On 2015-07-27 15:18, Michael Paquier wrote:
> On Sun, Jul 26, 2015 at 1:01 AM, Petr Jelinek wrote:
>> Yes that's what I meant by the change of checking order in the explanation
>> above. I did that because I thought code would be more complicated
>> otherwise, but apparently I was stupid...
>
> + In case the extension specifies schema in its control file, the schema
> s/schema/<literal>schema</>/
>
> +++ b/src/test/modules/test_extensions/.gitignore
> @@ -0,0 +1,3 @@
> +# Generated subdirectories
> +/results/
> +/tmp_check/
> test_extensions/.gitignore is missing /log/.
>
> Something also has not been discussed yet: what to do with new_version
> and old_version (the options of CreateExtensionStmt)? As of now if
> those options are defined they are not passed down to the parent
> extensions but shouldn't we raise an error if they are used in
> combination with CASCADE? In any case, I think that the behavior
> chosen should be documented.
>
I don't see why we should raise error, they are used just for the
top-level extension and I think it makes sense that way. CASCADE
documentation says SCHEMA option is cascaded to required extensions, do
we need to say something more than that (ie explicitly saying that the
old_version and new_version aren't)?
-- Petr Jelinek http://www.2ndQuadrant.com/ PostgreSQL Development, 24x7 Support, Training &
Services
