WordArt, clip art, and calendars: the hard parts of Publisher files
Extracting the text from a Publisher file is a solved problem. Several tools can do it. What separates a conversion you'd actually use from one you politely close is everything else: the decorative, fiddly, extremely-2003 details that make the document look like itself.
Those details are where we've spent most of the past year. A tour of the worst offenders:
WordArt
The arched headline over the bake sale flyer. Text bent into a full circle around a logo. Publisher's WordArt goes far beyond bold-and-colorful: each preset warps the letters along its own geometry, and getting it wrong is very visible. Most converters give up and render the text flat, which turns a poster back into a memo.
Our renderer reproduces the shapes, including the circular and ring presets where glyphs have to march around a curve one by one. This took an embarrassing amount of trial and error to get right. It was worth it. A surprising fraction of real-world .pub files lead with WordArt, because of course they do; it was the fun button.
Clip art and old picture formats
That clip art someone picked from the gallery in 2004 isn't a PNG. It's usually a WMF or EMF, Windows drawing formats from the 90s that store instructions ("draw a line here, fill this shape") rather than pixels. Modern software mostly pretends these formats don't exist, which is why converted newsletters so often have mysterious blank rectangles where the art used to be.
We render them. Lines, fills, embedded bitmaps, the odd coordinate systems, the pictures that are secretly rotated a quarter turn. If your 2006 newsletter had a cartoon turkey on it, the turkey survives.
Tables, especially calendars
Publisher's calendar wizard built a table. So did the price list template and the "business hours" box. Tables sound easy and are not: every cell can have its own borders, fills, and alignment, some cells have dark backgrounds with light reversed-out text, and some tables run their text vertically. Drop one border and a calendar stops looking like a calendar.
The engine reads the table structure directly from the file: column widths, row heights, per-edge borders, cell fills, and the text flow inside each cell, including the vertical kind.
Gradients, textures, and picture effects
The soft color fade behind a title, the washed-out photo used as a background, the drop shadow under a text box. These effects are stored as tiny recipes in the file, and each one had to be decoded and reproduced. When they're dropped, pages don't break exactly. They just go flat and gray, and the person who made the file notices immediately.
Why bother with any of this
Because the person converting a .pub file isn't converting a data structure. They're converting the newsletter they made when their kid was in third grade, and they can tell when it comes back wrong.
Publisher itself remains the only software that renders these files perfectly, and it's leaving. Our goal is to be a close second, and we test against tens of thousands of real documents to hold ourselves to that. When something still comes through wrong, that file becomes a bug report and, usually, a fix.
Try something with strange formatting and see. The preview is free.