On Sun, Jan 11, 2026 at 5:41 PM Chao Li <[email protected]> wrote:
>
> ---
> In origin.h:
>
> +/*
> + * Clear the per-transaction replication origin state.
> + *
> + * replorigin_session_origin is also cleared if clear_origin is set.
> + */
> +static inline void
> +replorigin_xact_clear(bool clear_origin)
> +{
> + replorigin_xact_state.origin_lsn = InvalidXLogRecPtr;
> + replorigin_xact_state.origin_timestamp = 0;
> + if (clear_origin)
> + replorigin_xact_state.origin = InvalidRepOriginId;
> +}
>
> Why does this function need to move to origin.h from origin.c?
>
>
> That’s because, per Ashutosh’s suggestion, I added two static inline helpers
> replorigin_xact_set_origin() and replorigin_xact_set_lsn_timestamp(), and I
> thought replorigin_xact_clear() should stay close with them.
>
> But looks like they don’t have to be inline as they are not on hot paths. So
> I moved them all to origin.c and only extern them.
Thank you for updating the patch.
I'm not even sure that we need to have setter functions like
replorigin_xact_set_{origin,lsn_timestamp} given that
replorigin_xact_state is exposed. While the reset helper function
helps us as it removes duplicated codes and some potential accidents
like wrongly setting -1 as an invalid timestamp etc. these setter
functions don't so much. So I think we can have the patch just
consolidating the separated variables. What do you think?
Regards,
--
Masahiko Sawada
Amazon Web Services: https://aws.amazon.com
