Skip to content

Prevent erroneous HashCollisions by using the same ctime for all rows.

vlorentz requested to merge generated-differential-D2977-source into generated-differential-D2977-target

'swh_content_add' tries to avoid this issue with a DISTINCT clause on the entire row; but it is useless because 'ctime' cells differ by a few microseconds. This commit ensures all ctime values are exactly the same, so they are filtered out.

An alternative would be to change 'swh_content_add' to do:

select distinct on (sha1, sha1_git, sha256, blake2s256, length, status) sha1, sha1_git, sha256, blake2s256, length, status, ctime from tmp_content

instead of:

select distinct sha1, sha1_git, sha256, blake2s256, length, status, ctime from tmp_content

but this is more verbose and there's no good reason to call 'now()' for every row.

Related to T2332

Migrated from D2977 (view on Phabricator)

Merge request reports