Re: [BUG] temporary file usage report with extended protocol and unnamed portals

On 4/20/25 00:42, Sami Imseih wrote:
>> (In my testing, the "temporary file:" message comes
>> out without any attached STATEMENT most of the time already, so this
>> isn't losing much as far as that's concerned.)

Indeed, this happens when using autocommit / implicit transactions.

But if you disable autocommit, and I think this is the most common
setting, we do have the attached statement.

> moreover, as I was looking into why the case mentioned earlier
>
> ```
> SELECT FROM foo ORDER BY a \bind
> ;
> SELECT 1 \bind
> ;
> ```
>
> does not show STATEMENT after the temp file logging, I realized
> it's because the temp files are cleaned up and reported at
> the end of transaction, which means that debug_query_string is NULL at the
> time the portal is dropped in the next query. This causes
> check_log_of_query to return false.
>
> /* query string available? */
> if (debug_query_string == NULL)
> return false;
>

Yes, we actually have two related bugs.

>> Perhaps a cleaner answer is to rearrange things in postgres.c
>> so that if there's a pre-existing unnamed portal, we drop that
>> before we ever set debug_query_string and friends at all.

> That did cross my mind as well, but I was trying to avoid doing this
> type of rearranging. I still rather not go down that path considering the
> case mentioned above will still not display the query text in a STATEMENT log.

IMHO, it would be nice to fix the explicit transactions case.

The bug with implicit transactions is less serious.

I can try to implement Tom's idea if we have a consensus.

Best regards,
Frédéric