SQLite

Timeline
Login

Many hyperlinks are disabled.
Use anonymous login to enable hyperlinks.

74 check-ins tagged with "experimental-mmap"

2013-04-03
21:23
Release resources prior to bailing out of the btreeCreateTable() routine following an OOM error. (Closed-Leaf check-in: b2a72be9 user: drh tags: experimental-mmap)
20:04
Remove an unreachable branch in the direct blob I/O logic of btree.c. (check-in: f97d7274 user: drh tags: experimental-mmap)
13:38
Fix a bug in the detection of iOS when setting the default mmap_limit. (check-in: ac1432b3 user: drh tags: experimental-mmap)
13:09
Change the mmap_limit pragma to report the new limit, or to report the existing limit if called with no arguments. Report the default mmap_limit as part of PRAGMA compile_options. Set the default mmmap_limit to 0 for all systems other than linux, mac, windows, and solaris. (check-in: 2d9f1327 user: drh tags: experimental-mmap)
11:52
Remove the restriction on using xFetch to load the root pages of tables and indexes. It appears to have been based on a misconception. (check-in: 5b082efe user: dan tags: experimental-mmap)
11:38
When moving a page to make way for the root page of a new table or index in an auto-vacuum database, save the positions of any cursors that may be holding xFetch references to the page being moved. (check-in: 9d9b1da5 user: dan tags: experimental-mmap)
11:17
In btree.c, save the positions of other cursors open on the same table when writing via an incremental blob handle. Otherwise, they may be left holding an out-of-date xFetch page reference. (check-in: 3f09fba1 user: dan tags: experimental-mmap)
10:50
Initialize the mmap_limit of temporary files to the configured mmap_limit. (check-in: 24bab759 user: drh tags: experimental-mmap)
03:53
Make use of xFetch, if it is available when doing a VACUUM or a backup. (check-in: 83bc37af user: drh tags: experimental-mmap)
01:26
Add the -mmap option to the command-line shell, for setting the default mmap_limit. (check-in: fc30d06c user: drh tags: experimental-mmap)
00:42
Fix the unix driver to check defined(_BSD_SOURCE) rather than just the plain _BSD_SOURCE macro. This fixes the build for OpenBSD. (check-in: 1dd42ef4 user: drh tags: experimental-mmap)
2013-04-02
20:55
Fix a resource leak in os_unix.c. (check-in: b29cda03 user: dan tags: experimental-mmap)
20:29
Adjustments to mmap1.test so that it works better on windows. It still gets some answers wrong, but it no longer crashes. (check-in: ee8d3cee user: drh tags: experimental-mmap)
20:19
Fix a bug in mmap2.test that was leaving an invalid sqlite3_log() callback installed, causing a crash. (check-in: c0cdaa07 user: dan tags: experimental-mmap)
20:13
Add assert to the Windows VFS pertaining to the size of the SIZE_T type. (check-in: 4eeea583 user: mistachkin tags: experimental-mmap)
18:33
Add test cases to mmap1.test and pagerfault.test. (check-in: 3050136b user: dan tags: experimental-mmap)
18:31
Change an assert in pager.c to acknowledge that it is possible for sqlite3PagerRollback() to return SQLITE_CORRUPT. (check-in: d641d3d2 user: dan tags: experimental-mmap)
18:25
Fix the build for Mac. (check-in: d36ce0bb user: drh tags: experimental-mmap)
15:37
Add new test file btreefault.test. (check-in: 7fe908af user: dan tags: experimental-mmap)
14:37
Fix a faulty assert() in the os_win.c VFS. (check-in: fd6ee549 user: drh tags: experimental-mmap)
12:04
Add test cases for errors in mmap() or mremap() is os_unix.c. (check-in: 3098a3c1 user: dan tags: experimental-mmap)
10:29
Proposed template preprocessor magic for activating mmap only on platforms where we know it works. (check-in: d96272f0 user: drh tags: experimental-mmap)
01:45
Reorder two conditions to make coverage testing easier. (check-in: 793ba4e9 user: drh tags: experimental-mmap)
00:41
Remove an unreachable condition from pager.c. (check-in: 3628e86b user: drh tags: experimental-mmap)
00:15
