If you load a bunch of packages at startup that take several seconds to load that is as good a benchmark as any. I have this at the end of my init.el:

    (run-with-idle-timer
     3 nil
     (lambda ()
       (let ((inhibit-message t))
         (message "Emacs ready in %s with %d garbage collections."
                  (format "%.2f seconds"
                          (float-time
                           (time-subtract after-init-time before-init-time)))
                  gcs-done))))

My startup had always been under a second. On a Mac, the command to get the path right is the longest pole. Remains so in native.

That said, I'll play with it some. Still looking for a broader benchmark, though.

Edit: Looks like I lied. Startup is about 2 seconds, but is mainly dominated by a network hit to refresh packages. (So, slower if I'm on a bad network... I should fix that.) Your script, though, is odd, as I can get to the Messages buffer and wait a second before I see that message show up. (Also, you can just eval (emacs-init-time))

I'm sure I got this code from someone else, was going to replace it with (emacs-init-time) myself but I suppose it's nice to only have the first two numerals after the decimal point which this code does and (emacs-init-time) doesn't.

Funny, coming back to this, if I recompile without native compilation and any other flags, I get faster startups. By upwards of 1 second. (So, from almost 2 seconds to almost 1 second.)

An emacs -q startup is roughly the same. And I maintain that many things feel snappier with the optimization flags.

Apropos of the path: have you tried setting your path in, say, `.zshenv`? When I started using that, I was able to shave off a significant amount of time because Emacs didn't have to fire up a full interactive shell to grab the env vars.

This is only on a Mac, right? I'll look into it, not a problem on my other machine.

IME the only thing -march=native does is it makes it impossible to boot your system on an older microarchitecture (which is trivial with Linux otherwise — I really don't understand those who bring their bad Windows habits and do a reinstall when moving between systems).

I mean, it would do that. But, I also have never just moved the hard drive between machines. I've only owned two machines in the last decade. So, not even something I have had a lot of opportunity to do.

Back to the point, though, I'm assuming that libc dynamically does most of the optimizations that native enables?