[KnifeLemon]

Refs #138. The sample from @cache51 there (input_p2.pdf, a garment measurement sheet printed from a spreadsheet) has every cell ruled, yet PDF → DOCX gave no table: 63 loose paragraphs.

Cause

The grid was found (40 rows × 17 columns), then thrown away because of one word. "Front NK Drop (from HPS to Edge)@ COLLAR NOTCH" is longer than its column:

• the spreadsheet cut it off at the cell, and the PDF only clips the rest;
• the text reader still sees the whole word "NOTCH", which then sits across the cell's right rule;
• the table finder takes any rule through a word as a sign of a chart or a strike-through, so the whole table became text.

Change

PdfDocumentReader.Tables:

• A word that runs on past a vertical rule stays in the cell it starts in.
• It's allowed for up to one in twenty of the grid's words (at least one), so a chart with labels across its lines still stays text.
• A horizontal rule through a word (a strike-through) still means no table.

Result on the sample

	Before	After
Word tables	0	1 (40 rows)
"…@ COLLAR NOTCH"	a loose paragraph	in its cell, next to 1/8 and 3 1/2

Merged cells, shading and the heavy frame come through too. Still different from the PDF, not part of this fix:

• rows come out a little taller, so the table runs onto a second page;
• the item labels come out bold;
• the jacket sketch lands under the table instead of beside it.

Test

PdfTableTests.Long_text_running_past_its_cell_stays_in_that_cell draws a 4 × 3 grid in code, with one label whose last word straddles its cell's right rule. It checks that the label, its tolerance and its value each land in their own cell. The 27 existing PDF table tests still pass, including Boxes, underlines and empty grids stay text.

🤖 Generated with Claude Code