 15 | |
61 | |
