Horton

Re: Compilation issues for HASH_STATISTICS and HASH_DEBUG options

Michael Paquier <michael@paquier.xyz>

View thread View on postgr.es
From: Michael Paquier <michael@paquier.xyz>
To: "Hayato Kuroda (Fujitsu)" <kuroda.hayato@fujitsu.com>
Cc: 'Tom Lane' <tgl@sss.pgh.pa.us>, "'pgsql-hackers@lists.postgresql.org'" <pgsql-hackers@lists.postgresql.org>
Date: 2025-08-15T03:41:53Z
Lists: pgsql-hackers 
On Fri, Aug 15, 2025 at 03:22:18AM +0000, Hayato Kuroda (Fujitsu) wrote:
>> Based on this, I think we should remove the HASH_DEBUG support.
>> It's been broken for six of the last nine years and only one
>> person ever noticed.  Moreover, if you were trying to find a
>> problem in dynahash, you'd probably want different debug logging
>> than what's here.
> 
> I didn't realize that it has been broken for a long time. Agreed to remove
> the HASH_DEBUG option.

One use case where I've been recently relying on dynahash is the
pgstats shmem code.  While I've not used these symbols, I'm pretty
sure that they could prove useful now that I know about them.  Andres,
an opinion perhaps?

> I have never used both options, but one point is that hash_stats() has been
> exported and extensions have called it. Is there a possibility that hidden
> developer is using?
> 
> Anyway, I have no strong opinion. Attached patch does 1) remove HASH_DEBUG and
> 2) fix broken code with HASH_STATISTICS.

It looks like I'm responsible for breaking one of them, at least,
sorry for that.  As far as I can see, the fixes are not complicated,
so I'd rather fix and backpatch things rather than removing both as
both combined can be quite powerful when debugging or improving this
code.  David, were you planning to do so?  I'd vote for keeping these
working.
--
Michael

Commits

  1. Remove HASH_DEBUG output from dynahash.c

  2. Use elog(DEBUG4) for dynahash.c statistics output

  3. Fix invalid format string in HASH_DEBUG code

  4. Fix failing -D HASH_STATISTICS builds

  5. Refactor initial hash lookup in dynahash.c

  6. Remove large fill factor support from dynahash.c.

  7. Remove broken and useless entry-count printing in HASH_DEBUG code.

  8. Partition the freelist for shared dynahash tables.