Обсуждение: Fix inconsistencies with code and beautify xlog structures description and fin hash_xlog.h

Hi hackers!

While doing some xlog-related task at my job I noticed inconsistency
between [0] & [1]. hash_xlog.h xl_hash_squeeze_page structure
description mismatches with actual code.
PFA fixing issue.

I also allowed myself to standardise description of
xl_hash_move_page_contents and xl_hash_init_bitmap_page, so
xl_hash_delete, xl_hash_init_bitmap_page and
xl_hash_move_page_contents describes their zeroth backup block akin to
each other.


[0]
https://git.postgresql.org/cgit/postgresql.git/tree/src/include/access/hash_xlog.h?id=567d27e8e2b752743626eb259ba75ecdc936eaf3#n147
[1]
https://git.postgresql.org/cgit/postgresql.git/tree/src/backend/access/hash/hash_xlog.c?id=567d27e8e2b752743626eb259ba75ecdc936eaf3#n830

-- 
Best regards,
Kirill Reshke

On Fri, 5 Sept 2025 at 11:17, I wrote:

> I also allowed myself to standardise description of
> xl_hash_move_page_contents and xl_hash_init_bitmap_page, so
> xl_hash_delete, xl_hash_init_bitmap_page and
> xl_hash_move_page_contents describes their zeroth backup block akin to
> each other.

Ahh... sorry for the noise, this part is wrong.
PFA v2


-- 
Best regards,
Kirill Reshke

> On 5 Sep 2025, at 11:24, Kirill Reshke <reshkekirill@gmail.com> wrote:
>
> PFA v2

Proposed change is correct and in line with a description of e.g. XLOG_HASH_MOVE_PAGE_CONTENTS.
I'd note that block is for lock only and record does not include image, but other such cases are not reported in
comments.


Best regards, Andrey Borodin.

On Wed, Sep 10, 2025 at 05:36:48PM +0500, Andrey Borodin wrote:
> Proposed change is correct and in line with a description of
> e.g. XLOG_HASH_MOVE_PAGE_CONTENTS.
> I'd note that block is for lock only and record does not include
> image, but other such cases are not reported in comments.

Right, the case of block 0 being registered in a MOVE_PAGE_CONTENTS
record to ensure that a cleanup lock is taken is documented in
_hash_squeezebucket()@hashovfl.c, for SQUEEZE_PAGE, or for DELETE.
Squeeze can also do that for buffer 1.

On top of your suggestions, I have double-checked the rest of the hash
records, and they seem to be in line with hash_xlog.h.  Will fix,
thanks.
--
Michael