[lgritz]

Tiles are emulated by buffering the whole image and writing it out as scanlines on close(). But close() took a null m_stream to mean "already closed", and m_stream is only created once the last scanline is written (never, on the OpenJPH path), so the buffered image was silently dropped, leaving an empty or truncated file. Test for the image objects instead. Tests added for both the OpenJPEG and OpenJPH writers.

Also, write_scanline() let y == height through, so copy_scanline() wrote a row past the end of the OpenJPEG component buffers, and negative y was not checked at all. And copy_scanline() subtracted m_spec.y a second time; that is latent only because check_open() zeroes the origin for this format.

Assisted-by: Claude Code / Claude Opus 5.5

[Copilot]

🟡 Changes recommended

The tiled round-trip tests omit invalid-row boundary cases, leaving the JPEG2000 scanline guard unprotected against a regression that could reintroduce a buffer overrun.

1 open issue

• Add JPEG2000 scanline boundary regression tests · New

What changed in this PR

This PR fixes JPEG2000 writer behavior so buffered tiled output is saved on close and invalid scanline indices are rejected before accessing component buffers.

Changes:

• Detect open writers using image objects rather than the output stream.
• Correct scanline bounds checks and remove duplicate origin subtraction.
• Add tiled-output round-trip tests for the OpenJPEG and OpenJPH writers.

File	Description
testsuite/​jpeg2000/​run.py	Adds an OpenJPEG tiled-output regression test.
testsuite/​jpeg2000/​ref/​out.txt	Records the expected tiled round-trip result.
testsuite/​jpeg2000/​ref/​out-icc.txt	Updates the ICC-enabled reference with the tiled result.
testsuite/​htj2k/​run.py	Adds an OpenJPH tiled-output regression test.
testsuite/​htj2k/​ref/​out.txt	Records the expected OpenJPH tiled round-trip result.
src/​jpeg2000.imageio/​jpeg2000output.cpp	Fixes close-state detection, scanline bounds, and row offsets.

_{Review effort: Balanced}