Many hyperlinks are disabled.
Use anonymous login
to enable hyperlinks.
50 descendants and 50 ancestors of 9803196dec85e3aa
|
2010-05-06
| ||
| 18:48 | Instead of transitioning to RECOVER state from CHECKPOINT when a recovery is required, perform the recovery while holding the CHECKPOINT lock. (check-in: bb0b6021e7 user: dan tags: trunk) | |
| 18:27 | If recovery is run before a checkpoint, change back to a CHECKPOINT lock before performing the actual checkpoint. (check-in: dc98ee169c user: dan tags: trunk) | |
| 18:04 | Only pass -1 as the size parameter to xShmGet when the caller does not care what size mapping is returned. Only call xShmSize when the underlying allocation (not the mapping) should be resized. (check-in: 637f51ddf9 user: dan tags: trunk) | |
| 17:28 | Fix a bug whereby an old snapshot could be checkpointed (and subsequent transactions discarded) if the last connection to disconnect from a WAL database happended to be holding an out-of-date wal-index header. (check-in: d0c0034b0b user: dan tags: trunk) | |
| 16:06 | After throwing an error to say that one cannot change into WAL mode within a tranactions, do not then go and change into WAL mode. (check-in: 56a17dae91 user: drh tags: trunk) | |
| 15:56 | Further coverage tests. (check-in: 98a9713e00 user: dan tags: trunk) | |
| 15:36 | Update configure scripts for WAL support. (check-in: 2edc5129f2 user: shaneh tags: trunk) | |
| 14:42 | Add tests to cover a couple of branches in wal.c. (check-in: 429f437fb7 user: dan tags: trunk) | |
| 13:36 | Unset a tcl variable before reusing it in wal.test. (check-in: e83efb232f user: dan tags: trunk) | |
| 12:15 | Add tests for some lock-contention cases. (check-in: f0fcb9c9a6 user: dan tags: trunk) | |
| 11:56 | Remove the noop-mutex implementations of mutex_held() and mutex_notheld() since they are both unreachable. (check-in: 6767b62a9a user: drh tags: trunk) | |
| 11:55 | Add two text files containing pager design notes to the doc/ subfolder. (check-in: ed817fc893 user: drh tags: trunk) | |
| 11:32 | Add test cases to test the libraries handling of corrupt wal-index headers. (check-in: 9465b267d4 user: dan tags: trunk) | |
| 07:43 | Add tests for handling errors returned by xShm VFS methods. (check-in: fbbcacb137 user: dan tags: trunk) | |
|
2010-05-05
| ||
| 20:00 | Change the behavior of the sqlite3_wal_hook() callback. It should now return SQLITE_OK or an error code and the error code is propagated back up the stack. If a checkpoint is desired, the callback should invoke sqlite3_wal_callback() itself. (check-in: 1b14195e05 user: drh tags: trunk) | |
| 19:09 | Further simplifications to the SHM locking logic. Remove the SQLITE_SHM_QUERY option. Unify the WRITE to READ and RECOVER to READ transitions. (check-in: e9bfa6e4ce user: drh tags: trunk) | |
| 19:04 | Test the handling of errors returned by the xShmXXX() APIs. (check-in: 72663123d6 user: dan tags: trunk) | |
| 18:46 | Do not compare page sizes on source and destination of backup until transactions are started and the page sizes are locked. This is a fix to check-in [7bd44794c4]. (check-in: ec7157788b user: drh tags: trunk) | |
| 18:20 | Simplifications to the SHM locking implementation in os_unix.c. (check-in: 9de05bfb09 user: drh tags: trunk) | |
| 16:23 | Prohibit backup if the destination is using WAL and has a different page size from the source. (check-in: 7bd44794c4 user: drh tags: trunk) | |
| 15:33 | When closing a WAL database, if the exclusive lock on the database file is obtained and the database successfully checkpointed, delete the wal-index file from the file system. (check-in: 2ac5d96c8e user: dan tags: trunk) | |
| 04:11 | Make sure the page size of the main database is fixed following a failed VACUUM attempt. Otherwise, two consecutive failed VACUUM attempts with attempted page_size changes on encrypted or WAL databases could result in database corruption. (check-in: 208e7d5d3a user: drh tags: trunk) | |
| 03:39 | Do not attempt to modify the page_size when running VACUUM on a WAL database. (check-in: 58a1a59be8 user: drh tags: trunk) | |
