Pavol Rusnak #cypherpunk #hacker #openhw #privacy #bitcoin #newmediaart

Idea Factory - Semantic Patches on Github

This post is a part of Idea Factory series.

Two years ago someone sent me quite weird pull request on Github. It took me some time to realize that the only change was the removal of whitespace at the end of the lines of all files. Commit message contained a link to some software development company so the whole effort could be considered as some kind of spam or publicity stunt.

But it sparked an idea!

When I was at FOSDEM I saw a great talk about Coccinelle which is a tool for semantic patches. These are similar to normal patches, but they contain something like expressions which are being matched during applying of the patch. That way they apply the change to usually more places of the codebase not just one (which is the case when using “normal” patches). Basically, you want to come up with most generic “rule” that fixes one particular issue, but it’s not triggered at any other time. Some examples how Coccinelle was used in Linux kernel development are here.

Another quite common example (which I’ve seen at many, many places while working as a package maintainer at SUSE) is the wrong usage of strncpy function. The last argument has quite different meaning than the one used in strncat function. It is the maximum length of the appended string, not the size of the whole output buffer. The patch fixing this issue is similar to the following:

--- source.c
+++ source.c
@@ -34,7 +34,7 @@
   return 0;
-strncpy(buf, player, sizeof(buf));
+strncpy(buf, player, sizeof(buf) - sizeof(player) - 1);
 if (verbose)

This patch obviously fixes the problem only at one particular place. Semantic patch that fixes all wrong usages of the function might look like this (but please don’t take my words for granted, I have not tested it):

identifier dest, src;
-strncpy(dest, src, sizeof(dest));
+strncpy(dest, src, sizeof(dest) - sizeof(src) - 1);

So back to the idea:

  1. Collect the most common programming mistakes people make.
  2. Create semantic patches for fixing them.
  3. Write a bot that scans Github for these issues and creates a pull request when necessary.
  4. PROFIT!

While writing this post I found this article which says that bots are generally not welcome on Github. But I think if the pull requests are reviewed by a person and the fixing rules are beneficial to everyone, then it might really be worth a try!