Track remaining HOP document-fidelity reports with independent reproduction cases #3

Scope and current evidence

Follow-up to the requested downstream HOP issue work. This is a reproduction/acceptance tracker, not a claim that the following upstream bugs reproduce in current OpenGeul or have been fixed. The keyboard reports #95/#100 and toolbar-label request #97 are handled separately in OpenGeul PR #2.

Priority 1 — table text must survive rendering and export

Upstream report: golbin/hop#99 (HOP 0.4.4, Windows 11). The reporter distinguishes intact HWP text from missing last-line rendering in a narrow table cell, including PDF/print output. Related: golbin/hop#78 .

• Obtain a redistributable, non-personal minimal fixture (or generate one); do not republish customer documents or installed font programs.
• Hold rhwp commit, input bytes and installed font families constant when comparing HOP, OpenGeul and rhwp. A moving online demo is not a version-controlled baseline.
• Confirm the tail text exists in independently extracted source text and the saved round trip.
• Confirm the same tail appears in rendered SVG/PDF text and is not clipped from the final visible page; a text-extraction-only assertion is insufficient for visibility.
• Cover narrow cells, multi-line text, merged cells, page splits and missing-font substitution. Preserve the original source on failed export.
• Fix only after an actual failing regression; keep screen, PDF and print-preview comparisons separate. Do not remove clipping globally as a workaround.

Priority 2 — legacy HWP save interoperability

Upstream report: golbin/hop#73 (reported on HOP 0.3.1). The report says saved files can fail to reopen and table assets/layout can break; the issue does not itself establish a cause.

• Reproduce on a minimal permitted fixture before declaring current affected scope.
• Verify OLE streams, text/control records and embedded-object counts before and after an unchanged save and a small edit.
• Reopen with a properly licensed independent reference application, not only the same rhwp parser that serialized the file.
• Add repeated save/reopen and special-character/control-content regression checks without normalizing away the failing input.
• Keep original files and explicitly document any unsupported content; do not silently discard records to make parsing pass.

Completion rules

A report is resolved downstream only with a reproducer, code change, passing regression and exact commit/run evidence. Browser success does not certify native Windows dialogs, real printers or all third-party fonts. Upstream issue closure/acceptance is a separate maintainer decision; do not close upstream issues on the strength of a downstream test alone.