Always send the SQLITE_FCNTL_MMAP_LIMIT pragma to the VFS, even if the limit is zero and even if the VFS does not support xFetch(). (check-in: 01ffdabb user: drh tags: experimental-mmap)
2013-04-01
22:42
Remove a debugging printf() accidently left in the previous check-in. (check-in: 8198cdd8 user: drh tags: experimental-mmap)
22:38
Add the SQLITE_CONFIG_MMAP_LIMIT configuration option for overriding the SQLITE_DEFAULT_MMAP_LIMIT compile-time setting. Enhance "PRAGMA mmap_limit" so that without a specific database name, it sets the limit on all database files and changes the default for any future databases that might be added using ATTACH. (check-in: 78141d0a user: drh tags: experimental-mmap)
18:25
Add a comment to xUnfetch on windows to say that the call to winUnmapfile() might be unnecessary. (check-in: e3c63c31 user: drh tags: experimental-mmap)
18:15
In windows, round down the size of the memory mapped region to the next smaller multiple of the page size. (check-in: 0e44ed1b user: drh tags: experimental-mmap)
17:58
Merge accidental fork. (check-in: 5f4437c0 user: dan tags: experimental-mmap)
17:56
Attempt to emulate mremap() on non-Linux systems by allocating a second mapping immediately following the first in virtual memory. (check-in: 4d67433d user: dan tags: experimental-mmap)
17:56
Bug fix in the winMapfile() subroutine: Be sure to record the map object handle in the sqlite3_file object. (check-in: ee4d188e user: drh tags: experimental-mmap)
17:22
Add xFetch and xUnfetch methods to the os_win.c VFS. (check-in: a1653a25 user: drh tags: experimental-mmap)
16:56
Ensure that a checkpoint does not use an out-of-date mapping. (check-in: a1040f03 user: dan tags: experimental-mmap)
14:35
Fix a typo in a log message in wal.c. (check-in: 83fc48d1 user: dan tags: experimental-mmap)
14:29
Fix a problem in btree.c that could cause a crash following an OOM. (check-in: b724aa58 user: dan tags: experimental-mmap)
2013-03-26
18:48
Fix a comment in os_unix.c. No code changes. (check-in: 72813b8e user: drh tags: experimental-mmap)
14:36
Change the name of the Pager.pFree field to Pager.pMmapFreelist. (check-in: 611bd824 user: drh tags: experimental-mmap)
14:16
In btree.c, save the positions of any open cursors before moving any pages around to auto-vacuum the database on commit. (check-in: 30c0a693 user: dan tags: experimental-mmap)
01:07
Previous check-in accidently left mmap turned off by default. This checkin fixes that. Unfortunately, shared.test is now segfaulting. All other veryquick tests appear to work, however. (check-in: a850c731 user: drh tags: experimental-mmap)
2013-03-25
23:09
Memory-mapped I/O is now on by default. The "PRAGMA mmap_limit(N)" can be used to issue a hint to the VFS to limit mmap space to N bytes. The VFS is free to ignore that hint if desired. However, if "PRAGMA mmap_limit(0)" is used, xFetch is never called. (check-in: 1b37c4ef user: drh tags: experimental-mmap)
20:50
Add munmap and mremap to the set of os interfaces that can be overloaded in os_unix.c. (check-in: 8776047b user: drh tags: experimental-mmap)
20:30
Use mremap() on Linux. (check-in: 431aecc8 user: dan tags: experimental-mmap)
19:57
Merge all recent trunk changes into the experimental-mmap branch. (check-in: a607d63f user: drh tags: experimental-mmap)
19:05
Add a test that simulates an error in mmap(). (check-in: 6ec7367d user: dan tags: experimental-mmap)
18:25
Simulate OOM errors in the sqlite3OsFetch() function. Run malloc.test as part of the "mmap" permutation. (check-in: 77443ef2 user: dan tags: experimental-mmap)
17:00
Change the signature of the xUnfetch method to "int (*xUnfetch)(sqlite3_file*, sqlite3_int64 iOfst, void *p)". (check-in: 115b8305 user: dan tags: experimental-mmap)
16:28
Remove unnecessary code to round the size of a memory mapping to 4KB from os_unix.c. Rename SQLITE_IOERR_MREMAP to SQLITE_IOERR_MMAP. Fix other small issues in os_unix.c. (check-in: dce35c01 user: dan tags: experimental-mmap)
