I've just uploaded DBD::ODBC 1.46_1 to the CPAN. In the process of
writing
http://www.easysoft.com/developer/languages/perl/sql-server-unicode.html
and https://github.com/mjegh/dbd_odbc_sql_server_unicode I discovered a
serious bug in the way DBD::ODBC can attempt to insert unicode
characters into char/varchar/longvarchar columns. This experimental
release fixes that issue but it does mean this release contains a
significant change in behaviour.
The issue ONLY applied to unicode builds of DBD::ODBC (the default on
Windows) and enabled in Linux/Unix via the -u switch to Makefile.PL.
The problem was that when inserting parameter data into
char/varchar/longvarchar columns DBD::ODBC ignored what your parameter
actually looked like and simply bound the parameter as the type
described by the database (SQL_CHAR). This meant that if you bound data
was unicode, the separate octets of the perl UTF-8 encoded data would be
inserted instead of the unicode characters. A simple example illustrates
this easiest:
Say you had a unicode euro in a perl scalar. This is U+20AC and is
encoded in UTF-8 as 0xe2,0x82,0xc2. If you inserted into a
char/varchar/longvarchar the database would receive it as 3 separate
chrs instead of 1 i.e., select len(mycol) from mytable would return 3
instead of 1.
There are a few situations when this did not apply 1) if you overrode
the bind type with SQL_WVARCHAR 2) if your ODBC driver did not support
SQLDescribeParam or you told DBD::ODBC not to use it.
A new test (45_unicode_varchar.t which has high verbosity set right now)
has been added to the test suite. Unfortunately, this test only runs to
MS SQL Server right now. If this test does not pass for you please
report it and the output to me as soon as possible.
You are strongly advised to test this release with your development
environment as I've not implemented a deprecation policy for this change
as yet. I'm hoping to release a full version as 1.46_1 is, BUT if it is
reported to me that this will cause too many people problems I'll
reconsider.
I full description of the issues and other problems can be found at
http://www.easysoft.com/developer/languages/perl/sql-server-unicode.html
and https://github.com/mjegh/dbd_odbc_sql_server_unicode.
Here is a full list of the changes since 1.45:
[CHANGE IN BEHAVIOUR]
As warned in release 1.45, the binding of unicode parameters to
char/varchar columns has changed significantly - see warning at
start of DBD::ODBC pod. If you don't attempt to insert unicode into
char/varchar columns or if you only inserted unicode into
nchar/nvarchar columns you should see no difference.
Previously if DBD::ODBC received an error or (SQL_SUCCESS_WITH_INFO)
from an ODBC API call and then the driver refused to return the
error state/text DBD::ODBC would issue its own error saying "Unable
to fetch information about the error" and state IM008. That state
was wrong and has been changed to HY000.
[BUG FIXES]
Some drivers cannot support catalogs and/or schema names in
SQLTables. Recent changes set the schema/catalog name to the empty
string (good reasons below) which causes "optional feature not
implemented" from MS Access (which does not support schemas - even
for a simply ping (which uses SQLTables)). Now we call
SQLCATALOG_NAME and SQLSCHEMA_USAGE on connect to ascertain support
which modifies SQLTables call.
[MISCELLANEOUS]
Added test 45_unicode_varchar.t for MS SQL Server only so far.
Martin
--
Martin J. Evans
Wetherby, UK