| 00:22 | Get the previous mutex fix working with SQLITE_DEBUG and with the amalgamation. (check-in: df19928f7e user: drh tags: trunk) | |
| 00:05 | When the in single-threaded mode, the sqlite3_mutex_alloc() interface still returns a non-NULL value. The mutex doesn't do anything, but it tests non-NULL. This way, extensions (or VFSes) that use sqlite3_mutex_alloc() can tell the difference between an OOM error and mutexes being disabled. (check-in: 451fd17575 user: drh tags: trunk) | |
|
2010-05-04
| ||
| 18:50 | When sqlite3PagerPagecount() is called without any locks, always return the physical file size, not the logical file size. (check-in: 4016b42228 user: drh tags: trunk) | |
| 17:38 | Be sure to release all wal-index locks before closing the wal-index. (check-in: b15b67fa14 user: drh tags: trunk) | |
| 17:20 | Relax an over-zealous assert() in sqlite3WalUndo(). (check-in: 8f9d22d58c user: drh tags: trunk) | |
| 16:33 | Take care not to invoke the xShmClose method of the VFS with a NULL shared memory object. (check-in: 1465d3842c user: drh tags: trunk) | |
| 15:20 | Add a test case to verify that log files containing pages that are not a power-of-two bytes in size are handled correctly. (check-in: c2bf693f93 user: dan tags: trunk) | |
| 14:47 | Fix problems with recovering wal files that use a page-size other than the default. (check-in: 1a391f3c55 user: dan tags: trunk) | |
| 11:06 | Fix a typo in walfault.test. (check-in: 232dbe8ece user: dan tags: trunk) | |
| 10:36 | Test that the correct number of padding frames are appended to the log file after committing a transaction in synchronous=FULL mode. (check-in: a60104aa7e user: dan tags: trunk) | |
|
2010-05-03
| ||
| 19:20 | Unset some global TCL variables prior to use in test where prior tests can have those same variables set to an array value. (check-in: 49bef00e5c user: drh tags: trunk) | |
| 19:17 | Fix a bug in the computation of the current time for the alternative CURRENT_TIMESTAMP functions used when SQLITE_OMIT_DATETIME is defined. (check-in: e7ded46b5e user: drh tags: trunk) | |
| 18:51 | Documentation stubs for sqlite3_wal_autocheckpoint and sqlite3_wal_checkpoint. (check-in: 313762427d user: drh tags: trunk) | |
| 18:22 | Fix a bug in the xCurrentTime() method of the os_unix.c VFS. (check-in: 551df11a56 user: drh tags: trunk) | |
| 18:01 | Add a test to simulate an OOM during log recovery to walfault.test. (check-in: 7ed85e87dd user: dan tags: trunk) | |
| 17:24 | Clear global variable "seconds" before use in walthread.test. (check-in: a35551a530 user: drh tags: trunk) | |
| 17:18 | Fix a couple of errors in WAL code that can occur in an OOM situation. (check-in: 9d3f3736be user: dan tags: trunk) | |
| 16:36 | Update the crash-test VFS in test6.c to pass-through the shared-memory methods to the real underlying VFS. This fixes the walcrash.test script. (check-in: ea09ff3791 user: drh tags: trunk) | |
| 16:30 | Change the VFS definition so that all methods take a VFS object pointer as their first parameter. (check-in: 43b5b07f0d user: drh tags: trunk) | |
| 15:58 | Have sqlite3_wal_checkpoint() handle a zero-length string in the same way as a NULL pointer. Fix "PRAGMA wal_checkpoint" so that it checkpoints all attached databases. (check-in: 7fecd21f45 user: dan tags: trunk) | |
| 14:32 | Use VFS method xCurrentTimeInt64 instead of xCurrentTime when it is available. Provide an implementation of xCurrentTimeInt64 for os_unix.c. (check-in: ab77b3ae6d user: drh tags: trunk) | |
| 14:08 | Merge the write-ahead-logging changes into the trunk. (check-in: de9ae443cc user: drh tags: trunk) | |
| 14:05 | If the sqlite3_wal_checkpoint() API is passed a NULL pointer in place of a database name, attempt to checkpoint all attached databases. (Closed-Leaf check-in: 27a5c09ce8 user: dan tags: wal) | |
| 13:37 | Make sure the mutex is held while calling sqlite3ApiExit() in sqlite3_wal_checkpoint(). Other cleanup of WAL logic. (check-in: 11a85b821a user: drh tags: wal) | |
