Bit surprised to see a PR since this has been discussed at various points over the last couple years.

I agree that output is too verbose, but I don't think this implementation is best exactly for the reasons mentioned: shared deps. What's the point of printing python and its dependencies 10 times if you have 10 python packages as roots.

To me it makes more sense to use traverse.traverse_tree(env.concrete_roots(), cover="nodes", depth_first=False) so every node is printed only once.

That's pretty trivial, except you need to have some callback to print section headers for abstract roots.

What's the point of printing python and its dependencies 10 times if you have 10 python packages as roots.

I can handle shared deps as well, although:

• I think if someone has an env with 10-20 py-* packages that they might not want to list them at all in the concretize output, at least one line at a time.
• For x->z, y->z, ! x->y, I wonder if it would make sense to show z but not dependencies of z, or otherwise indicate that y is truncated

@tgamblin ping since you mentioned an interest

That's pretty trivial, except you need to [...]

... and account for bugs ;)

Something like:

tree = traverse.traverse_tree(env.concrete_roots(), depth_first=False, key=traverse.by_dag_hash)
for (depth, edge) in tree:
  if depth == 0:
    # format corresponding abstract spec
  # format edge.spec

where key=traverse.by_dag_hash is sometimes needed (when specs come straight out of the concretizer they are separate instances)

@scheibelp take a look at #44489