Re: nRe: [PATCH v1] Show whether tables are logged in \dt+

On Sat, Apr 27, 2019 at 09:19:57AM +0200, Fabien COELHO wrote:
>
> Hello David,
>
> Patch v3 applies, but compiles for me with a warning because the indentation
> of the following size block has been changed:
>
> describe.c: In function ‘listTables’:
> describe.c:3705:7: warning: this ‘if’ clause does not guard...
> [-Wmisleading-indentation]
> else if (pset.sversion >= 80100)
> ^~
> describe.c:3710:3: note: ...this statement, but the latter is misleadingly
> indented as if it were guarded by the ‘if’
> appendPQExpBuffer(&buf,
> ^~~~~~~~~~~~~~~~~

Fixed.

> Make check fails because of my temp schema was numbered 4 instead of 3, and
> I'm "fabien" rather than "shackle".

I think the way forward is to test this with TAP rather than the
fixed-string method.

> > > > > > Included, but they're not stable for temp tables. I'm a little stumped
> > > > > > as to how to either stabilize them or test some other way.
> > > > >
> > > > > Hmmm. First there is the username which appears, so there should be a
> > > > > dedicated user for the test.
> > > > >
> > > > > I'm unsure how to work around the temporary schema number, which is
> > > > > undeterministic with parallel execution it. I'm afraid the only viable
> > > > > approach is not to show temporary tables, too bad:-(
>
> The tests have not been fixed.
>
> I think that they need a dedicated user to replace "shackle", and I'm afraid
> that there temporary test schema instability cannot be easily fixed at the
> "psql" level, but would require some kind of TAP tests instead if it is to
> be checked. In the short term, do not.

Checks removed while I figure out a new TAP test.

> I checked that the \di+ works, though. I've played with temporary views and
> \dv as well.

Great!

> I discovered that you cannot have temporary unlogged objects, nor
> temporary or unlogged materialized views. Intuitively I'd have
> thought that these features would be orthogonal, but they are not.

This seems like material for a different patch.

> Also I created an unlogged table with a SERIAL which created a
> sequence. The table is unlogged but the sequence is permanent, which
> is probably ok.

> I only have packages down to pg 9.3, so I could not test prior 9.1.
> By looking at the online documentation, is seems that relistemp
> appears in pg 8.4, so the corresponding extraction should be guarded
> by this version. Before that, temporary objects existed but were
> identified indirectly, possibly because they were stored in a
> temporary schema. I suggest not to try to address cases prior 8.4.

Done.

Best,
David.
--
David Fetter <david(at)fetter(dot)org> http://fetter.org/
Phone: +1 415 235 3778

Remember to vote!
Consider donating to Postgres: http://www.postgresql.org/about/donate