We tested PUB to PDF conversion: the results
Last updated
On October 27 and 28, 2026, we converted 23 public Publisher test files with Aretus and two other engines. On a 4-page newsletter, 0.0% of the words in our PDF collided with other text, against 48.6% for the open-source import on its own and 2.3% for a commercial PUB library. Every measurement and the method are below.
The test files
23 real Publisher files from public test suites that software projects use to check their Publisher support. No random downloads.
| Source | Files | What they are |
|---|---|---|
| Apache POI test files (test-data/publisher) | 21 | The same sample saved from Publisher 98, 2000, 2003/2007 and 2010; a newsletter template; a brochure; a résumé; a flyer; hyperlink tests |
| Apache Tika test documents | 1 | Simple text |
| LibreOffice's Publisher import tests (libmspub, files that must open) | 1 | A Publisher 97 file from a LibreOffice bug report |
- By the file format the open-source importer reads them as: 20 files in the Publisher 2002-and-later format, 2 in the Publisher 2000 format and 1 in the Publisher 97 format.
- A separate security set of 6 hostile files, made to crash converters: Apache POI's fuzzing cases and the Exploit-DB files LibreOffice tests against.
- Most of these files are simple test documents. Real church bulletins and school newsletters look more like the newsletter and the brochure, which is why those two are reported file by file.
- No public source pairs .pub files with PDFs made by Microsoft Publisher, so the output couldn't be scored against Publisher's own. Instead we looked at every page of every output, compared the same document saved from different Publisher versions, and measured text collisions automatically.
The engines compared
The same 23 files and the same fonts went through three engines:
- Open-source import, no repairs. The open-source Publisher importer that free office suites use, and that several free online converters say they use, run headless on its own with none of our repairs. It's also where Aretus starts.
- Commercial PUB library. A commercial .NET library for reading and converting PUB files, in its evaluation version, run with .NET 8 on Linux. It stands in for online converters and products built on it. It has no font fallback, so it was given the same substitute fonts as the other engines, renamed to the font names each file asks for. Its evaluation watermark was left out of the measurements.
- Aretus. The open-source importer plus our layout repairs (linked columns, text wrap, overflow) and free lookalike fonts.
How we measured
- Text collisions: we read the position of every word in each PDF (with pdftotext -bbox-layout) and counted the share of words that overlap another word by more than half a line's height.
- Repeated stories: we searched each PDF for repeated 8-word passages, the sign of a story printed again in every linked box.
- Everything else, from pictures to cut-off headings, by looking at every page.
- Speed: the median time per file, converting 4 files at a time. Aretus's time includes making the PDF, ODG and DOCX in one pass.
- Dates: the open-source import against Aretus on October 27, 2026; all three engines on October 28, 2026.
- Not compared: services that run Microsoft Publisher itself, and tools with their own renderers. Testing them means uploading the files to them and paying, so we didn't.
Text colliding with other text
Share of words overlapping another word by more than half a line's height. Lower is better. October 28, 2026.
| File | Open-source import, no repairs | Commercial PUB library | Aretus |
|---|---|---|---|
| Newsletter, 4 pages, linked columns | 48.6% | 2.3% | 0.0% |
| Brochure, 2 pages, many pictures | 3.8% | 2.1% | 0.0% |
| Flyer (title over a photo) | 15.4% | 0.0% | 0.0% |
| Résumé | 0.0% | 0.0% | 0.0% |
Files converted
| Open-source import, no repairs | Commercial PUB library | Aretus |
|---|---|---|
| 23 of 23 | 13 of 23 | 23 of 23 |
Of the commercial library's 10 failures, 8 are because a graphics component it uses only works on Windows, so it would probably convert them on Windows. Two are genuine: the Publisher 97 file isn't supported, and the Publisher 98 file crashes while being read.
What we saw on the pages
October 28, 2026.
| Aspect | Open-source import, no repairs | Commercial PUB library | Aretus |
|---|---|---|---|
| Linked text boxes | The whole story repeated in every linked box, spilling over everything | Flows correctly (it reads the real chain order) | Flows correctly within a page |
| Text wrap around pictures | Ignored: text runs under pictures | Couldn't judge: no pictures rendered on Linux | Rebuilt (square wrap) |
| Pictures | Present | Missing in this test, likely because of the same Windows-only component, so not counted against it | Present |
| Headings | Overflow and overlap | Some cut off (“Living and Learning in”, “What it means to be an America”) | Complete |
| Words | Intact | Often broken mid-word (“includ es”, “capturi ng”, “organizat ion”) | Intact |
| Fonts | Generic stand-ins (wide) | Our lookalikes, by design of the test | Lookalikes, 6 of them metric-compatible (same letter widths) |
| Overflowing text | Spills out of the box | Kept | Shrunk to fit down to 70% of its size; hidden only beyond that (2 boxes in the newsletter, 0 in the brochure) |
Open-source import against Aretus, file by file
October 27, 2026, before the overflow change described below.
| File or check | Open-source import, no repairs | Aretus |
|---|---|---|
| Converted | 23 of 23 | 23 of 23 |
| Newsletter (4 pages, linked columns) | Broken: every linked column repeats the story from its start; text overlaps pictures and other stories | Columns flow correctly, text wraps around pictures, nothing overlaps |
| Brochure (2 pages) | Broken: text runs under pictures, titles overflow their boxes | Text wraps around pictures; titles fit; a small overflow hidden in one panel |
| Flyer | Broken: the title overlaps the photo (Gill Sans MT drawn in a wide stand-in) | Correct, in a Gill Sans lookalike |
| Simple text, hyperlink tests, résumé | Good | Good |
| Publisher 98 and 2000 saves of the sample | Extra blank page; table missing; Arial bold italic drawn as a serif font | Blank page removed; table and font still wrong |
| Publisher 97 file | Converts; the content is mostly empty | Same |
| Fonts drawn in a generic stand-in | 9 of 12 non-default fonts | 0 (all mapped to lookalikes; 6 metric-compatible) |
| Hostile files | 6 of 6 rejected cleanly | 6 of 6 rejected cleanly, under 2 s each, nothing left running |
| Median time per file (4 at a time) | 1.3 s | 2.8 s (PDF, ODG and DOCX in one pass) |
Files with broken pages: 3 of 23 with the open-source import on its own, 0 of 23 with Aretus. With Aretus, the two Publisher 98 and 2000 files still have a table missing.
What we changed after the comparison
The commercial library kept whole stories where Aretus hid text that didn't fit. So on October 28, 2026 we changed how Aretus handles text too tall for its box: it's now shrunk until it would fall below 70% of its size, instead of being hidden as soon as it's 35% too tall. Hidden boxes went from 3 to 2 in the newsletter and from 1 to 0 in the brochure.
How the converter works, step by step: How the converter works.