MySQL Native Driver: Known Problems

From Apache OpenOffice Wiki
Jump to: navigation, search


This page lists the currently known problems with the MySQL Native Driver, aka MySQL Connector/OOo. A list of fixed problems is also available.

End User Observations

This section collects issues as observed by end users

Template for new entries

reported by NAME on 2009-mm-dd - needs review

Description goes here.

Severity: undecided

Reason: unknown

On windows, the extension crashs on some systems

reported by FS on 2009-04-09

On some Windows systems, if you install and use the extension, it will crash OOo.

Severity: required

Reason: The extension ships an own version of libmysql.dll. When Windows loads the driver library (mysqlc.uno.dll), it also implicitly loads the libmysql.dll, which mysqlc.uno.dll is linked against. Unfortunately, Windows first searches in the folders specified by your PATH environment variable. If it finds a libmysql.dll in one of those folders, this one is used - no matter whether it is really compatible.

Suggested solutions:

  • rename libmysql.dll, and link against this renamed version. Argh, not really.
  • link libmysql.dll statically
  • load libmysql.dll explicitly (using LoadLibrary). In this case, we would have control over where the lib is taken from
  • make libmysql a managed assembly, having a identifer, and a version, which changes with every incompatible API/ABI change (and differs between the debug and the release version). In this case, Windows would ignore all libmysql.dll versions found in the PATH which do not proper version.

TODO (Andrey): fix

Copying tables may not work properly

reported by NAME on xx/01/2009 - needs review

Needs further and systematic testing!

Original table:

 CREATE TABLE `source` (
 `id` int(11) NOT NULL,
 `col1` int(11) NOT NULL,
 `col2` char(1) default NULL,
 `col3` varchar(50) default NULL,
 PRIMARY KEY  (`id`),
 KEY `idx_col1` (`col1`),
 KEY `idx_col2` (`col2`)
 ) ENGINE=InnoDB DEFAULT CHARSET=latin1

Copy:

 CREATE TABLE `source2` (
 `id` int(10) NOT NULL,
 `col1` int(10) default NULL,
 `col2` char(3) default NULL,
 `col3` varchar(150) NOT NULL,
 PRIMARY KEY  (`id`)
 ) ENGINE=MyISAM DEFAULT CHARSET=latin1
  • Wrong column definitions (known issue)
  • NOT NULL not properly handled
  • Index information not copyied (same with JDBC, might be a Base issue)

Severity: undecided

Reason: unknown

TODO (Ulf): check what problems we actually have here, and how severe they are (compared with a connection vis ODBC/JDBC).

Wrong column width reported for text/binary columns

reported by Ulf on 09/01/2009 - needs review

We discussed this before and it might be fixed in the current source, but I work on an older snapshot. We return number of bytes, OO.org wants number of characters. A typical follow-up error is that copying tables may fail and/or give wrong results.

From an earlier mail exchange:

 Hintergrund: für eine VARBINARY(10)-Spalte in MySQL liefern wir aktuell
 40, für eine INTEGER-Spalte in MySQL liefern wir aktuell 11. Wir liefern
 momentan 40 für VARBINARY(10), weil das Feld maximal 10 Zeichen aufnehmen kann, die maximal 4 Bytes für das Encoding benötigen. Für ein  VARCHAR(255) wird sogar VARCHAR(255) der Wert von 3060 geliefert, weil hier der MySQL Server nochmal in einem seiner Datenfelder einen Mul  tiplikator von 3 einbaut (255 * 3 * 4 = 3060). 

