SQLite

Timeline
Login

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

50 check-ins occurring around 8191b512122c13d7.

2013-11-26
16:48
Change tclsqlite3.c so that it never invokes ctype macros with signed character arguments. (check-in: c07caabf23 user: drh tags: trunk)
16:20
Do not try to run the atof1.test test script on ARM hardware which lacks the "long double" type. (check-in: fafca560f2 user: drh tags: trunk)
15:45
Change the REAL-to-INTEGER casting behavior so that if the REAL value is greater than 9223372036854775807.0 then it is cast to the latest possible integer, 9223372036854775807. This is sensible and the way most platforms work in hardware. The former behavior was that oversize REALs would be cast to the smallest possible integer, -9223372036854775808, which is the way Intel hardware works. (check-in: 6f53fc7106 user: drh tags: trunk)
01:13
Fix comment. (check-in: 8364af392d user: mistachkin tags: winHdr)
01:00
Further work on Windows header file reform. (check-in: 540f552516 user: mistachkin tags: winHdr)
00:33
Merge updates from trunk. (check-in: c5ab437875 user: mistachkin tags: winHdr)
00:28
Better support for UTF-8 paths on Cygwin. (check-in: 9954327c0f user: mistachkin tags: trunk)
2013-11-25
23:42
Add SQLITE_CONFIG_WIN32_HEAPSIZE option to configure the maximum isolated heap size on Windows. (check-in: 914e6c9d88 user: mistachkin tags: trunk)
22:24
Change a while-loop into a do-loop in sqlite3VdbeRecordCompare() and add an assert() to verify that the do-loop really is appropriate. This gives about a 10% performance boost in that routine, or a 0.9% boost overall. (check-in: 011a6be439 user: drh tags: trunk)
21:49
Clarify docs for the SQLITE_CONFIG_WIN32_HEAPSIZE option. (Closed-Leaf check-in: 51e876074a user: mistachkin tags: winHeapSize)
21:41
Remove a NEVER() from a branch that is now reachable during OOM faults. (check-in: 18bdcb05c2 user: drh tags: trunk)
20:50
Optimizations to the sqlite3BtreeMovetoUnpacked() routine in storage engine making it about 17.8% faster, which in turn makes SQLite over 1.2% faster overall. (check-in: 032e89934f user: drh tags: trunk)
20:14
Return an SQLITE_CORRUPT error if the content size field of a table record extends off the end of a page. (Closed-Leaf check-in: b48c4e4021 user: drh tags: btree-optimization)
17:38
Uses shifts rather than division for arithmetic on the cell indices, since those indices are always non-negative. (check-in: 5bf2a3feeb user: drh tags: btree-optimization)
16:52
Optimize the skipping of the payload size field when doing a binary search for a rowid. (check-in: 55e5bfa231 user: drh tags: btree-optimization)
16:23
Arrange for sqlite3BtreeMovetoUnpacked() to always leave the cursor cell parse cache invalidated. This fixes issues with previous check-ins on this branch and also runs faster. (check-in: a74675607f user: drh tags: btree-optimization)
15:01
More optimizations to sqlite3BtreeMovetoUnpacked(). But there are failures in TH3. Committing this intermediate state to facilitate bisecting. (check-in: f80497be44 user: drh tags: btree-optimization)
14:10
More improvements to sqlite3BtreeMovetoUnpacked() performance. (check-in: 8868069823 user: drh tags: btree-optimization)
09:37
Correct line-endings in the new header file. (check-in: 94219b9f1f user: mistachkin tags: winHdr)
09:36
Initial work on isolating usage of the Windows header file. (check-in: 0d42c6b830 user: mistachkin tags: winHdr)
02:38
Performance improvements in sqlite3BtreeMovetoUnpacked(). (check-in: d0fb7acea7 user: drh tags: btree-optimization)
2013-11-24
23:18
Better support for UTF-8 paths on Cygwin. (Closed-Leaf check-in: 484162b6e5 user: mistachkin tags: cygUtf8)
01:14
Add the --scratch parameter to speedtest1. Improved error messages when misconfiguring memory parameters in speedtest1. (check-in: 8f3c767a30 user: drh tags: trunk)
00:46
The MEMSYS5 algorithm does not have to return the block with the lowest address. Any block of the appropriate size will do. Use the first block found on the freelist for the appropriate size for a performance improvement. (check-in: 12e612e8e7 user: drh tags: trunk)
2013-11-23
22:45
