On Fri, Nov 09, 2018 at 05:28:07PM +0100, Daniel Verite wrote: > But again COPY is concerned with importing the data that preexists, > even if it's weird, whereas a psql output formats are not.
Hm. I checked the contents of the patch in details which provide output consistent with COPY, but after looking at the global picture I am getting cold feet on this patch for a couple of reasons: - This stuff adds new code paths in the frontend mimicking what the backend already does for years, both doing the same thing.
From the original post:
"copy with csv can't help for the output of meta-commands
such as \gx, \crosstabview, \l, \d ... whereas a CSV format within psql does work with these."
Formatting is usually a client-side concern so this feature fits well there. The fact that we’ve provided a server interface for the same doesn’t preclude its potential desirability in the client.
- There are already three ways to fetch data in this format with COPY, \copy and file_fdw, with all three using the same code paths for option validations (I can see the arguments at the top of the thread for which COPY SELECT can actually do everything you want with?).
Not always conveniently.
- The experience is confusing, as the psql format uses different options than the backend to do the same things:
Yes, those who use psql need to learn its features. I’d posit that since this syntax is being learned anyway that transferring said knowledge to a newly added csv format will not be confusing. No more so that having to do something that is usually client-side (formatting) on the server in the first place. That we don’t fully replicate the server functionality does’t bother be. This is meant to be a simple and quick ability that handles 95% of use cases and defers to the more powerful server version for the outliers.
This patch is not simple - not due own complexity, but due current state of psql and output format support. The psql is aged software and implement new format well known is not simple.
The COPY statement has different purpose and because it has server side implementation, it cannot to cover client side space.
I afraid so there hardy be designed some better - and it is unhappy so psql has not native csv support yet.
Regards
Pavel
Feature-wise I’m on board. Given it’s already written I’d say it should go in unless there are code complexity and support concerns - which given the prevalence of other formats I have to believe there are not.