Re: GIN pageinspect support for entry tree and posting tree

On Wed, 14 Jan 2026 at 15:26, Roman Khapov <rkhapov(at)yandex-team(dot)ru> wrote:
>> Hi! Thank you 🙏 for your effort. 0003 looks good to me
>>
>> --
>> Best regards,
>> Kirill Reshke
>
> Hi! Thanks for the patch!
>
> I'v been reviewing your patch, and noticed there are some code logic
> that handles NULL values, but tests doesn't cover this scenarios.
>
> So, I added simple line at contrib/pageinspect/sql/gin.sql:
>
> INSERT INTO test1 VALUES (1, ARRAY[11, 111], ARRAY['a', 'b', 'c']);
> +INSERT INTO test1 VALUES (1, ARRAY[NULL, 222], ARRAY['d', NULL]);
> CREATE INDEX test1_y_idx ON test1 USING gin (y) WITH (fastupdate = off);
>
> And got unexpected result...
> As far as I understand, we shouldn't get error for the whole
> gin_entrypage_items executions when there are NULL-values columns at index key, but the
> output contains only the next error:
>
> SELECT * FROM gin_entrypage_items(get_raw_page('test1_y_idx', 1), 'test1_y_idx'::regclass);
> --[ RECORD 1 ]--------------
> -itemoffset | 1
> -downlink | (2147483664,1)
> -tids | {"(0,1)"}
> -keys | y=11
> --[ RECORD 2 ]--------------
> -itemoffset | 2
> -downlink | (2147483664,1)
> -tids | {"(0,1)"}
> -keys | y=111
> -
> +ERROR: invalid gin entry page tuple at offset 4
>
> Is the NULL values handle correct?
>

Nice catch!

I agree — the NULL handling seems wrong.
Here's a quick patch to fix it. What do you think?

--
Regards,
Japin Li
ChengDu WenWu Information Technology Co., Ltd.