| 12:14 | Have sqlite3_wal_checkpoint() populate the database handle error message and error code (as returned by sqlite3_errmsg() and sqlite3_errcode()). (check-in: ff234cf574 user: dan tags: wal) | |
| 11:05 | Add the "PRAGMA wal_autocheckpoint" command. Rename "PRAGMA checkpoint" to "PRAGMA wal_checkpoint". (check-in: 714e594726 user: dan tags: wal) | |
| 08:19 | Merge two wal leaves. (check-in: 23c0e6c3f3 user: dan tags: wal) | |
| 08:04 | Add the sqlite3_wal_checkpoint() and sqlite3_wal_autocheckpoint() APIs. (check-in: 9803196dec user: dan tags: wal) | |
|
2010-05-01
| ||
| 17:57 | Define an invariant to guarantee deadlock-free operation of SHM in os_unix.c and check that invariant with assert() statements. (check-in: 6af2dca75b user: drh tags: wal) | |
| 17:50 | Rework mutexes on the SHM implemention for os_unix to avoid a deadlock during WAL recovery. (check-in: 1a0f69bef2 user: drh tags: wal) | |
| 16:40 | Support compile-time option SQLITE_OMIT_WAL, for building without WAL support. (check-in: 9b230c43df user: dan tags: wal) | |
| 11:33 | Fix a typo in the walmode.test test script introducted yesterday. (check-in: dcabc90ca5 user: drh tags: wal) | |
| 11:19 | Fix the sqlite3_mutex_alloc() interface to return NULL (not segfault) when operating in threadsafe mode. (This is a general-purpose bug fix which really ought to be ported to trunk.) (check-in: 64840a3caf user: drh tags: wal) | |
| 08:30 | Add test case demonstrating deadlock during recovery of very large log files. No fix yet. (check-in: 63ea318eb1 user: dan tags: wal) | |
| 00:59 | Do not allow journal_mode=WAL if the underlying VFS does not support xShmOpen. (check-in: d1fcccecdc user: drh tags: wal) | |
|
2010-04-30
| ||
| 22:28 | Merge in changes from the trunk. (check-in: 76bf0eee1f user: drh tags: wal) | |
| 21:03 | Avoid assertion faults in queries using indices with redundant columns. Ticket [3dbdcdb14e7f41]. (check-in: 946e06cd2a user: drh tags: trunk) | |
| 17:47 | Change the SHM VFS logic in os_unix.c so that it does not hold an exclusive lock n the mapped memory when also holding a CHECKPOINT lock. This improves concurrency between readers and checkpointers. (check-in: 8660cda6f8 user: drh tags: wal) | |
| 17:28 | Remove some obsolete debugging parameters. (check-in: a012aed498 user: drh tags: wal) | |
| 17:05 | Fix a problem with resizing the wal-index mapping after the mapping has been extended by an external process. (check-in: 79d356fea6 user: dan tags: wal) | |
| 16:50 | Fix a problem with calculating the required size of the wal-index when appending frames. (check-in: 9526b11130 user: dan tags: wal) | |
| 16:48 | In the debugging output for SHM-LOCK in os_unix.c, use symbolic names for the lock states rather than raw numbers. (check-in: 2afc33de2b user: drh tags: wal) | |
| 16:41 | Fix a problem in the previous commit. Both unixShm.lockState and unixShm.readLock should be set to READ when downgrading from a WRITE to a READ lock (check-in: 90119fecbc user: dan tags: wal) | |
| 16:38 | Fix a bug whereby upgrading from a READ_FULL to a WRITE lock, and then back to a read lock was dropping all WAL locks. (check-in: 55f5af5e17 user: dan tags: wal) | |
| 16:24 | Reapply commits [837d82a929] and [c05e7dca17] that were accidentally overwritten. (check-in: 598de52700 user: dan tags: wal) | |
| 16:19 | Add xShmXXX() methods to the test VFS in test_devsym.test. (check-in: 1d20342424 user: dan tags: wal) | |
| 16:12 | Correctly record the fact that the SHM lock reached PENDING if it did so but failed to reach CHECKPOINT. (check-in: d9250e84ac user: drh tags: wal) | |
| 15:54 | Fix a couple uninitialized variables in the xShmLock method of the unix VFS. Improved debugging logic for xShmLock. (check-in: 69567c5fca user: drh tags: wal) | |
| 15:49 | When closing a WAL connection, attempt an exclusive lock on the database file. If the lock is obtained, checkpoint the database and delete the wal and wal-index files. (check-in: c05e7dca17 user: dan tags: wal) | |
| 15:24 | If a reader attempts to upgrade to a writer, but is not reading the most recent database snapshot, return SQLITE_BUSY. (check-in: 837d82a929 user: dan tags: wal) | |
