mirror of
https://github.com/storytold/lightcraft.git
synced 2026-10-24 10:09:01 -06:00
Decoders take a Mode: Header reads everything but the samples (DNG, CR2, ARW, RAF, RW2; PEF when it has crop tags). NEF/ORF (and PEF without crop tags) derive geometry/CFA from the samples and still unpack them. decode() stays bit-exact (debug-hash of every corpus/raw decode identical before/after); the corpus test asserts probe_info == decode().info() for all 37 files. Import of 14 corpus raws: 3.4 s -> 1.4-2.0 s; peak RSS through import 1.41-1.49 GB -> 1.12-1.22 GB. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>