Re: PATCH: pgbench allow '=' in \set

On Thu, May 7, 2015 at 2:18 PM, Fabien COELHO <coelho@cri.ensmp.fr> wrote:
> Since an expression syntax has been added to pgbench, I find that the
> readability of expressions is not great. An extra '=' improves the situation
> for me:
>
>    \set id = 1 + abs((:id * 1021) % (100000 * :scale))
>
> seems slightly better than:
>
>    \set id 1 + abs((:id * 1021) % (100000 * :scale))
>
> But that is debatable!

I would like to debate that.  :-)

Generally, I don't like optional noise words.  A good example of how
this can bite you is the whole problem with LOCK TABLE foo.  You can
also write that as LOCK foo.  What if you want to lock something
that's not a table?  If the word "table" were required, then you could
support LOCK VIEW foo or LOCK MATERIALIZED VIEW foo or even LOCK
FUNCTION foo().  But because the word TABLE is optional, that syntax
creates parser ambiguities.  We have not been able to agree on a
reasonable solution to this problem, and it has blocked implementation
of this important and useful feature for years.  While I can't foresee
what trouble specifically your proposal might cause, I have become
wary of such things.

I also don't like to restate what has already been said.  \set at the
beginning of the line tells you that you will be setting a variable.
Adding = or := only restates the same thing.  I agree it superficially
looks a little nicer, but I'm not sure it's really going to add
clarity, because it's basically just redundant.

-- 
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company