marin-community/marin#1690 - 2-way sync via github-pr.py (ryan-williams/git-helpers)

marin#1690.md

marin-community/marin#1690 "Workspace" step 1: experiments → marin

First step in the "uv workspace migration plan" (#1773):

• Initialize a uv workspace
• Move the marin package into lib/marin/
• experiments/ remain in "root" target, which depends on marin
• Next step: #1723 (Levanter integration)

flowchart LR
    experiments[experiments]
    marin[lib/marin]
    data_browser

    experiments --> marin

Loading

(data_browser stays independent, not a workspace member)

Changes

• Create workspace root pyproject.toml
• Move src/ → lib/marin/src/
• Update path references in workflows, Makefile, mkdocs config
• Update documentation GitHub URLs to use MARIN_DOC_BRANCH env var (defaults to main)

Structure After

marin/
  pyproject.toml        # Workspace root
  experiments/          # Stays at root, imports from lib/marin
  lib/
    marin/              # Workspace member
      pyproject.toml
      src/marin/

Migration Scripts

See workspace-migration scripts for:

• step1/main.py - Main migration script
• step1/test.sh - Test harness
• Full documentation

z3404539962-dlwh.md

🚀 🚀

z3409035636-percyliang.md

marin/lib/marin/src/marin is a bit of a mouthful - I suppose there's not an easy way to avoid this? Does marin and data_browser have to be in lib or could they be at the top-level?

z3409146290-ryan-williams.md

marin/lib/marin/src/marin is a bit of a mouthful

Agreed, though from the repo root it's "just" lib/marin/src/marin.

I suppose there's not an easy way to avoid this? Does marin and data_browser have to be in lib or could they be at the top-level?

• marin/src/marin is possible (again, relative to repo root). We could have all member root dirs be top-level in repo (instead of under lib/)
• Per latest on #1773, I'm thinking leaving data_browser out of "workspace" machinery (i.e. not a "member") is the better approach.