SQLite

Changes On Branch omit-column-cache
Login

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

Changes In Branch omit-column-cache Excluding Merge-Ins

This is equivalent to a diff from 21235d9a to a500893b

2018-08-04
20:30
Remove the column-cache from the code generator. The column-cache has been a persistent source of bugs for years and with recent improvements in the performance of OP_Column, it no longer provides a benefit. After the column cache is removed, the binary is almost 2KB smaller and the speed-check.sh performance test is over 3 million cycles faster. (check-in: cdff3af7 user: drh tags: trunk)
20:12
Remove a testcase() that is no longer reachable without the column cache. Provide an assert() to help prove that the testcase is no longer reachable. (Closed-Leaf check-in: a500893b user: drh tags: omit-column-cache)
17:15
Fix comments that were made obsolete by the removal of the column cache. (check-in: 2041231d user: drh tags: omit-column-cache)
15:16
Ensure that all expressions that are to be evaluated once at the start of a prepared statement (the Parse.pConstExpr expressions) pass the sqlite3ExprIsConstantNotJoin() test. It is not sufficient to pass just the sqlite3ExprIsConstant() test as that would allow through column references that are bound to constants by the WHERE clause in the constant propagation optimization. This fixes a problem discovered by OSSFuzz. (check-in: 8bc7f84c user: drh tags: trunk)
2018-08-03
23:04
Completely remove the column cache logic, which has been a persistent source of bugs for many years. Due to recent enhancements to the performance of the OP_Column opcode, removing the column cache actually makes speed-check.sh run faster. Removing the column cache also saves about 1,800 bytes of code space. (check-in: 3f5f60cd user: drh tags: omit-column-cache)
20:19
Fix the handling of sub-queries with LIMIT clauses by the optimization activated by compile-time symbol SQLITE_COUNTOFVIEW_OPTIMIZATION. (check-in: 21235d9a user: dan tags: trunk)
15:58
Fix the OP_SeekRowid opcode so that it has no type-change side-effects on the key register in P3. This fixes an obcure problem that arises when doing equi-joins between a table with a TEXT column against another table with an INTEGER PRIMARY KEY. The original problem was discovered when OSSFuzz created such a query and hit an assert() in OP_VerifyTabCol that was specifically designed to catch these kinds of errors at run-time. Test cases for this fix are in TH3. (check-in: fa94b49e user: drh tags: trunk)

Changes to src/build.c.

Changes to src/delete.c.

Changes to src/expr.c.

Changes to src/insert.c.

Changes to src/pragma.c.

Changes to src/select.c.

Changes to src/sqliteInt.h.

Changes to src/test1.c.

Changes to src/update.c.

Changes to src/vdbe.c.

Changes to src/vdbeInt.h.

Changes to src/vdbeapi.c.

Changes to src/vdbeaux.c.

Changes to src/vdbemem.c.

Changes to src/where.c.

Changes to src/wherecode.c.

Changes to test/btree02.test.