What is the point of this configure check? This change calls localtime() unconditionally. Which is reasonable since no OS fish runs on should not have an implementation for that function.

It's not localtime, it's localtime_r which is sometimes not available on Win32 IIRC.

AC_CHECK_FUNCS defines a preprocessor symbol if the function is available, but there is no check for this in the code (and no alternative handling).

@zanchey: Do we need to support platforms which don’t define localtime_r? I’m really not wild about using localtime, since it’s not thread-safe. Would it be acceptable for configure to just halt with an error if localtime_r is undefined?

Somewhat surprisingly we don't currently have a single localtime() call in the code so it's use in this change would be thread-safe since it is the only thread that would ever call it. However, that's putting down a land mine. So this is a case where a fallback function needs to be defined if the platform doesn't provide it. The fallback should just use a simple global lock around the call to localtime() to ensure the single internal result buffer isn't overwritten before it's contents are copied into the user supplied result buffer. Look in the code for instances of scoped_lock locker(lock); //!OCLINT(side-effect) for how to do that.

I should also note that you change has another bug: You don't call tzset() (and neither does the current fish code). It's possible the internal timezone state is being set indirectly via some other call but you shouldn't rely on that.

It looks like it's going to be mingw that may not have this - seems like it can be expected on Cygwin? Perhaps it's not an issue?