Message-ID: <4A8D9CA0.1040607@pjmodos.net>
Date: Thu, 20 Aug 2009 20:57:36 +0200
From: Petr Jelinek <pjmodos@pjmodos.net>
User-Agent: Thunderbird 2.0.0.22 (Windows/20090605)
MIME-Version: 1.0
To: PostgreSQL-development <pgsql-hackers@postgresql.org>
Subject: Re: GRANT ON ALL IN schema
References: <603c8f070908050951s3e5df452se4c610f01b80bb7d@mail.gmail.com>
	<4A7FE7F902000025000296E3@gw.wicourts.gov>
	<4A803A89.9020509@dunslane.net>
	<2CCE9C1D-919D-45D5-BFC2-C7723C45064E@hi-media.com>
	<162867790908150434j7aef36ecm202e81c4bb4740fa@mail.gmail.com>
	<4A86B5D5.6060807@dunslane.net> <4A871F46.1040002@agliodbs.com>
	<5D5ECAE0-F2B0-479B-9594-50AAF18F10ED@hi-media.com>
	<20090815223102.GS5407@samason.me.uk>
	<1250378139.18992.6.camel@vanquo.pezone.net>
	<603c8f070908152051mbd1d706hb0e29f5cfaf2f151@mail.gmail.com>
	<4A87854B.3070707@dunslane.net>
	<1250427428.26280.2.camel@vanquo.pezone.net>
	<395.1250439160@sss.pgh.pa.us>
In-Reply-To: <395.1250439160@sss.pgh.pa.us>
Content-Type: text/plain; charset=windows-1250; format=flowed
Content-Transfer-Encoding: 7bit

Tom Lane napsal(a):
> Peter Eisentraut <peter_e@gmx.net> writes:
>> Well, I don't know if we really need to call it "lambda", but I fully
>> expect to be able to use these "ad hoc functions" as part of other
>> expressions.
> 
> Why would you expect that?  To be used in an expression, you'd also need
> decoration to tell the function argument types, result type, volatility
> properties, etc etc (your proposed lambda notation is far too
> simplistic).  I think you're moving the goalposts to a point where we'd
> need ANOTHER, simpler, mechanism to accomplish the original intent.
> And frankly, all of the user demand I've heard is for the latter not
> the former.  By the time you get into specifying function properties
> you might as well just create a function.
> 

I agree with Tom here, doing it the way Andrew and Tom agreed on will be 
*way* easier and will give us most of the benefit (as Heikki said "90% 
of the usability with 10% of the trouble").

I volunteer to do this feature too.

The implementation as I see it would create function in pg_temp 
namespace, call it and then drop it. Any other implementation would imho 
mean rewriting procedure language api.

I am unsure if we should try to make the name of the function unique, 
since it should not collide with anything if we allow just one statement 
at a time (transactional DDL wins again), or am I mistaken here ?

Also do we want the LANGUAGE option to be at start or at the end or 
anywhere (like it's in CREATE FUNCTION). The reason I am asking this is 
that if we let user to put it on both sides then the LANGUAGE keyword 
can't be optional (what Dimitri Fontaine wanted).

And last thing I am wondering is if we want to allow DO to return rows 
(probably by creating the function with SETOF record as return type) ?
I am guessing not here since if user wants to run something often then 
he should crate a function.

Otherwise this should be quite straightforward (I have working code 
already).


-- 
Regards
Petr Jelinek (PJMODOS)