A story of bug fixing
9 April, 2016 § Leave a comment
I spent some time last night looking in to a bug in the Firefox preferences as part of the Outreachy internship I’m mentoring. This bug is about how the context menus in our preferences were squished horizontally compared to normal context menus. I’ve put screenshots of the two below so you can compare.
Looking at bug 1247214, I knew that context menus outside of the preferences looked fine but context menus in the preferences were broken. This meant I had something I could compare side-by-side when looking at the CSS.
I used the inspector and looked at the applied styles for a
xul:menuitem and compared them between the preferences and non-preferences. I was surprised to see that they were nearly identical, and what differed was inconsequential.
After walking up the DOM from the
xul:menuitem and seeing that the two instances were still nearly identical, I had to think of a different route. Somewhat by chance I realized/remembered that there were children nodes of the
xul:menuitem that could possibly be affected. In the Inspector I expanded the
xul:menuitem and looked at the rules for the
xul:label inside of it. Immediately I noticed that the
-moz-margin-end were being set to zero and the rule looked like it wasn’t intended to affect context menus. Unsetting that rule fixed the bug.
So that describes how I found the issue with the bug. Figuring out a fix for it wouldn’t be that difficult now. I also wanted to find what changeset introduced the bug so I could learn more about why it was there.
Using MXR, I pulled up the `hg blame` for the file and went to the offending line. Clicking on the revision number on the left of the line brings up some of the diff from the changeset. At the top of this new page, I clicked on “changeset” to see the full changeset. This unfortunately was a refactoring changeset and wasn’t what actually introduced the offending lines. I then clicked on the parent changeset and based off of that revision found the file again and ran `hg blame` on it. Now the offending line was linked to the changeset that introduced the bug.
After reading through the bug that introduced the changeset I knew that I could safely make this change as the lines were never intended to alter context menus, and I also knew what to look for and test while making the change so I wouldn’t regress the original bugfix.
All of this being said, I probably could have found this quite a bit faster by using a tool called mozregression. mozregression is a tool that will help you run a binary search on all of our nightly builds (continuous integration ones too). As you are likely already aware, binary search is very fast, letting mozregression help narrow down the day that the regression got introduced from a range of 10 years to a single day in about 7 builds.
If I had taken the mozregression route I probably could have saved myself about 45 minutes of debugging and digging around. But both routes work well, and sometimes I learn more from one than the other.
I know this write-up was long, but I hope it was valuable.