Re: search_path vs extensions

On May 25, 2009, at 2:16 AM, Dimitri Fontaine wrote:

> Proposal: pg_extension, a new dedicated system schema for extensions
> Good:
>  It's easy to see SQL objects (\df) of extensions (think contribs) you
>  installed, and as extension developpers are required to use it, you
>  don't have to care about it any more.
>
>  As you have only one namespace for everyone, the collisions are
>  detected early.
> Not good:
>  As you have only one namespace for everyone, collisions prevent users
>  from installing several extensions using the same SQL object name, so
>  we'd need a way for extension authors to share a catalog of free
>  names, like internally we do for systems OIDs in the bootstrap,
>  IIUC. But in a distributed fashion.
>
>  We would have to add ways for the user to see which extension which
>  object belongs to, so you'd have extension | schema | object_name
>  columns in all \dX things, e.g.

I like this, although I'd want to be able, as a user, to override that  
default and tell an extension to install in some other schema. That  
would allow me to immediately overcome conflicts, and to organize my  
extensions if I want, rather than throw them all in one place.

> Proposal: Separate search_path into components: pre_search_path,
>  search_path, post_search_path
> Good:
>  This allows to easily separate who changes what: typically DBAs will
>  edit pre and post search_path components while application will care
>  about search_path the same way as now.
> Not good:
>  2 new GUCs (but no new semantics, and defaults to empty)

I don't follow this at all. How to the three components effect  
behavior? And what does this mean for where extensions are installed  
in schemas?

Best,

David