On 2024-04-07 06:33 +0200, Tom Lane wrote:
> Erik Wienhold <ewie@ewie.name> writes:
> > I'm surprised that the comment is not skipped by the scanner at this
> > point. Maybe because the parser just reads the raw expression between
> > WHEN and THEN with plpgsql_append_source_text via read_sql_construct.
>
> > How about the attached patch? It's a workaround by simply adding a line
> > feed character between the raw expression and the closing parenthesis.
>
> I don't have time to look into this on this deadline weekend,
Sure, no rush.
> but what's bothering me about this report is the worry that we've made
> the same mistake elsewhere, or will do so in future.
Right. At the moment only make_case is affected by this because it uses
the raw expression for rewriting. I checked other uses of
read_psql_construct (e.g. IF ... THEN, FOR ... LOOP) and they don't show
this bug.
> I suspect it'd be much more robust if we could remove the comment from
> the expr->query string. No idea how hard that is.
I slept on it and I think this can be fixed by tracking the end of the
last token before THEN and use that instead of yylloc in the call to
plpgsql_append_source_text. We already already track the token length
in plpgsql_yyleng but don't make it available outside pl_scanner.c yet.
Attached v2 tries to do that. But it breaks other test cases, probably
because the calculation of endlocation is off. I'm missing something
here.
--
Erik