This also kills test_write_replay.py file since it does not test anything
more than what is currently tested in test_replay.py.

Or you could just use a random value every time.

The proposal is good imo. Maybe add milliseconds just to be sure though, in case the client is scripting multiple uploads at the same time

I think you should change check_archive to detect Django's name changes (see D3192)

No test?

I wonder if there should be a test checking the snapshot isn't referenced if Storage.snapshot_add errors

this should be two commits, though

"ok then"

In D3191#77774, @ardumont wrote:

What will be the output of that pre-commit check (when it fails)?

In D3191#77776, @olasd wrote:

Good idea, but unfortunately I think this will complain when running pre-commit -a, if a file hasn't seen an update recently. This would need to check whether the last year matches the last commit in git log for that file.

improve comments

It should be easy to fix

fix typo

Upping priority, as we are hitting this issue in production

LGTM, but I think it's missing some tests:

fix migration, it was dropping a function that is still in use.

rebase

• fix typo
• rebase

fix mismatch

We wouldn't need that if pytest had an equivalent for assertCountEqual :/

What is the motivation for this?

In D3171#77196, @douardda wrote:

Maybe, but that would require we keep the info "this is an anonymized object" somewhere, which is not the case for now. This idea can be dealt later, maybe?

fix typo + use SWHID as first argument

Shouldn't we make anonymized objects error when their compute_hash() method is called?

rebase

fixup of the rebase: alter index metadata_provider_type_url to change its name, instead of creating metadata_authority_type_url which is a duplicate.

Make 'visit' depend on 'origin' being provided.

apply @ardumont's changes to the migration.

I applied most of your comments, except:

apply most comments

undo push on the wrong diff (for real this time)

undo push on the wrong diff

big rebase

big rebase

for the context, we need to use the SHWIDs themselves, not the sha1_git that is bound to version 1 of SWHIDS

I don't think it makes sense to make this difference, because if a metadata only has origin in its context, it means it's context-free on all "axis" but the origin already. Having also no origin context is just one more axis it would be context-free on.

Hmm actually, "artifact" isn't great either, because we'll want to use it to mean extrinsic metadata , which aren't artifacts. Ugh...

In D3152#76595, @douardda wrote:

In D3152#76561, @vlorentz wrote:

And I'm not a big fan of the name. What about object_type, as we already use this name in other places?

yes I also agree, but not a big fan of object_type either. However I have no better candidate to propose, I'm afraid. entity_type maybe ?

Great! You know how little I like isinstance ;)