[sequoiap]

Draft: depends on a PyroLab release. This needs a PyroLab release that includes BYUCamachoLab/pyrolab#96 (PyroLab's own move to pydantic 2). Until that release exists, pyproject.toml pulls pyrolab from git through [tool.uv.sources]. That only affects development installs, not what PyPI users get. Before releasing AutoGator, delete that section and change the pyrolab[rto] dependency to require the new release.

Why

gdsfactory<7 held AutoGator on pydantic 1, and PyroLab is moving to pydantic 2. Newer gdsfactory doesn't help: 7.x requires pydantic 2 but pins an old gdstk, and 8+ is built on KLayout. AutoGator only used gdsfactory to read a GDS file's top cell and its polygons. gdstk, which gdsfactory 6 used underneath, does that directly.

Changes

• GDS loading (circuits.py): uses gdstk.read_gds(...).top_level() instead of gf.import_gds. If a file has more than one top-level cell, it now raises an error that names them. gdsfactory is no longer a dependency, so 17 fewer packages get installed.
• pydantic 2:
  • model_dump_json / model_validate_json replace json / parse_file, so existing profile JSON files load unchanged.
  • ConfigDict(extra="forbid").
  • Stage axes are explicitly Optional[...] = None.
• Keyboard bindings (controllers.py) use pydantic-settings and read environment variables with the prefixes AUTOGATOR_KEYBOARD_ and AUTOGATOR_GUI_.
  • This fixes a bug: with no prefix, the HOME binding picked up the shell's HOME variable (e.g. /home/user).
  • ⚠️ Anyone who overrode a binding with an unprefixed environment variable must add the prefix.
• Python 3.11+, matching PyroLab.

Verification

AutoGator has no test CI and only stub tests, so I checked these by hand:

• Every module imports on Python 3.11 and 3.13.
• A profile JSON written in the old format loads and round-trips, and unknown keys are rejected.
• The environment prefixes work, and HOME no longer leaks into the bindings.
• I compared loadGDS on master (gdsfactory 6, pydantic 1) with this branch on the same synthetic chips. The polygon counts and polygon hashes matched, and so did the outcome.
  • My synthetic layouts don't contain any recognizable circuits, so both versions failed the same way.
  • Please also check it on a real chip GDS before merging.

🤖 Generated with Claude Code