Re: proposal: possibility to read dumped table's name from file

Hi

pá 4. 11. 2022 v 14:28 odesílatel Julien Rouhaud <rjuju123(at)gmail(dot)com>
napsal:

> On Fri, Nov 04, 2022 at 07:19:27AM -0500, Justin Pryzby wrote:
> > On Fri, Nov 04, 2022 at 07:59:01PM +0800, Julien Rouhaud wrote:
> > > On Thu, Nov 03, 2022 at 10:22:15PM +0100, Pavel Stehule wrote:
> > > > čt 3. 11. 2022 v 5:09 odesílatel Julien Rouhaud <rjuju123(at)gmail(dot)com>
> napsal:
> > > > updated patch attached
> > > >
> > > > big thanks for these comments and tips
> > >
> > > Thanks for the updated patch! As far as I'm concerned the patch is in
> a good
> > > shape, passes the CI and I don't have anything more to say so I'm
> marking it as
> > > Ready for Committer!
> >
> > +1
> >
> > I started looking to see if it's possible to simplify the patch at all,
> > but nothing to show yet.
> >
> > But one thing I noticed is that "optarg" looks wrong here:
> >
> > simple_string_list_append(&opts->triggerNames, optarg);
>
> Ah indeed, good catch! Maybe there should be an explicit test for every
> (include|exclude) / objtype combination? It would be a bit verbose (and
> possibly hard to maintain).
>

yes - pg_restore is not well covered by tests, fixed

I found another issue. The pg_restore requires a full signature of the
function and it is pretty sensitive on white spaces (pg_restore). I made a
mistake when I partially parsed patterns like SQL identifiers. It can work
for simple cases, but when I parse the function's signature it stops
working. So I rewrote the parsing pattern part. Now, I just read an input
string and I try to reduce spaces. Still multiline identifiers are
supported. Against the previous method of pattern parsing, I needed to
change just one regress test - now I am not able to detect garbage after
pattern :-/. It is possible to enter types like "double precision" or
"timestamp with time zone", without needing to check it on the server side.

When I wroted regress tests I found some issues of pg_restore filtering
options (not related to this patch)

* function's filtering doesn't support schema - when the name of function
is specified with schema, then the function is not found

* the function has to be specified with an argument type list - the
separator has to be exactly ", " string. Without space or with one space
more, the filtering doesn't work (new implementation of pattern parsing
reduces white spaces sensitivity). This is not a bug, but it is not well
documented.

* the trigger filtering is probably broken (on pg_restore side). The name
should be entered in form "tablename triggername"

attached updated patch

Regards

Pavel