Re: csv format for psql

David G. Johnston wrote:

> Unaligned format could grow its own fieldsep if it wanted to but it can
> also just use the default provided fieldsep var whose default value is
> pipe. If it did grow one it would need to understand "not set" in order to
> preserve existing behavior. We don't have that problem with csv.

fielsep is already owned by the unaligned format. No other format
uses fieldsep currently. Why would it needs to grow its own?
I don't quite see what you mean here.

For csv, Fabien and Peter expressed the opinion that we shouldn't
create another fieldsep-like variable specifically for it, but instead
reuse fieldsep. That's what my latest patch does.

Now it turns out that sharing fieldsep comes with some problems.

1. since a single variable can't have two different default values,
we have either to accept '|' as a default csv separator, or introduce
some tricks to automagically commute the value.

2. when a user does \pset fieldsep ';' there are doubts
on whether this should affect the current mode only, or both,
or even the other one when the user goes back and forth between
formats. Any proposal where a \pset var value affects by
side-effect another pset variable is inconsistent with what has been
done until now with these variables, and I expect that
side-effects are not exactly sought after in general.

3. the command-line cannot express: use character 'X' for unaligned
and 'Y' for csv for the rest of the session. The current patch provides
'--csv' to select both the csv format and a comma separator, but
when combined with -F, the end result depend on the relative position
of the options, which may feel unintuitive or buggy.
Again we could tweak that, but only by being inconsistent
with how we process other options.

Personally I think the benefit of sharing fieldsep is not worth these
problems, but I'm waiting for the discussion to continue with
more opinions.

Best regards,
--
Daniel Vérité
PostgreSQL-powered mailer: http://www.manitou-mail.org
Twitter: @DanielVerite