And:

 Später am Abend fiel es mir wieder ein... wir ziehen getColumnDisplaySize() intern im Connector/C++ ran für ConnectionMetaData::getColumns() -> COLUMN_SIZE.
 rs_data.push_back(my_i_to_a(buf, sizeof(buf)-1, (long) rs3_meta->getColumnDisplaySize(i))); // Column size 

Pointer: MySQL ODBC file utitlity.c get_column_size

Severity: required

Reason: unknown

TODO (Chris): check and classify severity

Renaming a view "makes" it a table

reported by Ulf on 09/01/2009

Tested with the JDBC driver. Create a view and rename it in Base. Base changes the icon in the table list and shows the renamed object as a table :-)

clu: works also wrong with native mysql driver -> works fine with odbc driver

Severity: low

Reason: unknown

Removing records from views fails

reported by Ulf on 08/01/2009

Removing records from SQL views neither works with the JDBC driver nor with the native driver. Base sends a misformed query like: "DELETE FROM `test`.`v` W"

Severity: low

Reason: Removing records from views is in general not possible in Base. The fact that "Edit/Delete record" is enabled for a view (as in any table which the user does not have the DELETE privilege for) is a bug, which I fixed in the CWS.


No way to set MySQL specific table attributes

The Base table editor does not give access to table attributes. Not even basic ones such as the Engine (MyISAM: non-transactional, InnoDB: transactional).

Severity: 1.1

Base does not recognize schema changes

After connecting to a database and opening a table once, Base will not recognize changes applied to the DB schema meanwhile when opening the table in the table editor again.

Severity: low

that's not nice, but consistent with other DB/Drivers. For this purpose, there's View/Refresh Tables ...

Table column comments not synced between MySQL and Base

Base table column comments are not synchronized with the MySQL DB and its schema. Existing comments are not displayed in Base, and entering comments in the table editor is not propagated to MySQL.

Severity: later

That's a known issue with all database types. The column description as displayed in Base is purely client-side, and stored within the .odb file only. There's also an issue for this, but I'm too lazy too search for it right now ...


Default values not properly processed

The Base table editor neither properly sets default values nor does it re-engineer default values properly.

Severity: 1.1

Again, this is a known issue. The default value displayed in the UI is a so-called "control default", which is applied to controls used to enter data into the given field. The DB-side default for a column is a different property, API-wise, and currently not evaluated at all. Probably not even properly fetched by most existing drivers.

Changing this is possible, but probably requires UI changes. First, we would need to define how the control default and the DB default should interact in the UI. A possible scenario would be to drop the UI support for the control default, and always use the DB default (even in controls), as long as the driver supports providing/accepting DB defaults.

BIGINT values crippled

Large BIGINT values are displayed in Base using scientific notation: 1e+15. If one changes the display format to number #.### the first 14 values of a large number (9223372036854775807) are displayed properly but then some rounding takes place, for example: 9223372036854800000. Connector/C++ can handle long long (L64) values properly and Base does use getLong() nevertheless the displayed value seems wrong.

Severity: later

I bet that's because of the number formatter ... processing numbers for display is done using a office-wide number formatter component. Unfortunately, it works with double-precision values only, which imposes a precision loss for certain values. This applies to "too-large" values, as well as fractional values with "too many" digits. This is a general problem in Base, and not limited to the MySQL Native Driver.

A possible solution would be to 'not employ the number formatter for certain column types. (Effectively, this means not using a FormattedField for the respective table column in the data view, but a NumericField, which internally works with long integer values of arbitrary (?) length.) This would solve the BIGINT issue, though not the too-many-digits issue for fractional values.

fieldtype is changed after saving

reported by CLU on 12/01/2009

in table edit view: create a field from f.e. type bool & save → tinyint

Severity: desired

Reason: maybe the same problem like 1.10 & 1.11

(Ulf) Duplicate and not an error in C/C++. There is no boolean type in MySQL. However, its handled in the JDBC driver case, the best we can do is emulate it in the same way.

(FS): Really, if there is no BOOL type, I tend to think the driver should not offer it. This way, users are not tempted to use it. Finally, this is a dedicated MySQL driver, it must not necessarily emulate features it does not really have.

