Re: guc comment changes (was Re: Getting a move on for 8.2 beta)

On Mon, Sep 18, 2006 at 01:13:45PM +0200, Peter Eisentraut wrote:
> Joachim Wieland is in the process of reworking the original feature patch
> (resetting commented out parameters) in a much more compact form. But it
> turns out that there are a couple of very tricky situations involving custom
> variables, which may need more discussion than we have time for now.

The problem with custom variables is the definition of their default value.
First I defined it to be "variable does not exist" and tried to have a close
analogy to what happens on set/set local/reset with normal variables.
However that means that custom variables have to vanish (where other
variables show their default value) and revive (where other variables show a
changed value). For example if you have:

begin;
savepoint a;
set foo.var to 3;
(here we delete foo.var from the configuration file and send SIGHUP)
savepoint b;
reset foo.var;

There seem to exist quite a few possible definitions as to what happens if you
run "show foo.var" after the "reset" directly or after rolling back to the
savepoints a and b.

The result of the discussion with Peter was that all that seems not worth
doing since it is a rare special case. In the current form of the patch if you
delete a custom variable from the configuration file it gets deleted
internally and all reference to it results in error. There is still one
special case which is:

begin;
set foo.var to 3;
(here we delete foo.var from the configuration file and send SIGHUP)
commit;

This transaction will fail because commit cannot change the value of the
variable to 3 as requested by the user.

I have the patch almost ready in the described form, if there is any chance
it will make it into 8.2 I will clean it up and post it ASAP but Peter wrote
me that chances are close to zero and so I stopped working on it for now.

Joachim

--
Joachim Wieland joe(at)mcknight(dot)de
GPG key available