Hi,

On Thu, Nov 19, 2026 at 03:58:54PM -0600, Nathan Bossart wrote:
> Tar-format dumps include an undocumented restore.sql script that can be run
> with psql after extracting the archive and hand-editing its $$PATH$$
> placeholders.  pg_restore never reads it.  This script has needed a number
> of fixes over the years (e.g., commits bf0945e863, ff75219e9f, 3c5d9d914f,
> and 146e03e79a), and it still fails to restore table data for dumps taken
> with --inserts.  Furthermore, it is generated by a second pass over the
> TOC, which forced tar-only hacks into the statistics code added by commits
> 7d5c83b4e9 and 9c02e3a986.  Those hacks run all the attribute statistics
> queries a second time just for this file.
> 
> We could spend more time fixing restore.sql, but by all indications, it
> sees minimal use, and "pg_restore -f" can generate a script from the same
> archive.  IIRC this file was originally added in case users did not have
> access to pg_restore, but that seems like a relatively unlikely situation
> these days.  So, I think we should just remove it, as per the attached.
> 
> Thoughts?

Thanks for the patch!

+1 for removing it. Keeping a second TOC pass and tar specific machinery for an
undocumented script does not seem worth it (especially since pg_restore -f 
provides
the supported way to generate a script).

Having said that, the patch LGTM.

Regards,

-- 
Bertrand Drouvot
PostgreSQL Contributors Team
RDS Open Source Databases
Amazon Web Services: https://aws.amazon.com


Reply via email to