A check for (a + b) / 2

[cushon]

Original issue created by kevinb@google.com on 2013-04-04 at 03:45 PM

---

Fact: some percentage of all occurrences of "(a + b) / 2" out there are bugs, or bugs waiting to happen.
(http://googleresearch.blogspot.com/2006/06/extra-extra-read-all-about-it-nearly.html)

Unfortunately, there's no reasonable fix we can suggest that is guaranteed to work for all possible values of a and b... except Guava's IntMath.mean(a, b) / LongMath.mean(a, b) methods.

If there is ever a place in error-prone for "checks for which the fix requires Guava", then it might be worth having a check that forces users of "(a + b) / 2" to change to use mean(). Most of those users probably weren't vulnerable to the bug, but we can't tell which they are.

If we don't do this, we'll probably just have to give up on ever catching that bug, and we will sadly probably give up on these Guava methods entirely, because literally NO ONE is using them (why would they? They already know how to get the mean of two numbers!).

[cushon]

Original comment posted by alexeagle@google.com on 2013-04-27 at 07:43 PM

---

I imagine we could determine that the code is being compiled with Guava available in the classpath, which would mean we could be confident in requiring it to be used here. But it would be pretty strange to add Guava to your classpath and then your compile breaks.