On 20/10/2025 04:36, Michael Paquier wrote:
> On Wed, Oct 15, 2025 at 05:39:38PM +0300, Heikki Linnakangas wrote:
>> We could store xl_crc unaligned. IIRC all the structs that follow that are
>> already stored unaligned.
>
> If we do just that, would there be any downside in moving xl_crc to be
> just after xl_prev? That would keep the record assembling part
> simpler, because XLogRecord is treated as a single object, where the
> code relies on a MAXALIGN64() to make sure that the header is aligned.
The CRC calculation looks like this:
> /*
> * Now that xl_prev has been filled in, calculate CRC of the record
> * header.
> */
> rdata_crc = rechdr->xl_crc;
> COMP_CRC32C(rdata_crc, rechdr, offsetof(XLogRecord, xl_crc));
> FIN_CRC32C(rdata_crc);
> rechdr->xl_crc = rdata_crc;
If 'xl_crc' is not the last field, the above needs to be changed to skip
'xl_crc', or fill it with zeros or something. Similarly in the code that
verifies the checksum. But sure, we could do that.
- Heikki
