I was recently looking at a PDF document and suddenly noticed Evince (GNOME’s PDF reader) was using a lot of memory for a ~100 page article that mostly contained text. Like a good FLOSS citizen, I looked upstream if there were any issues that matched what I had just seen, and that’s where I found my first clue on the problem. Someone reported that if you tried to load a document with a lot of annotations, memory consumption would immediately go up by several hundreds of Megabytes!
Since I contributed to GNOME already, I figured I might as well take a look if there was any quick fix. I built Evince using jhbuild and created a heaptrack trace. By analyzing that and looking at the source code, I figured out that Evince was loading each annotation and creating a window for it. Each window instance contained a spell checker, and that of course needed to maintain a complex set of rules and known words.
In the end, I created an MR that kind of fixes the problem by removing the line that loaded all the annotation windows at the very beginning. It’s a first step to improve the memory consumption, and even though you could still end up with the same memory usage after individually opening up each annotation, you at least won’t suffer from this if you have no plans of doing that. Long term, probably the spell checker should be shared over the whole document (and maybe also lazily loaded).
One of the big takeaways I wanted to call out in this post though, is that somebody originally reported the issue and with just enough detail, so that I could recognize the same symptoms for my situation, and even make a partial fix for the problem. So if you ever run into a bug: please take the time to go to the project’s issue tracker and file a bug. It’s a small but important step to help FLOSS projects move forward :-)