Nested Schemata, in a Standard-Compliant Way? - Mailing list pgsql-general

From Raymond Brinzer
Subject Nested Schemata, in a Standard-Compliant Way?
Date
Msg-id CANasJHnxbKhE+z0Gmcx+A5t-CCsZYzJJhYPR4-syvvOoP8pzjQ@mail.gmail.com
Whole thread Raw
Responses Re: Nested Schemata, in a Standard-Compliant Way?  (Tom Lane <tgl@sss.pgh.pa.us>)
List pgsql-general
Greetings.

For some people the "what?" and "why?" of this will be immediately
obvious from the title, but I'm going to spend a little time on those
before "whether?" and "how?"

We have schemata.  They're namespaces; very convenient for organizing
things.  They let you group tables and other entities together, and,
by setting search_path, only see the ones which presently interest
you.

In fact, they're pretty similar to directories in a filesystem...
except that they don't nest.  Imagine a filesystem where you could
have directories, but the directories could only contain files, not
other directories.  (Like the first Unix on the PDP-7, or DOS before
2.0.)

You could, of course, use your own delimiters.  And we do; often along
the lines of: schema.category_subcategory_table.  You can't really use
these to establish context, however. The system doesn't recognize
category_subcategory as a "place".  So you can't easily deal with a
subset of your tables, and the combination of many tables and long
names tends to be messy.

So, for example, I'd like to be able to say something like this:

SELECT * FROM /projects/contacts/people;

Or:

cd /projects/contacts;
SELECT * FROM people;

We use / for division, so that probably isn't plausible, but it makes
for a familiar example.

I'm wondering whether such a feature could be added, without breaking
either existing code, or compliance with the SQL standard.  For
instance, borrowing :: from languages like Ruby and Perl:

SELECT * FROM ::projects::contacts::people;  -- Absolute path
cd ::projects;                               -- Session-specific
SELECT * FROM contacts::people;              -- Relative path

I'm not necessarily saying this is the best delimiter, but the colon
isn't valid in unquoted identifiers, so it's probably a choice which
would have minimal impact.

Now, you could do a fair job of this just within the client, but my
thought is that this would be better if actually supported by the
database.  For instance, having representation in the system tables.

So, then:  can it be done?  Should it be done?  I can say easily that
my database life would be better for having this, but there do tend to
be those nasty lurking problems which aren't obvious.

-- 
Ray Brinzer



pgsql-general by date:

Previous
From: Tom Lane
Date:
Subject: Re: statement_timeout vs DECLARE CURSOR
Next
From: Tom Lane
Date:
Subject: Re: Nested Schemata, in a Standard-Compliant Way?