refactor diff to defer loading content data after creating content object.

apply comment

Add missing hunk

4c070f98ca09255a54fb3fc9efbcf9d6e099ac96

e10924140b409e77edc2d7501dbcfea4b46329b5

60c3aa16cf8778c7413606e2a532cfc47966d63c

Looks good to me, thanks!

use git order

In D2703#64548, @olasd wrote:

We already have lots of issues with loaders having an "optimistic" usage of memory, so I'd be tempted to introduce a DiskBackedContent (DiskContent?) model, inheriting from Content, with a path attribute as well as a lazy data attribute that will read data from disk the first time it's accessed.

What's the difference?

The difference between what and what?

add doc for max_content_length.

In D2703#64521, @olasd wrote:

That'd be a lot more work downstream, but wouldn't it make sense to have get_data return model objects directly instead of dicts?

By default, from_disk will not pull the full data from disk into memory, so by doing this there's a good chance you'll end up with a Content object that has an empty data attribute.

reword comment

That explains the inconsistent behavior I noticed in tenacity... thanks!

I see you did not annotate some of the functions. Do you need help with those?

5f4452e0d71cf0f1491df115915b9854890a5cb1

update commit message to mention swh-scheduler

rebase

But then, errors would pass silently...

Shouldn't the new test be in a new tests module?

update doctest

In D2685#64238, @olasd wrote:

I guess we should have the RPC client/server pair re-raise ValueErrors? @vlorentz?

rebase

update copyrights

update copyright

rebase

rebase

In D2684#64107, @anlambert wrote:

Abandoning this and let @vlorentz properly fixing the issue

change base commit