Week 6 + Week 7: Fixing bugs & writing tests

Saturday, 18 July 2026  |  Ojas Maheshwari

I spent the last 2 weeks mostly on fixing the bugs that were breaking font subsetting for annotations.

One such bug was about deleting the original font which we talked about in the last blog.
And there were some more edge-cases and type bugs.

For eg: An object could be a ref but I directly do obj->getStream() instead of obj.fetch(xref).getStream() which crashed the code.

I spent 3-4 days on writing some tests for the font subsetting. I just test 2 things for now.

  1. Single annotation test: Here we test if subsetting behaves well for a single annotation. We load a file, add an annotation, save it, re-open it, and check if the annotation is using a properly subsetted font which has glyphs only for the characters we added.

  2. Linked annotation test: We try to check if subsetting for a particular annotation leaves all the other annotations intact. So, we load a PDF file which has 2 annotations that use the same font, modify the 1st annotation, save and re-open, check if the subsetting works well for the 1st annotation, and check if the font binary data for the 2nd annotation is intact.

These tests required approximately 400 lines of code which is slightly surprising. This is almost the same amount of code as the subsetting logic in FontSubsetter.cc

I also used core poppler code for these tests and not an API like poppler-qt5 or poppler-qt6 because I needed a lot more control to test these things.

I also switched from Vim + terminal coding workflow to QtCreator. This was basically because I am not very good at GDB right now and wanted a visual debugger to make work easier. Switching to CLion just for it's debugger felt weird.

So I started using QtCreator and it feels okayish right now. I use it with it's FakeVim plugin.

Now, I need to work on:

  1. The extra font bug in poppler: Basically, when poppler tries adding fonts for a text, it seems to be including an extra wasteful font such as Cantarell in my case. I need to investigate why this happens.

  2. Make our own splitTextByFont: Right now, I just save the font-string mappings when the AP stream is generated in an annotation. This ensures consistency between what is being written to the AP stream and what is visible to the subsetter.
    However, this is not very good both performance-wise and code-wise. We store the mappings for annotations that we might never subset. And it requires change in the internal code.
    Therefore, I have to create my own splitTextByFont function inside my FontSubsetter class which takes the font and returns the appropriate font-string map.
    We need to be careful that we don't cause inconsistencies between what's visible to the subsetter and what actually gets added to the AP stream because that might cause rendering issues.

  3. Fixing more bugs: Because bugs never end in moving software, do they?

Thanks for reading! Have a great day ☺️