Re: Table modification

>1) I don't want to retrieve source from logs. I want to analyse the database
>and generate the source from what is known to work and compile correctly.
>There is no better test for an object than to build it in PostgreSQL.

When an object is broken, its source code disappears or cannot be retrieved
because PostgreSQL internals are based on OIDs.
Example: if you create a view based on a function and drop the function,
you will not be able to read the view source code.

When several developers are working on the same database at the same time,
you cannot guarantee consistency at PostgreSQL level.
Developer1 will be working in function1, while developer2 will be working
no function2 based on function1. Developer circular dependency. Boom!

When an object is broken, with your system, you have to retrieve source
code from revision logs.
I don't know if we can speak of ACID rules, but ... this is not a
consistent and atomic choice.

>2) Yes, compilation must be in dependancy order (which Publishing is). I
>don't know how we could ever get over my circular dependency problem though.
>That *would* require logs to figure out the sequence of queries required to
>create the circular dependency in the first place. And yes, I do agree that
>that is nasty...
>A good heated debate does brighten up the day don't you think?!

Yes my friend. Indeed, we are discussing about the (future) schema features
of PostgreSQL.
We need more input on PostgreSQL core team schema and dependency projects
... or start our own project.

/Later, Jean-Michel