-R collects stages from all of the repo

[skshetry]

Bug Report

We are building repo.graph and using path to search for stages in the graph when -R instead of just reading files from the path directory.

Context

https://groups.google.com/a/iterative.ai/g/support/c/H_c36GuAsPM/m/6mLNdPIRAgAJ

[efiop]

This is by-design though, we are being paranoid in every operation. But it does make sense to revisit this in general.

[skshetry]

Had a conversation with a user (~100k .dvcfiles, 525s to load), where this would have been really helpful.

[dberenbaum]

Bumping this priority down for now.

[dberenbaum]

A customer mentioned a similar issue where they are specifying a target in dvc exp run pipeline_one/dvc.yaml but DVC still collects stages from pipeline_two/dvc.yaml.

[skshetry]

That is expected, as dvc has to build a graph. dvc repro dvc.yaml means reproduce all the dependencies of the stage in dvc.yaml and the stages in dvc.yaml.

That is unrelated to this issue though, which is about --recursive (eg when using dvc checkout --recursive etc).