279fcddc23 · When switching wal file in wal2 mode, ensure that a checkpointer cannot take a lock that causes a writer to return a spurious SQLITE_BUSY error. Forum post [forum:63ba0208a6 | 63ba0208a6]. · Updated 2026-11-09 14:42:58 +00:00
2794e4c059 · Experimental: fts5vocab leaves range constraints and ORDER BY term to the core when PRAGMA encoding is denied, ignored, or otherwise not confirmed UTF-8. AI-assisted proof of concept for https://sqlite.org/bugs/forumpost/b0c03988e894. The previous candidate returned SQLITE_AUTH from xCreate/xConnect when an authorizer denied PRAGMA encoding, so the vocab table would not open, and it treated a missing pragma row as UTF-8, so an authorizer IGNORE kept the wrong UTF-16 memcmp plans. Equality constraints stay consumed. SQLITE_NOMEM and any other prepare, step, or finalize error still fails the constructor. Add ext/fts5/test/fts5vocab5.test. · Updated 2026-11-09 07:07:04 +00:00
439841af28 · Experimental: fts5vocab range constraints on term and ORDER BY term are left to the core when the database text encoding is UTF-16. AI-assisted proof of concept for https://sqlite.org/bugs/forumpost/b0c03988e894. The FTS5 index visits terms in memcmp() order of their UTF-8 encoding, which is the BINARY collation order only in a UTF-8 database; fts5VocabBestIndexMethod() consumed <, <=, >, >= constraints and set orderByConsumed regardless of the encoding, so in a UTF-16 database range queries dropped rows and ORDER BY term returned UTF-8 order. Read PRAGMA encoding once in xCreate/xConnect and keep equality constraints. Add ext/fts5/test/fts5vocab4.test. · Updated 2026-11-09 05:20:35 +00:00
bd874ce46e · Experimental AI-assisted proof of concept (Claude Fable 5.1), not for trunk: in sqlite3SubqueryColumnTypes(), use multiSelectCollSeq() for every compound subquery column rather than only when the left-most term has no collating sequence, so that any special handling of a term inside multiSelectCollSeq() (such as a multi-row VALUES term) also applies to the column of the subquery. No behavior change on its own. Bug forum post 52e3f2057f81. · Updated 2026-11-09 05:06:33 +00:00
e49a697692 · Experimental AI-assisted proof of concept (Claude Fable 5.1, xhigh): in sqlite3WhereEnd(), also retarget the OP_IsNull opcodes of the second and subsequent components of a row-value IN past the OP_IfNoHope. Those jumps landed on OP_IfNoHope before OP_Affinity had converted the prefix registers, so a NULL in a later component of the first RHS row visited could abandon the IN loop on a false-positive. Adds tests in6-3.140 through in6-3.150. Report: https://sqlite.org/bugs/forumpost/4420a6b645f2 (Bug 2026-11-04T00:58:35Z). Based on trunk 1fea2d9c18. Private local branch; maintainer review required. · Updated 2026-11-09 04:03:15 +00:00
85de66bb07 · Experimental: json_each() and json_tree() escape a double-quote, and a backslash in a TEXTRAW label, when building the fullkey and path columns. AI-assisted proof of concept for https://sqlite.org/bugs/forumpost/53eb99b97a30 by Claude Fable 5.1 (xhigh) under Codex Fleet Supervisor oversight. jsonAppendPathName() copied the stored label text between quotes without escaping, so a double-quote stored unescaped in a TEXT5 label from a single-quoted JSON5 string, or in a TEXTRAW label from an unquoted path label, produced a path that the path parser could not read back. Add tests to json502.test. Based on trunk 1fea2d9c18. Private local branch; maintainer review required. · Updated 2026-11-09 02:57:37 +00:00
f429a5da87 · Experimental: json_error_position() reports the position of non-whitespace text that follows a complete JSON value. AI-assisted proof of concept for https://sqlite.org/bugs/forumpost/370226230047 by Claude Fable 5.1 (xhigh) under Codex Fleet Supervisor oversight. jsonConvertTextToBlob() returned failure for trailing text without recording pParse->iErr, so jsonErrorFunc() reported the parser's last stored offset: 0, or the offset of the ']' or '}' that closed an empty container. Set iErr to the start of the trailing text. Add tests to json101.test. Based on trunk 1fea2d9c18. Private local branch; maintainer review required. · Updated 2026-11-09 02:57:34 +00:00
a942c98416 · Experimental AI-assisted proof of concept for Bug Forum df6692fc080c: a 'utc' modifier that converts the time no longer discards an earlier 'subsec'. Claude Opus 5.5 high investigation and candidate under Codex Fleet Supervisor oversight. Based on trunk 1fea2d9c18. Private local branch; maintainer review required. · Updated 2026-11-09 02:13:38 +00:00
97c66fe9a6 · Experimental AI-assisted proof of concept for integrity_check report [bugs:/info/b2f9f58ae6bf|b2f9f58ae6bf]: report surplus entries in a partial index by comparing the number of rows that satisfy the index WHERE clause with the entry count from the b-tree walk, as was done before check-in [cc294c041b4c7a04]. Claude Fable 5.1 xhigh candidate under the Codex fleet supervisor. Based on trunk [1fea2d9c18fc5f84]. Maintainer review required. · Updated 2026-11-09 01:13:30 +00:00
12ff9dd06d · Experimental AI-assisted proof of concept for dbstat reports 14031a27db, 113506a561, 7e102e0e60. Claude Opus 5.5 high candidate; Gemini 4 Argon high review; Codex fleet supervisor validation. Based on trunk 9a42b5817055e6ad. Maintainer review required. · Updated 2026-11-07 20:33:24 +00:00
781d8c538b · Experimental AI-assisted proof of concept for date modifier reports 7757ac1132, 3cc3a0207e, e0fe679e7c. Claude Opus 5.5 high candidate; Gemini 4 Argon high review; Codex fleet supervisor validation. Based on trunk 9a42b5817055e6ad. Maintainer review required. · Updated 2026-11-07 20:33:06 +00:00
578f753d88 · Return SQLITE_LOCKED if an ALTER TABLE DROP COLUMN statement is attempted when there are other statements belonging to the same connection reading the database. Report [bugs:/info/2026-10-18T20:41:32Z | 2026-10-18T20:41:32Z]. · Updated 2026-11-02 20:11:43 +00:00
19dc478897 · Add a refcount to the WASM/JS auxdata dtor binding, keyed on the input Function object. Add a test which triggers the dtors more frequently (further demonstrating how costly this automated conversion is). · Updated 2026-10-29 23:01:58 +00:00
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?