Detection of nested function calls

[robe]

This is a topic that Hugo Mercier brought up to PostgreSQL Hackers and is being widely discussed now and something Oslandia has been prototyping ideas of.

​http://www.postgresql.org/message-id/526A61FB.1050209@oslandia.com

It involves the loss of power we suffer when deserializing/serializing and affects all of us, particularly for writing sql/plpgsql functions where nesting is very common.

Thanks - Hugo for bringing the issue to the forefront. I'm hoping something comes of this and we can leverage changes upstream in PostGIS and PostGIS family of projects

[robe]

Milestone renamed

[pramsey]

I have performed a couple experiments with "expanded object headers" and they are quite unpleasant but can be used to remove serialize/deserialize overhead. For most of our use cases, though, our deserialize step just _is not that bad_ and the penalty for using EOH (at least one full copy into a parent memory context, and added complexity in call handling) is not worth it. Only for deserializations where there is a fair amount of overhead (like a decompression, or calculations) does EOH start to make sense.