Re: [PATCHES] Proposed patch for sequence-renaming problems - Mailing list pgsql-hackers

From Bruce Momjian
Subject Re: [PATCHES] Proposed patch for sequence-renaming problems
Date
Msg-id 200510011810.j91IAfQ08710@candle.pha.pa.us
Whole thread Raw
In response to Re: [PATCHES] Proposed patch for sequence-renaming problems  (Tom Lane <tgl@sss.pgh.pa.us>)
Responses Re: [PATCHES] Proposed patch for sequence-renaming problems
List pgsql-hackers
Tom Lane wrote:
> Bruce Momjian <pgman@candle.pha.pa.us> writes:
> > Tom Lane wrote:
> >> Ripping out ALTER SCHEMA RENAME is not a solution unless you have a path
> >> to a solution later with more work.
> 
> > Well, we are only giving ourselves a few weeks to solve this, and I
> > think a hack to make it work cleanly for users is better than supporting
> > two function names perpetually.
> 
> Well, if you are dead set on having only one function name, then I think
> the best solution is this:
> 
>     * only one function, taking regclass
> 
>     * add an implicit text-to-regclass coercion
> 
> With this, nextval('foo') is early binding and nextval('foo'::text) is
> late binding, and existing dumps are going to continue to behave as late
> binding unless changed manually.
> 
> The implicit coercion is a bit risky, but in practice these are likely
> to be the only functions in the system that are declared to take
> regclass, so the odds of the implicit coercion firing unexpectedly seem
> low.
> 
> Does that sound like a workable compromise?

Personally, I _love_ it.  I hope others do as well.  :-)

Let me explain why I thought two function names would be confusing.  We
have been telling people to use nextval() since we added sequences in
6.4, and since 99% of people would want early binding (with
dependencies), I think making them all move to a new function name would
be a long-running education effort.  If it can be avoided, that is
better.

I think the solution you propose is great because:
o  it fixes SERIAL dependencyo  it allows old dumps to load with no change in behavioro  it allows new nextval() calls
tohave early binding, unless   ::text is used
 

In fact, the use of ::text is dump is the only thing that is _allowing_
this migration idea to work.

I think I can easily explain this issue in the release notes, and show
how users can update their schemas to the new behavior.

--  Bruce Momjian                        |  http://candle.pha.pa.us pgman@candle.pha.pa.us               |  (610)
359-1001+  If your life is a hard drive,     |  13 Roberts Road +  Christ can be your backup.        |  Newtown Square,
Pennsylvania19073
 


pgsql-hackers by date:

Previous
From: Tom Lane
Date:
Subject: Re: \d on database with a lot of tables is slow
Next
From: Tom Lane
Date:
Subject: Re: [PATCHES] Proposed patch for sequence-renaming problems