[postgis-devel] Small regression test updates

[Andreas 'ads' Scherbaum]

Hi,

Am 20.07.2011 10:37, schrieb Sandro Santilli:
> On Wed, Jul 20, 2011 at 10:33:30AM +0200, Andreas 'ads' Scherbaum wrote:
>
> I know, but how does that give an actually predictable order ?
> Isn't this just exploiting a PostgreSQL implementation detail ?

How so?

The regression tests run a single SELECT which returns one or more rows, 
but every time only a single column.


Right now the sort order of the rows is not predictable. Most of the 
time you end up with the expected order, but this is not guaranteed by 
the database. As mentioned in my initial mail, we run into the problem 
that we don't see the same order every time we run the regression tests 
- in our distributed environment the resulting rows come from different 
PostgreSQL nodes and are not sorted in any way. Which node is faster 
ends up delivering the first row for the result set.

My patch changes this by applying a sort order. Every time the 
regression tests run with the "order by", the database sorts the results 
and you are guaranteed to get the same order - given the fact, that 
everything works well and you actually get the same data back from the 
SELECT.


What I do is using the column number in my patch instead of the column 
name - which is valid according to the SQL spec. This is nothing 
specific to PostgreSQL, it works in other databases too.


Regards,

-- 
				Andreas 'ads' Scherbaum
German PostgreSQL User Group
European PostgreSQL User Group - Board of Directors
Volunteer Regional Contact, Germany - PostgreSQL Project