On May 29, 11:48 am, Kevin Field <kevinjamesfi...@gmail.com> wrote:
> On May 29, 11:35 am, robertmh...@gmail.com (Robert Haas) wrote:
>
>
>
> > On Fri, May 29, 2009 at 7:59 AM, Kevin Field <kevinjamesfi...@gmail.com> wrote:
> > > On May 28, 5:19 pm, da...@kineticode.com ("David E. Wheeler") wrote:
> > >> On May 28, 2009, at 12:53 PM, Kevin Field wrote:
>
> > >> >> Can pgTap check for a regex instead if just a string?
>
> > >> > That's the other option, if the pgTAP author is willing...if the
> > >> > SQLSTATE thing doesn't work out I guess we'll have to go down that
> > >> > road.
>
> > >> Patches welcome. ;-)
>
> > >> http://github.com/theory/pgtap/tree/master/
>
> > >> I'm getting a new version ready to release as I type.
>
> > > Thanks, great to know. :) Although, I do think changing plperl is
> > > the more proper option, so I'm going to try there first...
>
> > It seems to me that removing line numbers from PL/perl error messages
> > is not a good solution to anything. Line numbers are extremely useful
> > for debugging purposes, and getting rid of them because one particular
> > testing framework doesn't know how to use regular expressions is
> > solving the wrong problem.
>
> You're right, but that's not what I'm proposing...
>
> > I'm also a bit confused because your original post had a line number
> > in the PL/pgsql output, too, just formatted slightly differently. Why
> > doesn't that one cause a problem?
>
> The difference is, in PL/pgsql they're in the CONTEXT: line, whereas
> in plperl they're in the error line. This is inconsistent; if we fix
> it, we don't need to add kludge to pgTAP.
>
> But later in the thread the desired fix became not changing perl but
> instead making a way to report error codes from plperl, which is what
> I'm attempting to do with my rusty C skills soon. plperl should have
> ereport() *anyway*, as I believe Tom had insinuated.
BTW, I noticed in exec_stmt_raise() in src/pl/plpgsql/src/pl_exec.c
that the comment still says "throw it with elog()" rather than "ereport
()" even though ereport() is used in all places but one in the
function:
default: elog(ERROR, "unrecognized raise option: %d", opt->opt_type);
Should this be changed to:
default: ereport(ERROR, (errmsg_internal("unrecognized raise option: %d",
opt->opt_type)));
...along with the comment?
