I am not sure we want to keep that dateutil.parser.parse. Its function input
is a plain string (quite large) and its output is sometimes unexpected [1]

We already discussed in other modules that we should converge on using iso8601
dates (well use that module at least). So might be, it'd be best to use more
conventional timestamp formats, iso8601 sounds like one sensible choice for at
least one format.

I imagine we have that behavior already so that might be another task entirely.
I just mention it here since we are reworking endpoints.

Wouldn't this be a good time to also move the URL to a query parameter? (Because of the double-slash issue that can happen on clients)

Update: Doc fix

Update: Fix test syntax in django template (no line break allowed, sigh ...)

Update: Remove redundant error message in template and prevent django.template.base.VariableDoesNotExist exception

Ack, I will land the diff and push a new swh-web tag.

I should be able to work something out for that one.

we are missing the test case about revision ok

I added an extra check for the revision if the directory is None in order to improve error message.

Update: Improve error message for missing revision

D3042 handles the branch with non resolvable revision case.

I have updated D3041 to restore previous behavior and look what I got now:

Update: Do not filtered out branches with a non resolvable revision (restore previous behavior)

Apparently, the changes I introduced in D3041 makes the above issue disappear ... I need to dig further on this.

Apparently, the changes I introduced in D3041 makes the above issue disappear ... I need to dig further on this.

In D3041#73696, @ardumont wrote:

I think it's fine :D

It's "slightly" big ;)

I see where the issue is, diff incoming ;-)

Rebase

Update: Use simpler hardcoded URL in test

Update: Add test

Rebase

It seems Phabricator reopened the task automatically with my last comment, that was not intended.

@moranegg , for the branch case the anchor will be the revision it points to. For your example, it will be

• add assert mock*.*.called checks

I got the exact same situation when I updated the mercurial loader to swh-model objects.

Update: Upgrade mathjax to 3.0.5 (version was released after I pushed the diff)

Yes sure, but I meant to self-host it instead of using a CDN

Is it really worth it?

I am not sure if explicitly setting the key type when instantiating a SortedList instance is a real necessity.
How about simplifying a little bit in order to simply write sha1s = SortedList[bytes] ?