gap

Author: knopkem

OPENJPH_RS_HTJ2K_BLOCKERS.md

@@ -0,0 +1,55 @@
+# openjph-rs HTJ2K migration blockers
+
+## Current blocker
+
+The current remaining blocker for switching `dicom-toolkit-rs` HTJ2K over to
+`openjph-rs` is the broad irreversible 9/7 parity gap (`irv97`).
+
+## Current evidence
+
+Running:
+
+```text
+cargo test -p openjph-core
+```
+
+currently fails in:
+
+```text
+tests/integration_encode_decode.rs
+```
+
+with:
+
+```text
+38 passed; 50 failed
+```
+
+Running:
+
+```text
+cargo test -p openjph-core irv97
+```
+
+reproduces the same blocker bucket directly.
+
+Representative failing tests:
+
+- `dec_irv97_64x64_rgb`
+- `dec_irv97_gray_tiles`
+- `enc_irv97_decomp_0`
+- `enc_irv97_16bit_gray`
+- `enc_irv97_tiles_33x33_d5`
+
+Representative current MSE values:
+
+- many 8-bit RGB failures: about `5464.9844`
+- 16-bit failures: about `1_039_004_000` to `1_053_217_000`
+
+## Recommendation for `dicom-toolkit-rs`
+
+Do **not** switch the active HTJ2K backend to `openjph-rs` yet.
+
+The next required step is a dedicated `irv97` parity pass against local C++
+OpenJPH. Only after that bucket is green should the backend swap be retried and
+the downstream HTJ2K integration tests be rerun.