[Firebird-devel] What the Borland changes do.

[Ann W. H. <aha...@ib...>]

At 11:16 PM 5/24/2001 -0400, Claudio Valderrama C. wrote:

>Hello, can anybody reproduce <a bug involving rdb$dbkey
>with our latest sources or FB-0.95?
>
>I get "invalid column reference", a good description of what happens,

That's one of the changes we picked up from the Borland tree.
Here's a description of the changes the Borland team released with
their "certified" build.


         - example databases are now owned by sysdba

Not a big deal.  We should make sure that the qli message database is
no longer owned by Brett Bandy.

         - rdb$dbkey in the select list of a query with GROUP BY was 
dereferencing
           a null pointer

Claudio - here's the answer to your question.  In DSQL pass1.c, the
routine invalid_reference did not have code to handle rdb$dbkey.  That
routine is called to validate the selected columns in a query that includes
a group by.  Thus any reference to an rdb$dbkey in a grouped query would
involve trying to generate blr for a null node - works very badly.

         - a NULL state in a GROUP BY item would crash the server

The changes to evl.c (moving a call to EVL_make_value so it's handled
only if the value isn't null) should fix this.

         - initialize shadow_file to NULL_PTR _before_ calling SETJMP

This is part of the SDW_start routine in code that handles potential
errors.  It probably corrected the error that a missing shadow file
would keep a database from being opened.

         - don't let the server use environment variables
           ISC_USER and ISC_PASSWORD

Make the ISC_set_login routine conditional - works for classic, not
for shared server.  There's a second change to utl.c that moves an
sprint out of conditionally compiled code - no functional change.


         - if the server dies, ibguard was not able to restart the server

In the CNTL_main_thread routine of cntl_guard.c, there is a large if
statement that checks that everything is going well then and, if so
goes into a long wait.  Unfortunately the next line of code assumes
that it was the wait that failed, not one of the six conditions in
the if.  The new code adds a status which is cleared iff the if
test passes.

         - changing build number and license strings
         - some cosmetic changes

Boring stuff.

         - fix ods 9 problem

The code in ini.e adjusts the system metadata structures if the ODS
version is less than 10.  It removes references to two fields (RDB$PRECISION
and RDB$CHARACTER_LENGTH) from RDB$FUNCTION_ARGUMENTS).



Regards,

Ann
www.ibphoenix.com
We have answers.