Install pixi on your machine and pixi install to set up the environment.
Then pixi run jupyter lab or set up a Jupyter kernel to allow VS Code to execute notebooks using the default pixi environmnet (LLMs can tell you how).
On Pippin:
- Just double click
start_all.batand follow the instructions to start the Pyro nameserver, the job server and the worker. - Or go inside
job_serverand double click the.shexecutables to start server/worker individually.
On Pippin or your own machine:
pixi run [nameserver|server|worker]to start each of them individually.pixi run monitorto live-monitor the job queue (-n 20for more entries,-u aliceto filter by user).
You can pass the --mock option to the worker to use mock mode (use mock QICK and Yoko instead of real hardware and gitignored local folders instead of real D drive locations on BF5).
Pending major refactoring, slab code that is actively used in this repo is now vendored inside slab as a module.
pyproject.toml takes care of making sure the python modules inside this repo can be correctly imported without hacking path variables:
If you add "abc" to the "include" list under [tool.setuptools.packages.find], the folder abc will become an importable module (such that you can import abc in your code anywhere).
No import multimode_expts or import .anything or sys.path hacking!
- Move config versions outside repo
- Experiment should initialize using
station.imif available instead of constructing its own im every time (saves time and allows skipping the dependency injection in runners for mock instruments) - Switch jobs.db file read/write to interacting with the job server
- Delete the binary files that don't belong in git. hdf5 data, images etc. Move them to the data/log folder accompanying each experiment.
- The few
slabfiles are simply vendored in their original cursed state. If we straighten up the Experiment class and allow say proper file-reading functionalities, we can actually easily use theanalyzefunctions in each child class to do the analysis instead of collecting everything infitting_display_classesor such. - Deal with accumulated log files and old jobs.db entries
- Migrate to tProcv2 at some point
- Consider using pydantic for complex config dictionaries.
- Sequencer
- Let the worker run jobs from any worktree, not only
main: one child process per job, started from the job's worktree env, with the code version (commit, dirty flag) in the HDF5 file. Findings, proposed design and open decisions:docs/log/2026-11-12_worker-worktrees-and-flat-namespace.md, part 1. - Replace the import-all-classes hack in
experiments/__init__.pywith a lazy name lookup that keepsmeas.SomeExperiment. 15 class names have more than one definition and must be resolved first. Do this before, or with, the worker item above. Same doc, part 2.