the original_artifact JSON properties in revision.metadata for all imported tar/zip files look this:
{ "original_artifact" : [ { "name" : "dejagnu.texi.tar.gz", "sha1" : "9a5380aa7c2a9fb7f84036b223cbde07d9db7e67", "sha256" : "68f5a356bae22bfef8d09c6314d2c508016507b197e298bcb210a2cdf674cb93", "sha1_git" : "26a738cc74f236e7e2e1ee1a00817dfe5716aa2d", "archive_type" : "tar" } ] }
A length field is missing and we should add it, both to the tarball loader and to the already imported data.
Rationale: the lack of a length field is inconsistent both with the layout of the content table and with what the git loader does.
Note that the expected JSON Schema is now available in swh-storage under sql/json/.