Calc has a few issues with row height calculation, which are somewhat interrelated.

To reproduce:

1. Open a blank spreadsheet.
2. Format cells A1:B4 to auto-wrap text.
3. Fill cells B1:B4 with text, about 6 rows per cell.
4. Apply optimum row height if necessary.
5. Zoom in (about 140%).
6. Verify row height, and apply optimum row height again.
7. Zoom back to 100%.
8. Verify row height, and apply optimum row height again.
9. Fill A1 with long text (longer than B1:B4 combined).
10. Merge A1:A4, apply optimum row height if necessary.

Expected behavior:

* Zooming in and out should not affect row height calculation, i.e. steps 6 and 8 should not result in row height changes.
* Editing text should not affect row height changes, unless the cell being edited is the single tallest cell in its row either before or after editing.
* If a merged cell needs to be taller than the sum of its constituent rows, the rows should be enlarged (e.g proportionally).

Actual behavior:

* When zooming in or out, text may suddenly become too long for the cell to accommodate.
* When editing text in a row, even a one-liner next to a cell containing multiple rows (previously set to optimum height), the row may shrink slightly, causing the long text to no longer fit.
* Merged cells are as tall as their constituent rows, never taller, even if their content does not fit.

Additional comments:

* Size calculations should be done in a display-neutral manner; same goes for the actual layout. This is also occasionally an issue with printing: text fits into one row on screen but is wrapped in the printout due to minute differences in layout and text length being close to one full row.
* It looks as if the code has two different routines for row height calculation in different places, with different results. Code deduplication would probably solve that.
* Merged cell height is probably a tricky issue, with multiple possible solutions. If nothing else works, extend the last row of the merged cell so the content fits. Proportionally enlarging cells would give nicer results, but may become difficult if different merged cells span overlapping ranges of rows (say, A1:A2 and B2:B3).