You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Version srcset projection semantics for reproducible receipts #60
The html_attribute_urls projection format needs an explicit parser-semantics version so receipts remain reproducible when srcset parsing changes. A parser correction can change the resulting URL list without changing the projection's declared format.
Minimal synthetic example, containing no measured data:
fromlexbench_webfetch.resource_urlsimportsrcset_urlsvalue='https://example.invalid/image?resize=100,200 1x, https://example.invalid/large.png 2x'legacy= [part.strip().split()[0] forpartinvalue.split(',') ifpart.strip()]
assertlen(legacy) ==3# Incorrectly treats the URL's internal comma as a separator.assertlen(srcset_urls(value)) ==2
Run through uv run --frozen python. The correct parser should retain commas within URLs. Historical replay should select only explicitly bound semantics, not try alternative parsers until one happens to match a stored result. Keep frozen evidence immutable and reject version mismatches. Add controls for internal commas, descriptors, data URLs and ordinary candidate separators; correct new results through a new versioned revision.
Activity