"Bert Huijben" <b...@qqmail.nl> writes:
>> Index: subversion/libsvn_wc/wc-queries.sql
>> ================================================================
>> ===
>> --- subversion/libsvn_wc/wc-queries.sql (revision 1692140)
>> +++ subversion/libsvn_wc/wc-queries.sql (working copy)
>> @@ -818,6 +818,7 @@ WHERE wc_id = ?1
>> OR IS_STRICT_DESCENDANT_OF(local_relpath, ?2))
>> AND op_depth = ?3
>> AND presence NOT IN ('base-deleted', 'not-present', 'excluded',
> 'absent')
>> +ORDER BY local_relpath
>>
>> -- STMT_INSERT_WORKING_NODE_FROM_BASE_COPY
>> INSERT INTO nodes (
>
> Yes this is the right fix for that problem.
>
> While the code that processes this query result doesn't rely on the ordering
> of the result, the maintainer mode checks do rely on that ordering.
>
> This fix (r1567080) is part of a batch of fixes that I found by asking
> Sqlite to return result rows backwards in all cases where queries didn't ask
> for an explicit ordering. (Now triggered by setting
> SVN_SQLITE_REVERSE_UNORDERED_SELECTS during compilation of libsvn_subr; but
> I don't think that support was already available in 1.7)
I backported SVN_SQLITE_REVERSE_UNORDERED_SELECTS to 1.7 and the above
fix to STMT_INSERT_DELETE_FROM_NODE_RECURSIVE is sufficient to PASS the
regression tests when reverse is enabled. Reverting r1687152 is the
most reliable way to avoid a regression but fixing
STMT_INSERT_DELETE_FROM_NODE_RECURSIVE is probably sufficient.
--
Philip Martin
WANdisco