A much simpler fix is to simply change MEMSYS5 so that it takes any free block of the appropriate size (the first on the list of free blocks) rather than searching for the one with the smallest address. This is also faster than using the min-heap algorithm. Need to research to verify that the allocator still satisfies the Robson proof, however. (Closed-Leaf check-in: 8191b51212 user: drh tags: memsys5-performance)
21:30
Use a heap-based primary queue rather than a linked list to store the available free blocks of each size in MEMSYS5, since this provides faster access to the first available block. (Closed-Leaf check-in: 7d2cdfad0e user: drh tags: memsys5-performance)
21:29
Add newlines at the end of some error messages in speedtest1. (check-in: 6b98f0af7a user: drh tags: trunk)
11:45
Report errors from sqlite3_exec() and sqlite3_config() in speedtest1. Fix a bug in the main testing logic that was found by these error reports. (check-in: 659f1a98ae user: drh tags: trunk)
04:32
Fix the order of parameters to SQLITE_CONFIG_PAGECACHE in the speedtest1.exe program. (check-in: dbe85ef6d2 user: drh tags: trunk)
04:22
Adjust MSVC makefile to avoid using forward slashes in file names. (check-in: facf6deaa6 user: mistachkin tags: trunk)
04:16
Add the "speedtest1.exe" test program. (check-in: fb6f0c6d77 user: drh tags: trunk)
00:27
Add experimental sqlite3_config option to control the native Win32 heap size. (check-in: f09f11e94b user: mistachkin tags: winHeapSize)
2013-11-22
21:32
Fix harmless compiler warning. (check-in: f336c18fb7 user: mistachkin tags: trunk)
00:49
Minor correction to the batch build tool for MSVC. (check-in: 22144c9df2 user: mistachkin tags: trunk)
2013-11-21
23:37
Fix a harmless clang warning in the command-line shell. (check-in: 3d47a556f0 user: drh tags: trunk)
22:02
Slight change to the --timer output on wordcount for better display on windows: Avoid showing the full pathname of the executable. (check-in: b9e047b9e3 user: drh tags: trunk)
21:59
Changes some offset and amount parameters from "int" to "u32" to avoid harmless signed/unsigned comparison warnings. (check-in: 4e8c5d0795 user: drh tags: trunk)
21:40
Modify wordcount so that timer information appears on standard error instead of standard output. Rename the run-wordcount.bash script to run-wordcount.sh and simplify it so that it stands a better chance of running on non-GNU systems. (check-in: 586c11ed7c user: drh tags: trunk)
21:23
Do not reuse factored constants that might have had their encodings changed. (check-in: 487f20366c user: drh tags: trunk)
20:48
Fix the code generator to honor turning off constant expression factoring. (check-in: 882622662d user: drh tags: trunk)
19:27
Add the --timer option to the wordcount test program. (check-in: a89fdf8755 user: drh tags: trunk)
19:05
Remove a test from sqlite3VdbeMemFromBtree() which was unnecessary, and after the recent OP_Column refactoring, unreachable. (check-in: 23667f3ba0 user: drh tags: trunk)
17:24
Make sure the OP_Next and OP_NextIfOpen opcodes are numbered close together for efficiency in switch() statements. OP_Prev and OP_PrevIfOpen too. (check-in: d4ccf0f5c6 user: drh tags: trunk)
16:08
When one or more arguments to a function are constants, try to factor out just those arguments into initialization code. (check-in: 50d350abbc user: drh tags: trunk)
14:44
Remove the obsolete TK_CONST_FUNC token type. (check-in: 9b4217f055 user: drh tags: trunk)
14:33
Add the ability to factor constant functions out of inner loops. But do not factor out non-constant functions, like random(). (check-in: 1b0f779e19 user: drh tags: trunk)
04:18
Another improvement to OP_Function and an improvement to OP_Move. (check-in: 70b056fb6f user: drh tags: trunk)
03:43
A simple change to the OP_Function opcode improves overall performance by about 0.5%. (check-in: b890eefd57 user: drh tags: trunk)
03:12
Performance optimization to the OP_Next and OP_Prev opcodes. (check-in: ecaac28a2e user: drh tags: trunk)
01:33
Refactor the OP_Column opcode to make it clearer and easier to maintain. Overall, performance tests show about a 1% speed increase with this change. (check-in: 972881c6d3 user: drh tags: trunk)