14:31
Do not return SQLITE_IOERR when the user attempts to open a small file that is not a database with mmap enabled. Instead return SQLITE_NOTADB. (check-in: bbcaab3e user: dan tags: experimental-mmap)
13:50
Fix a case in the pager where an xFetch() reference was being leaked following an OOM error. (check-in: 5885ba6c user: dan tags: experimental-mmap)
2013-03-23
21:00
Replace the sqlite3_io_methods.xMremap interface with sqlite3_io_methods.xFetch and xUnfetch. (check-in: 1431be95 user: dan tags: experimental-mmap)
17:29
Improve a comment in wal.c. No code changes. (check-in: 60b9f5e4 user: dan tags: experimental-mmap)
14:20
Fix a case where a checkpoint operation could write to an invalid part of a memory mapped region. (check-in: 8dbe89d0 user: dan tags: experimental-mmap)
12:15
In the winMremap VFS function, unmap the region prior to attempting to truncate the file. (check-in: 8870c4cc user: mistachkin tags: experimental-mmap)
10:09
Do not unmap the mapped file region in winTruncate as there may be outstanding pointers to it. (check-in: f57a9c91 user: mistachkin tags: experimental-mmap)
09:56
Preliminary changes to support the xMremap VFS method on Windows. (check-in: 75a85a1c user: mistachkin tags: experimental-mmap)
2013-03-22
20:15
Update wal mode tests so that they work with the mmap test permutation. (check-in: f7295872 user: dan tags: experimental-mmap)
19:56
Fix a potential NULL-pointer dereference in btreeSwapOutMmap(). (check-in: e81ccdcd user: drh tags: experimental-mmap)
19:17
Fix a case where the mapped part of a database file may be written during a backup operation. (check-in: c8d67aef user: dan tags: experimental-mmap)
18:20
Allow the database file to be memory mapped in wal mode. (check-in: d190ddab user: dan tags: experimental-mmap)
17:46
Add a fix for the assert() statements added by the previous commit. (check-in: 19345416 user: dan tags: experimental-mmap)
08:58
Add assert statements to os_unix.c to ensure that any mapped region of the database file is not being read or written using the xRead() or xWrite() methods. (check-in: 765615f9 user: dan tags: experimental-mmap)
2013-03-21
20:39
Fix cases where xRead() was being used to read from a memory mapped part of the database file. (check-in: 5c9e9df2 user: dan tags: experimental-mmap)
20:00
Avoid calling xRead() on a part of the database file that is memory mapped. (check-in: c8eac290 user: dan tags: experimental-mmap)
15:57
Fix a problem when opening a write-transaction while there exist read-only b-tree cursors in mmap mode. (check-in: 32e0bbb7 user: dan tags: experimental-mmap)
14:47
Do not use the Linux mremap() call. Use the same strategy for xMremap() as on OSX instead. (check-in: 5ed8ad78 user: dan tags: experimental-mmap)
2013-03-20
18:25
Optimize the xMremap method in os_unix.c some. (check-in: 9529ed88 user: dan tags: experimental-mmap)
14:26
When possible, use memory mapping when appending new pages to a database file. (check-in: 14135da3 user: dan tags: experimental-mmap)
10:07
Add test file mmap1.test. (check-in: aee1f53a user: dan tags: experimental-mmap)
2013-03-19
19:28
Add the sqlite3_io_methods.xMremap() method to the VFS interface. Also "PRAGMA mmap_size". (check-in: 6183f1bd user: dan tags: experimental-mmap)
2013-03-16
20:19
When possible, use memcpy() to and from the mapped region instead of xWrite() and xRead(). (check-in: f8ca5622 user: dan tags: experimental-mmap)
2013-03-15
19:13
Fix a dropped error code in pager.c. (check-in: 022fdc98 user: dan tags: experimental-mmap)
18:29
Allow read-only cursors to use mmap pages even if there is an open write transaction. (check-in: b387e2f9 user: dan tags: experimental-mmap)
2013-03-14
18:34
Use mmap() to read from the database file in rollback mode. This branch is unix only for now. (check-in: 6f21d9cb user: dan tags: experimental-mmap)