This contains a bug that was mentioned on a Google programmer's blog some time back. You add posmin and posmax together, possibly inducing overflow or promotion to floating point values.

You should of course take the difference of posmin from posmax, divide it by 2, and add the result to posmin.

This came up on perl5-porters mailing list some time ago: the core was also vulnerable to this error. See the summary on use.perl for the background on this issue.