GDA missing records retest (long)

[Phil Longstaff]

Mark Johnson wrote:
> While looking at the PostgreSQL error log, I also noted the following error:
> ERROR:  relation "books" does not exist
> STATEMENT:  INSERT INTO books (guid, root_account_guid, 
> root_template_guid) VALUES ('be5d5b80c1a8f36ec33139cb27440ca7', 
> '8f436420bffa2ba45515192c895b083b', '2c54584a7995ce449f2fd371771ef562')
> PROBLEM: the books table is not created.  I checked all 3 DBs.

When the gda backend saves to a db, it deletes all of the tables and
then recreates them.  Between these, it calls a libgda function to
refresh its knowledge of the scheme.  This function retrieves the new
schema and removes entries from its internal table list which are not in
the new schema.  Unfortunately, this function has a bug so that all
entries except the 2nd one are removed.  Then, when the gda backend
tries to recreate the tables, it doesn't recreate that table because it
"already exists".  This means 1 table is not created until the next time
the db is opened.  At this time, the missing table is not in the libgda
table list, so the gda backend creates it.

When I run into this problem, I see it with the billterms table.  I',
not sure, but I guess it is possible that it might also affect the books
table if it was the 2nd table in the list (accounts is 1st).

As I workaround to the bug, I could try a direct operation on the table
and see if there is the right error message.  The real fix is to fix the
libgda bug, of course.

Phil