Rebase

Update from review comments

Fix docstring

Remove dead code

Add special case for revisions, compress with ZST

database -> database to create

Expand command description

Rebase

Since D6676 the specialized methods of swh/graph/backend.py have all been removed and replaced by a generic proxy layer that calls all the methods in a completely transparent fashion, so this specific issue appears to be obsolete now.

Right, this is what T3855 will cover, but there is also a dependency on T1971 to figure out where this will go exactly (package summary vs sphinx)

I filed T3855 to document the architecture of the java code of swh-graph so that it stays on our radar.

I agree in general, but I disagree for the specifics of this diff. This is a description of the design of a specific component, which could be copied in an architecture presentation if it makes sense, but is certainly not an architecture presentation in itself.

SwhBidirectionalGraph: add UML diagram of class hierarchy

@vlorentz this is all written in the docstrings of the classes, which will be added to swh-docs once https://forge.softwareheritage.org/T1971 is implemented (which is going to be easier to do thanks to this diff :-))

Fix more buggy Graph -> SwhBidirectionalGraph

Remove Graph -> SwhBidirectionalGraph strings

Here, I changed it so that all the paths go to the correct node.

Make random_walk tests deterministic

Rebase

Minor comment to avoid doing url parsing manually, otherwise LGTM

I have just one security consideration, otherwise LGTM. Thanks!

The coverage report does not seem to agree though ;)

No, we want to check that random_walk can reach its actual destination.

I made a temporary fix in D6893, it doesn't solve the underlying issue but greatly decreases the probability of it happening. I'm not quite sure what would be a proper test for this endpoint, but this is at least enough to fix this issue in particular.

Yes, this is implicitly all tested under the compression pipeline test.

Add simplify step, fix various review comments

why the need for SortOutputHandler attributes to become final?

Fix src/dst inversion, add regression test