Remove what nothing uses, and keep one copy of the sample data

[adulbrich]

Housekeeping after the 5.0.0 release. Two commits: removals, then the example/ consolidation.

Removed, each verified unreferenced

path	what it was
target/	rust-analyzer's flycheck scratch, committed once. Carried another machine's absolute paths and a v3.1.0 build's warnings. Now in .gitignore.
public/SunAperture*.png	replaced by the lucide Aperture glyph; nothing has referenced them since
.gitmodules	declared submodules/radiance_pipeline, a directory that does not exist
binary_paths.json	four empty tool paths, from before the tools were compiled into the app
extra/	the lab's original ldr-to-hdr.sh and an R script, wired to nothing
e2e-tests/test/inputs/CR2/output.hdr	a 66 MB pipeline output, committed inside the inputs directory

stages.ts cited extra/ldr-to-hdr.sh:197 as the provenance of pcomb -h. That is load-bearing rather than trivia — it is the evidence the flag was inherited rather than chosen, which is why #241 nearly removed it and why removing it broke the pipeline. The comment now says where the script was and that the history has it.

One copy of the sample data

There were two sets of calibration files: example/, which the README sends users to, and e2e-tests/test/inputs/, which the suites run. They had drifted, and not harmlessly.

example/fisheye_corr.cal fills the region outside the lens circle with 1; the fixture copy fills it with 0. Both carry the same comment:

We also clear the region outside the r=[0,.5] circle to black.

The copy users were pointed at contradicted its own documentation, and nothing exercised it, so nothing caught it. The fixture copy wins.

The other three example/ files were redundant under vaguer names — response_function.rsp, vignetting.cal and calibration_factor.cal are byte-identical to Response_function.rsp, vignetting_f5d6.cal and CF_f5d6.cal, whose names at least say which aperture they belong to.

So the fixtures move to example/, and everything reads from there: both end-to-end suites, the hdrgen benchmark, and the README. Sample data the tests exercise is sample data that stays correct.

The move costs no history — git stores content, so a rename reuses the same blobs.

Verification

435 unit tests, all three projects typecheck, lint clean, the benchmark's fixture resolution, and the full 18-frame browser pipeline end to end.

Note on ordering

A history rewrite is planned to purge ~800 MB of dead blobs (dist/ from the PyInstaller era, tmp/output*.hdr, the vendored hdrgen binaries, a committed installer). This should merge first, or it will need recreating against the rewritten history.

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
lumilab	Ready	Preview	Aug 7, 2026 11:02am