| 14:39 | Separate the concepts of underlying storage size and mapped size in the VFS shared-memory implementation. (check-in: 4cbe49f13f user: drh tags: wal) | |
| 11:43 | Add a missing walIndexUnmap() call to the checkpoint code. Change a couple of SQLITE_CANTOPEN constants to SQLITE_CANTOPEN_BKPT. (check-in: 1f9e8c5c27 user: dan tags: wal) | |
| 10:24 | Fix an assert() in sqlite3WalRead(). (check-in: 2e7a0050e1 user: dan tags: wal) | |
| 10:06 | Add missing mutexes to unixShmClose(). (check-in: a4741cb54d user: dan tags: wal) | |
| 09:52 | Fix a couple of assert() statements in os_unix.c and wal.c. Combine sqlite3WalIsDirty() with sqlite3WalUndo(). (check-in: a8f958be80 user: dan tags: wal) | |
| 09:32 | Add a missing walIndexUnmap() to sqlite3WalSnapshotOpen(). (check-in: 72b95fde15 user: dan tags: wal) | |
| 06:02 | Merge latest bugfix into wal branch. (check-in: 265e66e698 user: dan tags: wal) | |
| 05:57 | Zero the "immediate FK constraint counter" associated with a statement object when sqlite3_reset() is called. Fix for [c39ff61c43]. (check-in: f660be615a user: dan tags: trunk) | |
| 02:13 | The first 6 WAL tests now work. It's a start. (check-in: a92c1851da user: drh tags: wal) | |
|
2010-04-29
| ||
| 22:34 | Refactor wal.c to use the VFS. This check-in compiles and links and works ok as long as you leave WAL turned off, but WAL does not work. (check-in: 62db5fa3b6 user: drh tags: wal) | |
| 16:40 | Untested implementation of the shared-memory dead-man-switch. (check-in: 706611283e user: drh tags: wal) | |
| 15:17 | Progress towards a VFS that will support WAL. Locking code is in place but is untested. Still no support for the DMS. (check-in: 1bde41cf08 user: drh tags: wal) | |
| 14:58 | Close all open database connections at the end of wal.test. (check-in: 3cc55a7568 user: dan tags: wal) | |
| 14:51 | Fix a bug in the WAL checkpoint code causing SQLite to use an inconsistent cache in a subsequent transaction. (check-in: d1cadeed4e user: dan tags: wal) | |
| 08:47 | Add tests to walthread.test. (check-in: 9e891e7543 user: dan tags: wal) | |
|
2010-04-28
| ||
| 18:17 | Add a test to walthread.test for changing between WAL and rollback modes. (check-in: da229e44bd user: dan tags: wal) | |
| 17:49 | Merge two "wal" leaves. (check-in: 13d2d5a66e user: dan tags: wal) | |
| 17:48 | Change walthread.test so that tests can be run with either multiple threads or multiple processes. (check-in: 25f85f6872 user: dan tags: wal) | |
| 17:21 | Changes to the interface design for the xShmLock method of the VFS. (check-in: 348409de26 user: drh tags: wal) | |
| 14:42 | Rename the sqlite3_log_hook() to sqlite3_wal_hook(). Added comments to wal.h. (check-in: bbc385111b user: drh tags: wal) | |
|
2010-04-27
| ||
| 18:49 | Merge two "wal" leaves. (check-in: 8c2d43babd user: dan tags: wal) | |
| 18:43 | Add a test to check that a checkpointer releases its locks before returning, even if it has been prevented by a conflicting lock from checkpointing the database. (check-in: be44349d2b user: dan tags: wal) | |
| 14:37 | Fix a problem with identifying a corrupt database in WAL mode. (check-in: c8658da446 user: dan tags: wal) | |
| 11:49 | Add the xShmRelease() method to the VFS. The os_unix.c implementation of the shared memory logic is still non-functional. (check-in: ed715b47c5 user: drh tags: wal) | |
| 06:49 | Run trans.test and avtrans.test as part of the "wal" permutation. (check-in: c889331051 user: dan tags: wal) | |
| 05:42 | Fix a problem that occurs when one process causes the log-summary file to grow and then a second process attempts to read the database. (check-in: b51a5f8bc6 user: dan tags: wal) | |
| 01:56 | Merge in recent changes from the trunk (check-in: 7a0ac682c3 user: drh tags: wal) | |
|
2010-04-26
| ||
| 19:17 | Optimization: Convert an ORDER BY clause into a no-op if the query also contains a GROUP BY clause that will force the same output order. (check-in: ca9d86baf7 user: drh tags: trunk) | |