The value of Float.nan ** 0. expression depends on the version of glibc: it's nan with glibc 2.25 or later, and 1. for older versions.

This is caused by changes in the handling of signaling NaN values in glibc to confirm to TS 18661-1. It affects OCaml's Float.nan, which is a signaling NaN. In this example, the implementation of pow in glibc changed (see issue 20916.), but there may be other functions.

This changes the behavior of some existing ocaml code in subtle ways. How should it be handled?

@xavierleroy