Re: BUG #18947: TRAP: failed Assert("len_to_wrt >= 0") in pg_stat_statements

On 2025/06/11 7:39, Michael Paquier wrote:
> On Tue, Jun 10, 2025 at 08:14:40PM +0900, Michael Paquier wrote:
>> Hmm. I am wondering whether this would be enough; these are not the
>> only places where we use select_no_parens and select_with_parens. I'd
>> be curious to see what you think a fully-bloomed fix should look like.
>
> By the way, this report also means that none of the problematic query
> patterns are run in the main regression test suite, or we would have
> noticed something immediately when running the recovery test
> 027_stream_regress.pl that loads PGSS.
>
> So we have a problem in test coverage with query normalization
> patterns. But we also have a second, older, problem in terms of
> coverage regarding the parentheses patterns that are assumed as
> supported in the parser. Even if we finish by reverting the nested
> query tracking support for now (let's settle things before beta2 not
> after it), we should definitely improve the main regression test suite
> in this area to make sure that reintroducing it safe.

Sounds reasonable to me.

Also I agree with reverting the commit and revisiting the feature in v19.

Regards,

--
Fujii Masao
NTT DATA Japan Corporation