(CLU): agree with FS .. every proposed type has to work, any unsupported type must not be shown

TODO (Ulf): analyze reason, and suggest possible solution: "emulate" BOOL like the JDBC driver does (might yield other problems), or drop BOOL from the list of supported types

Currency Value

reported by Mechtilde - needs review

This note relate to 4.2. I can't input currency values. I get there 0.00.

TODO (Chris): check and confirm or deny

Developer Observations

This section collects observations interesting for developers only. It's likely the describe the root cause for another problem listed in the End User Observations section. In this sense, they might be duplicates, but we currently just use this as reminder list whenever we notice a problem ...

Template for new entries

reported by NAME on 2009-mm-dd - needs review

Description goes here.

Severity: undecided

Reason: unknown


getProcedureColumns is a stub

reported by Ulf on 2009-01-27

http://pastebin.com/m75809

Severity: later

Reason: unknown

getImportedKeys()/getExportedKeys()

reported by Ulf on 2009-01-27 - needs review

Most likely Connector/C++ does not distinguish between "" and NULL parameter values for catalog and schema as requested by the JDBC specs. Do we need the differentiation?

Severity: undecided

Reason: unknown

TODO (Ulf): do we still/really have a problem here? Severity?

getWarnings() (C/C++)

reported by Ulf on 2009-01-26

getWarnings() always returns a warning object, regardless if there is one or not. Should it return NULL? JDBC compliance test expects NULL.

Severity: undecided

Reason: Connector/OO.org

getWarnings() (C/OOo)

reported by FS on 2009-01-29

C/OOo's getWarnings is implemented as dummy only. It should at least retrieve the warnings from the underlying C/C++ connection, and concert them to css.sdbc.SQLWarnings as appropriate.

Severity: later

Do we need any of those?

ERROR: SQL Exception ODatabaseMetaData::getCrossReference: feature not implemented.
ERROR: SQL Exception ODatabaseMetaData::getProcedures: feature not implemented.
ERROR: SQL Exception ODatabaseMetaData::getUDTs: feature not implemented.
ERROR: SQL exception ODatabaseMetaData::supportsIntegrityEnhancementFacility: feature not implemented.
ERROR: SQL Exception ODatabaseMetaData::supportsResultSetConcurrency: feature not implemented.
ERROR: SQL Exception: OResultSet::getObject: feature not implemented.
ERROR: SQL Exception ODatabaseMetaData::getExportedKeys: feature not implemented.

FS: supportsIntegrityEnhancementFacility is used to determine relationship support, but there's meanwhile an exception in the code which enables relationships for MySQL, regardless. (using this method is incorrect, speaking strictly, anyway.) supportsResultSetConcurrency is not used by default, but its usage can be enabled by the user on a per-database basis. So, it might be better to implement it.

getObject - not sure.

The other three are not used, I think.


DatabseMetaData::getColumns() COLUMN_SIZE

DatabseMetaData::getColumns() takes COLUMN_SIZE from getDisplaySize().

Affected: DECIMAL/FLOAT/VARCHAR.

Severity: required

Note: DisplaySize is not the same as ColumnSize. The former is a hint for UI processors which are to display the data in a tabular way. Currently, Base does not make use of this attribute. The ColumnSize's semantics is described in the Java API, and Base uses it in the very same semantics. As a consequence, COLUMN_SIZE is what needs to be tweaked.

DatabseMetaData::getColumns() DECIMAL_DIGITS

DatabseMetaData::getColumns() does not set DECIMAL_DIGITS. Therefore, Base does not show it properly.


Potential Issues

This section serves as collection of issues which might become a problem to end users, though we did not yet (bother to) create scenarios where users are hit by those (potential) problems.

TIME value range

TIME data type may have a significantly larger value range than the Base counterpart.

Severity: undecided

TINYINT / BOOLEAN

Educate Base users on MySQL Server type mappings, for example BOOL[EAN] -> TINYINT.

Personal tools