v4.5.1

Compatibility

• Fixed a crash on Python 3.9: load_plugins() called importlib.metadata.entry_points()
  with a group= keyword that Python 3.9 doesn't support there. ctrl-kd now works on
  Python 3.9 again, not just 3.10+.

Printed-mode fidelity

This release is measured against real, factory-default WordStar 7 output (not just our
own previous exports), and fixes a batch of small placement bugs found that way:

• Superscript and subscript now use the correct size and horizontal spacing for each font
  family, instead of one fixed value.
• Header and footer lines can now contain inline style toggles (bold/italic/underline
  partway through the line) without corrupting the text.
• Running heads now honor a page's own left/right offset instead of always using page 1's.
• Right-aligned tabs in headers and footers now recompute their padding on every page
  (previously only correct on the page it was first calculated for).
• Footnote reference markers now join correctly to their note text.
• Page numbers are now on by default, matching stock WordStar 7 (previously off unless
  requested).
• Automatic single-line spacing is now 1.0x, matching WordStar 7's own default (previously
  1.2x, borrowed from a customized install).
• Top margin now uses the document's .mt setting alone, instead of double-counting it
  with the header margin.
• The footnote area reserves the same vertical space stock WordStar 7 does.
• Running head lines now use the page's default margins correctly.
• Fixed an off-by-one in right-tab column placement.
• A typed paragraph indent combined with a style-driven .pm indent is no longer
  double-counted.
• HTML output: fixed a rendering glitch where a tab-and-leader run in a centered row
  (e.g. a symmetric table-of-contents line) produced an oversized gap instead of a
  single space.

Testing infrastructure

• New coordinate-level test tier that checks PDF output against real WordStar 7 printer
  output, page by page.
• New tooling to track and record page-by-page visual review against real printed pages
  (paper scans), so a reviewed page is a recorded fact instead of tribal knowledge.
• CI now runs the test suite on every push, not just manually.

Known state at release

• Reduced (public) test suite: 846/846 passing.
• Coordinate-level fidelity tier: 14 of 18 captured documents match real WordStar 7
  output exactly. 4 have named, unresolved residuals:
  • LJ6DTP — parked by design; its own printing-hack features are out of scope for
    this tier.
  • LYING and WARPRAYR — small residual font-metric drift from a substituted font
    (ctrl-kd never embeds fonts; the base-14 substitute doesn't have identical metrics
    to the original).
  • -SCREEN — a test-harness limitation matching short Greek/math symbol fragments
    across a font switch, not a rendering bug.
• Paper-scan visual review: 69 pages catalogued, all recorded as unreviewed pending
  human review — tracked and testable, not silently skipped.