Zack discusses mysterious alignments in the kernel; and discovery and invention.
Mysterious Alignments in the Kernel
Sudip Mukherjee reported a build failure with the mainline kernel source tree. This is fairly unusual. Most patches get tested to see if they’ll compile on a variety of different systems before the patch is accepted, but a build failure can happen.
After noticing the failure, Sudip did a
git bisect – a fast way to jump half the distance to the problem, then half the remaining distance, and so on. It’s one of the most common techniques for identifying which patch actually caused a bug. Sudip used it to track the problem down to a small patch that compared the number of bytes used by two different variables that needed to be the same.
Linus Torvalds is often interested in build failures or crashes in the mainline tree because he maintains that tree himself. Linus responded right away to Sudip. He said he had looked at the data size comparison, and it seemed like a good patch to him. He said, “That BUILD_BUG_ON() looks entirely correct, and if that sizeof() doesn’t work, then the code doesn’t work.” In other words, the check was appropriate. Linus added, “I’m not seeing what could go wrong in there.”
Use Express-Checkout link below to read the full article (PDF).