On Wed, Mar 07, 2018 at 08:04:05PM +0100, Fabien COELHO wrote: > > >> psql --csv -c 'TABLE foo' > foo.csv > >> > >>With a -c to introduce the command. > > > >This seems pretty specialized. If we're adding something new, how about > > > > psql --format=csv -o foo.csv -c 'TABLE foo' > > > >Or we could stick with: > > > > psql -P format=csv -o foo.csv -c 'TABLE foo' > > Currently "-P format=csv" uses the unaligned formating separators, i.e. '|' > is used. I was suggesting that a special long option could switch several > variables to some specific values, i.e. > > --csv > > Would be equivalent to something like: > > -P format=csv -P fieldsep=, -P recordsep=\n (?) -P tuples_only=on ...
We have some inconsistency here in that fewer table formats are supported, but I think asciidoc, etc., do this correctly via invocations like:
psql -P format=asciidoc -o foo.adoc -AtXc 'TABLE foo' > I.e. really generate some csv from the data in just one option, not many. > > But this is obviously debatable.
I suspect we'll get requests for an all-JSON option, HTML tables, etc., assuming we don't have them already.
I'm hoping we can have that all in one framework. I get that setting each of tuples_only, fieldsep, recordsep, etc. might be a bit of a lift for some users, but it's not clear how we'd make a sane default that made choices among those correct for enough users.
For example, do we know that we want tuples_only behavior by default? A lot of people's CSV tools assume a header row.
I am for default header - it can be disabled by -t option
