From the review of #106 and #108.
#106 fixed the one json.dump(f(), open(p, "w")) where evaluation order was load-bearing. The remaining writes in build_subsets.py (every edge and cell file, plus subsets_summary.json) and the census still use json.dump(doc, open(...)) without closing the handle. There is no truncation risk there because the document is built first, but every run emits ResourceWarnings, and an interrupted write can still leave a partial file in assets/fleet/ that assemble_page.py --check will not notice.
#106 explicitly deferred the stronger fix: write to a temporary file and rename, across the four scripts that write derived data. This records that deferral so it is not lost.
From the review of #106 and #108.
#106 fixed the one
json.dump(f(), open(p, "w"))where evaluation order was load-bearing. The remaining writes inbuild_subsets.py(every edge and cell file, plussubsets_summary.json) and the census still usejson.dump(doc, open(...))without closing the handle. There is no truncation risk there because the document is built first, but every run emitsResourceWarnings, and an interrupted write can still leave a partial file inassets/fleet/thatassemble_page.py --checkwill not notice.#106 explicitly deferred the stronger fix: write to a temporary file and rename, across the four scripts that write derived data. This records that deferral so it is not lost.