I have uploaded DBD::ODBC 1.15_1 to CPAN. See below for a list of the
Changes. I'd appreciate any testing from anyone using DBD::ODBC as there
have been some very significant changes in error handling and describing
result-sets and repercussion of problems building/using on WIN64.
I also have a couple of outstanding questions re DBI::DBD development:
1. can I now throw away any DBIh_EVENT2 calls as they are now noops?
e.g. DBIh_EVENT2(drh, ERROR_event, DBIc_ERR(imp_drh),
DBIc_ERRSTR(imp_drh));
2. the set_err docs say setting err to "" is an informational state. I
am using that for some SQL_SUCCESS_WITH_INFO msgs. Under what conditions
do/can you see those?
3. I have a post on dbi-dev subject "Resetting odbc_err_handler - help
with dbd_db_STORE_attrib". For now I have replaced the code in question
with a test for SvOK. If anyone knows what the original author may have
meant by "if(valuesv == &sv_undef)" I love to hear from you.
Minor changes to 20SqlServer.t test for SQL Server 2008 (Katmai).
Timestamps now return an extra 4 digits of precision (all 0000) and
the driver reported in dbcc messages has a '.' in the version which
was not handled.
New FAQ entry and test code for "Datetime field overflow" problem in
Oracle.
Changed all ODBC code to use new SQLLEN/SQLULEN types where
Microsoft's headers indicate, principally so DBD::ODBC builds and
works on win64. NOTE: you will need an ODBC Driver Manager on UNIX
which knows SQLLEN/SQLULEN types. The unixODBC driver manager uses
SQLLEN/SQLULEN in versions from at least 2.2.12. Thanks to Nelson
Oliveira for finding, patching and testing this and then fixing
problems with bound parameters on 64 bit Windows.
Added private_attribute_info method DBI introduced (see DBI docs)
and test cases to 02simple.t.
Fairly major changes to dbd_describe in dbdimp.c to reduce ODBC calls
by 1 SQLDescribeCol call per column when describing result
sets. Instead of calculating the amount of memory required to hold
every column name we work on the basis that (num_columns + 1) *
SQL_MAX_COLUMN_NAME_LEN can hold all column names. However, to avoid
using a large amount of memory unnecessarily if an ODBC driver
supports massive column name lengths the maximum size per column is
restricted to 256.
Changed to avoid using explicit use of DBIc_ERRXXX in favour of newish
(ok, DBD::ODBC is a bit behind the times in this respect)
DBIh_SET_ERR_CHAR. This involved a reworking or the error handling
and although all test cases still pass I cannot guarantee it has no
other effects - please let me know if you spot differences in error
messages.
Fixed bug in 20SqlServer test for multiple results that was passing
but for the wrong reason (namely, that the odbc_err_handler was being
called when it should not have been).
Fixed bug in odbc_err_handler that prevented it from being reset so
you don't have an error handler. Looks like the problem was in
dbd_db_STORE_attrib where "if(valuesv == &PL_sv_undef)" was used to
detect undef and I think it should be "if (!SvOK(valuesv))".
Improvements to odbc_err_handler documentation.
Added 10handler.t test cases.
More tests in 02simple.t to check NUM_OF_FIELDS and NAMES_uc.
Bit of a tidy up:
Removed some unused variable declarations from dbdimp.c.
Lots of changes to DBD::ODBC tracing, particularly in dbd_describe,
and dbd_error2 and login6.
Removed a lot of tracing code in comments or #if 0 as it never gets
built.
Changed dual tests on SQL_SUCCESS and SQL_SUCCESS_WITH_INFO to use
SQL_SUCCEEDED.
Martin
--
Martin J. Evans
Easysoft Limited
http://www.easysoft.com
