MariaDB Connector/ODBC 2.0.16 Changelog
The most recent Stable (GA) release of MariaDB Connector/ODBC is:
MariaDB Connector/ODBC 3.2.4
Download Release Notes Changelog About MariaDB Connector/ODBC
Release date: 8 Feb 2018
For the highlights of this release, see the release notes.
The revision number links will take you to the revision's page on GitHub. On GitHub you can view more details of the revision and view diffs of the code modified in that revision.
- Revision #d93b0b8
2018-02-04 23:01:24 +0100
- [ODBC-134] The fix and the test-case. Fetch would fail, if unbound column contained NULL, and for that column some arbitrary descriptor field was set by application(but not value/len/ind buffer ptrs). The bug affected ADO, as that is something it can do with CursorLocation adUseClient in some cases
- Revision #d506530
2018-02-01 10:20:13 +0100
- For the info type SQL_SCHEMA_TERM/SQL_OWNER_TERM the connector returned wrong length. Or more exactly - did not write anything into application's length buffer. The fix and the test-case. Possibly fixes ODBC-109.
- Revision #e7e12ae
2018-01-28 23:21:04 +0100
- [ODBC-119] The fix and the test-case. The connector ordered SQLStatistics results using wrong columns. That could cause MS Access to pick wrong column as a unique index. And that, in its turn, could cause mangled data shown for the linked table. Fix of the test-case in catalog2 that failed against 10.3 due to new privilege. Added test of SQL_GROUP_BY info to ODBC-123 test-case. It had similar problem(and tested in similar way) Fixed build in VS2017
- Revision #ce44b4d
2018-01-17 14:48:56 +0100
- Small fix of the testcase for SQLProcedureColumns - reference values for RADIX field were wrong for 2 parameters. Has to be 10(or theoretically 2) for numeric types
- Revision #8f34a4e
2018-01-26 16:21:06 +0100
- Fix bug in ABI analogous to [ODBC-123]
- Revision #dbc6409
2018-01-21 23:38:16 +0100
- [ODBC-131] The fix and the test-case. While linking a table, MS Access threw the error when it received unexpected length for SQLSMALLINT and SQLINTEGER columns from SQLColumns resultset. New graphic images for the installer on Windows.
- Revision #e84bdcc
2018-01-13 18:02:15 +0100
- Temporary turned off tests atainst MaxScale in Travis
- Revision #4f8bdb9
2018-01-12 17:50:40 -0500
- Addition to the fix of ODBC-123 and the testcase for that problem - for SQL_CATALOG_LOCATION info type the connector incorrectly wrote to the buffer SQLUINTEGER, while it has to be SQLUSMALLINT
- Revision #a22bf42
2018-01-10 20:54:38 +0100
- [ODBC-126] The fix and testcases. Some of internal structures weren't reallocated in accordance with the next resultset fields count. Beyond the case described in the report, they cover the case of statements batch. It had similar and own problems. Removed unused property from the statement structure.
- Revision #b085030
2017-12-27 21:46:48 +0100
- [ODBC-123] The fix and the test-case. LibreOffice sets SQL_ATTR_USE_BOOKMARKS attribute, but does not actually use bookmarks. The connector was not ready for that, and would throw wrongly error, and then crash. Fixed a number of compilation warnings.
- Revision #99ccc9e
2017-12-13 12:07:27 +0000
- Don't space pad parameters with non-zero scale
- Revision #528df54
2017-12-19 19:16:55 +0100
- Fix of waring from recent pull request
- Revision #4ab2f82
2017-12-13 10:49:18 +0000
- Return correct scale for SQL_C_NUMERIC
- Revision #7c445fb
2017-12-06 23:56:35 +0100
- Addition for the fix of ODBC-120 - optimization only used for FORWARD_ONLY cursors in this branch. Correction of merge error - tests detected Travis in wrong way
- Revision #c077544
2017-11-17 18:40:02 +0100
- Fixed connection procedure call in couple of tests, since 1 parameter type was changed in previous commit, and these references were overlooked. Skipping connstring in Travis. Fixed SQL_API SQLTablePrivilegesW - it would always return empty resultset if CatalogName is NULL. That was not intended, and its ANSI counterpart doesn't behave like that.
