build: set pyproject and uv as the dependency managers

.gitignore

@@ -49,4 +49,8 @@ scratch/
 sample_data/
 external_ckpts/
 external/MANO/
+
+# setuptools build output
+build/
+dist/
 docs/

README.md

@@ -40,16 +40,32 @@ git clone [email protected]:GaTech-RL2/EgoVerse.git
 cd EgoVerse
 uv venv emimic --python 3.11
 source emimic/bin/activate
-uv pip install -r requirements.txt
 uv pip install -e .
 uv run pre-commit install
 ```
 
+`pyproject.toml` (locked in `uv.lock`) is the dependency spec for the
+package. Three things live outside it: `requirements.txt` (used only by the
+Eva robot `Dockerfile`), `requirements-ray.txt` (the Ray data-processing
+image) and `scale_sensor_fusion_io`, needed only by the Scale converter under
+`external/scale/scripts` (`uv pip install scale-sensor-fusion-io`).
+
+For the Pi0.5 policy also install the `openpi` fork from the submodule. It is
+installed without its own dependencies (its `lerobot` pin needs `av>=14.2`,
+which conflicts with our `av==12.0.0`), then its `transformers` patch is copied in:
+```
+git submodule update --init external/openpi
+uv pip install --no-deps -e external/openpi
+cp -r external/openpi/src/openpi/models_pytorch/transformers_replace/* emimic/lib/python3.11/site-packages/transformers/
+```
+Re-run the `cp` after any `uv sync` or `transformers` reinstall, which silently
+undoes the patch. See `pi05.md` for details.
+
 ### Conda
 ```
 git clone --recursive [email protected]:GaTech-RL2/EgoVerse.git
 cd EgoVerse
-conda env create -f environment.yaml
+conda create -n emimic python=3.11
 conda activate emimic
 pip install -e .
 pre-commit install

pyproject.toml

@@ -64,10 +64,27 @@ dependencies = [
     "scaleapi",
     "openai",
     "scikit-learn>=1.8.0",
+    "scipy",
 ]
 
+# Not declared: `openpi` (egomimic/algo/pi.py). The fork at external/openpi pins
+# lerobot -> av>=14.2, which conflicts with av==12.0.0 above, so it cannot be a
+# resolvable extra. Install it without its dependencies, as the lab venv does:
+#     uv pip install --no-deps -e external/openpi
+# then apply openpi's `transformers_replace` patch (README, "Installation").
+# `uv sync` reinstalls transformers and silently undoes that patch.
+#
+# Also not declared: `scale_sensor_fusion_io` (PyPI: scale-sensor-fusion-io),
+# imported only by the Scale converter under external/scale/scripts, which is
+# not part of this package; install it by hand when running that converter.
+# The Ray data-processing image keeps its own requirements-ray.txt.
+
 [tool.setuptools.packages.find]
 where = ["."]
+include = ["egomimic*"]
+
+[tool.setuptools.package-data]
+egomimic = ["hydra_configs/**/*.yaml", "resources/*"]
 
 [tool.pytest.ini_options]
 testpaths = ["tests"]

tests/unit/test_packaging.py

@@ -0,0 +1,79 @@
+"""A non-editable install must ship the training stack and nothing else.
+
+Builds the wheel with the configured build backend (no pip / uv needed) and
+inspects its contents: Hydra configs and robot resources must be inside,
+``external/`` and ``tests/`` must not be.
+"""
+
+import shutil
+import subprocess
+import sys
+import zipfile
+from pathlib import Path
+
+import pytest
+
+REPO = Path(__file__).resolve().parents[2]
+
+
+@pytest.fixture(scope="module")
+def wheel_files(tmp_path_factory) -> list[str]:
+    out = tmp_path_factory.mktemp("wheel")
+    # setuptools reuses a stale build/lib (deleted modules would linger in the
+    # wheel); build from a clean slate and leave the checkout as we found it.
+    shutil.rmtree(REPO / "build", ignore_errors=True)
+    code = (
+        "import sys; from setuptools import build_meta; "
+        f"print(build_meta.build_wheel({str(out)!r}))"
+    )
+    proc = subprocess.run(
+        [sys.executable, "-c", code],
+        cwd=REPO,
+        capture_output=True,
+        text=True,
+        timeout=600,
+    )
+    shutil.rmtree(REPO / "build", ignore_errors=True)
+    assert proc.returncode == 0, proc.stderr[-3000:]
+    wheel = next(out.glob("*.whl"))
+    with zipfile.ZipFile(wheel) as zf:
+        return zf.namelist()
+
+
+def test_wheel_does_not_ship_external_or_tests(wheel_files):
+    stray = sorted(
+        {n.split("/")[0] for n in wheel_files}
+        - {"egomimic"}
+        - {
+            n.split("/")[0]
+            for n in wheel_files
+            if n.split("/")[0].endswith(".dist-info")
+        }
+    )
+    assert stray == [], f"top-level entries other than egomimic in the wheel: {stray}"
+
+
+def test_wheel_ships_hydra_configs(wheel_files):
+    packaged = {
+        n[len("egomimic/hydra_configs/") :]
+        for n in wheel_files
+        if n.startswith("egomimic/hydra_configs/") and n.endswith(".yaml")
+    }
+    on_disk = {
+        str(p.relative_to(REPO / "egomimic/hydra_configs"))
+        for p in (REPO / "egomimic/hydra_configs").rglob("*.yaml")
+    }
+    assert "train_zarr_cartesian.yaml" in packaged
+    assert "paths/default.yaml" in packaged
+    assert (
+        on_disk - packaged == set()
+    ), f"hydra yamls missing from wheel: {sorted(on_disk - packaged)}"
+
+
+def test_wheel_ships_robot_resources(wheel_files):
+    # model_x5.xml's meshdir points at egomimic/robot/eva/stanford_repo, which
+    # is not packaged, so the MJCF only loads from a checkout. Shipped anyway:
+    # the URDFs beside it are self-contained and the robot code resolves the
+    # XML relative to egomimic.__file__.
+    assert "egomimic/resources/model_x5.xml" in wheel_files
+    assert "egomimic/resources/model_arx.urdf" in wheel_files