The releases generated by swh-loader-tar aren't referenced by any occurrences: we generate occurrences pointing at revision objects, but not at releases.
Actions:
After discussion, we will stop generating dangling releases and keep the occurrences pointing at revisions.
Yep. Of course we will also need to remove all dangling releases that have been generated thus far.
Of course we will also need to remove all dangling releases that have been generated thus far.
Indeed
Actions undertook to clean dangling releases:
$ select id from person where name='Software Heritage' and fullname='Software Heritage <robot@softwareheritage.org>' and email='robot@softwareheritage.org'; > 3661419
$ select * from origin where type='ftp'; id | type | url | lister | project ---------+------+-------------------------------------------------------------------------+--------+--------- 4423668 | ftp | rsync://ftp.gnu.org/gnu/3dldf | | 4423671 | ftp | rsync://ftp.gnu.org/gnu/3dldf | | 4423974 | ftp | rsync://ftp.gnu.org/gnu/GNUinfo/Audio/francais | | ...
Checking manually those are indeed only gnu (there aren't that much).
All dangling releases that were created targets the same revision as the occurrence that were created at the same time:
$ select count(r.id) from occurrence occ inner join origin ori on occ.origin=ori.id inner join release r on (r.target=occ.target and r.target_type='revision') where ori.type='ftp' and r.author=3661419; count ------- 8911
$ delete from release where id in (select r.id from occurrence occ inner join origin ori on occ.origin=ori.id inner join release r on (r.target=occ.target and r.target_type='revision') where ori.type='ftp' and r.author=3661419);softwareheritage=> select count(*) from release r where r.author=3661419 and r.synthetic;
count
-------
0
(1 row)