- Revision #d9c0ff4
2017-12-03 16:25:51 +0100
- Adding MariaDB 5.5 server to Travis tests. Skipping one tests with 5.5 servers(and MySQL <5.7), since it doesn't make sense there.
- Revision #27daadc
2017-11-27 22:32:02 +0100
- Fixing remaining issues in Travis Made (most of) tests framework global variables static, as one of them clashed with variable in connector's env, and that caused problems in some cases. Also fixed few tests in connstring, that expected test connection to use password. Skipped one of tests there(in connstring) on travis.
- Revision #0f69c54
2017-11-23 01:30:48 +0100
- [ODBC-120] Fix of the performance issue. We did redundant calls of mysql_stmt_data_seek. They are needed for different type of cursors, positioned operation, array fetch etc. But in forward_only cursor it only significantly slows down execution.
- Revision #553bc88
2017-11-14 13:02:13 +0200
- Fix of testcases failing in Travis. Mostly that is adding cursor closing, as older UnixODBC versions require it even there it is not really required. One test in catalog2(bug50195) was expecting, that db server is on the same host with tests(that is not the case in Travis). Added Travis detection in the tests framework.
- Revision #714acf9
2017-11-09 13:48:59 +0200
- Changed DSN parsing/saving/reading tests(connstring.c) in order they do not fail with unixODBC. UnixODBC has buggy ini cache(which seemingly will be fixed in 2.3.5): if dsn is changed, and then read again, application will get old cached value. That doesn't occur if unixODBC is built with `--enable-inicaching=no` Thus tests have been changed to create and use individual DSN's, rather then one common for all tests DSN.
- Revision #514a892
2017-10-19 15:51:54 +0200
- Initial Travis setup
- Revision #20c43cc
2017-10-09 19:58:56 +0300
- Fixed one small memory leak - SQLSetCursorNameW leaked recoded (into ANSI or utf8 encoding) name. Added freeing of allocated memory in a couple of tests, so they don't create noise in valgrind. Enabled test of "where current of" - it was skipped with misleading "unsupported" message, and it doesn't fail.
- Revision #e475a3d
2017-10-09 16:54:20 +0300
- Enabled test of updateable cursor (SQLSetPos with SQL_UPDATE), it was skipped with misleading "unsupported" message, and it doesn't fail.
- Revision #9ea9f21
2017-10-06 21:17:19 +0200
- Fix of the memory leak. It could occur if prepare, or direct execution in case of 10.2, failed, and Stmt handler is reused. Also removed unused field from Stmt struct and its references (was only freed in few places).
- Revision #bb86bf1
2017-10-04 23:48:52 +0200
- [ODBC-115] The fix and the testcase. Fix and the testcase. Wrong rc and sqlstate for numeric overflow (but not for fractional truncation). Now the connector returns 22003 and SQL_ERROR. Also changed int SQLFetch couple of switches to switch on Concise_Type, and not Type, that was seemingly wrong. Changed logo in the READMC.md
- Revision #33116e2
2017-08-28 12:27:17 +0200
- [ODBC-110] The Fix and the testcase of the crash in case of columns unbinding after stmt execution and before fetching data. The reason was freeing of 2 internal arrays that are allocated during execution and reset at each fetch. SQLFreeStmt(SQL_UNBIND) just shouldn't free them.
- Revision #064a189
2017-07-31 02:30:34 +0200
- [ODBC-105] Starting from v. 10.2.7, server encloses COLUMN_DEFAULT in the INFORMATION_SCHEMA.COLUMNS table, in single quotes for literal strings. Connector now considers the server version, when constructing the query for SQLColumns.
- Revision #ab661fb
2017-06-22 16:56:39 +0200
- Merge pull request #15 from jacquesg/include-guard
- Revision #cd24ea5
2017-06-22 14:12:19 +0200
- correct include guard
Comments
Comments loading...
Content reproduced on this site is the property of its respective owners,
and this content is not reviewed in advance by MariaDB. The views, information and opinions
expressed by this content do not necessarily represent those of MariaDB or any other party.