Thanks @olasd for this piece of information.

Fix tests

Few additions to the original diff

stupid arc tool

fix typo reported by ardumont

• Commit missing stuff.
• Fix commit again.
• Fix requirements.txt
• Add missing tests.

Please give a bit of context in your commit messages (in git). Like these 2 commits are dealing with tests for example. And 'More keys' is not a good commit message. Like 'fix a bug' is not a valid commit message, generally speaking.

A bit 'how you doing' there, but 1/ it's kind of consistent with other similar modules in other swh packages, and 2/ who cares?

LGTM, but hey, these are puppets!

Fix indentation in swh-vault/entrypoint.sh

Actually not needed

rebased

Important rework of the config handling

I'm (re)working this diff, so wait a bit before spending time reviewing it.

typos

Let's keep on, but I really don't like the turn this code takes...

lgtm

rebased and fixes reported by aardumont and vlorentz

Add missing hunks in tox.ini and requirement-test.txt

I'm a bit puzzled that with so many comments, this diff has been accepted! It should not have been so IMHO.

In D1090#23117, @vlorentz wrote:

In D1090#23077, @douardda wrote:

One drawback of this function extraction is that you now have to keep an intermediate result (the origins list). Couldn't this be avoided using generators?

That's a very small list (I'd be surprised if it ever gets over 10KB). And starting with D1092, we'll download the whole list at once anyway.

In D1080#23137, @vlorentz wrote:

(Actually, after thinking more about this, a class with setUp and tearDown looks like better than my decorator solution.)

Also give a look at Dramatiq https://dramatiq.io/

In D1092#23086, @douardda wrote:

I see no link between this hunk and the diff's title. Did I miss something?
BTW, I do prefer the version before the diff: easier to read.

ask rework just for the sake of the fixture question

LGTM

I see no link between this hunk and the diff's title. Did I miss something?
BTW, I do prefer the version before the diff: easier to read.

Could you please explain in the commit message why you do this diff? Also the 'assert False' looks really strange. What is its practical purpose?

One drawback of this function extraction is that you now have to keep an intermediate result (the origins list). Couldn't this be avoided using generators?

Why not make this decorator a pytest fixture?

This lacks a unit test: just add a test for the new method. It's nice to also have the test for the revision_metadata_add() call, but there should be a test that ensure the _check_duplicates() behaves ok (once again, with valid and invalid inputs), and ensure the newly added validator works in every modified method. I count 7 of them patched (plus the hunk in in_memory.py). I would expect (functionally) 8 tests.

Also, you add many defensive statements, but none of them are actually tested, unless I'm mistaken. Once again, please add tests for these invalid input resiliency checks.

As always with this kind of diff, I'm quite puzzled: we add a lot of 'if' conditions (on types), but what are the unexpected types that the code may encounter? As is, we just (silently) ignore a bunch of values.

Kill (useless) --create-tables and --with-data cli command options