It is easier to keep track of it in Grafana.

RunAll is not practical, because we need to mount the graph on a tmpfs
at some point. Let's keep that step manual for now, which means
RunExportCompressUpload and DeanonymizeOriginContributors will be
called separately.

I cannot find how to make WebGraph log to stderr instead of stdout,
so it seems to be the only way.

It ensures luigi will be available after having installed swh
development environment.

It can be cumbersome to set paths for all (recursives) dependencies of
the task we want to run; this CLI endpoint takes care of most of them.

It is going to get large, with the future addition of tasks to generate
the license dataset and the citation dataset.

This triggers a bug in ListOriginContributors, causing it to include
"null" as a contributor.
A future commit will fix this.

This Java script (and related Luigi tasks) traverse the graph in
topological order, building up the set of all contributors to a
node and its ancestors, then dump the value of this set for every
origin node they encounter.

This allows readers to efficiently get ancestors of nodes with low indegree
(ie. most revisions), as it avoids a random access / API call.

It's hard to handle correctly, and it does not seem useful to use
outside the compression pipeline.

In practice, we use separate directories for this.