In early October of 2014, I started collecting changes that I saw fly by on llvm-commits that I thought would be straight-forward to automate. I was trying to be pretty conservative, so these tend to be pretty basic things: fixing deceptive white space around an if clause, removing the name of a method from it’s doxygen comment, removing a couple of syntactically redundant semi colons, and things of similar complexity. These weren’t chosen because they were interesting, but precisely because they were not.
In the 66 days since I started collecting, I’ve saved 105 unique commits. That’s a bit less than 2 per day, and only about 1.6% of the 6,500 commits made to LLVM in that time.
Let’s assume that each of those changes took an average of 15 minutes on the part of their author. That’s not too much more than a single build and test cycle, so it seems like a reasonable estimate. At roughly $2 per developer minute, we can guesstimate that each of these changes cost about $30. Taken together, these 105 changes consumed about 26 hours of developer time at a cost of a bit over $3,150.
This gives us a value for straight forward code maintenance activities of roughly $1,400 per month (or roughly $50 per day.)
If anything, this is an extremely low estimate. I know several of these changes required review, and at least a couple of them broke the build and had to be reverted. We could probably add in several more hours of developer time just for that alone.
Now this is only a small fraction of the roughly $88,000 in development time going to the project as a whole each month*, but it’s still pretty material.
* Using the same logic as above: 15 minutes per change, $2 per developer minute, 6500 changes, llvm repository only. It goes without saying that this is a massive understatement of the actual value of the contributed work.