Re: Emacs 28 on OSX: emacsclient.c:1415: warning: implicit declaration of function 'openat'

[Paul Eggert]

On 4/17/22 02:01, Mattias Engdegård wrote:
> `openat` was introduced in OS X 10.10, but shouldn't gnulib have a substitute? 
> Paul?

Gnulib's openat substitute is so heavyweight that Emacs doesn't use it. 
(The substitute implements openat via fchdir-and-fchdir-back which is 
dubious for Emacs.)

Instead of using Gnulib openat, I installed the attached patch into the 
emacs-28 branch so that Emacs uses openat only on platforms like GNU 
that define O_PATH (as these are the only platforms where Emacs passes a 
value other than AT_FDCWD to openat's first argument). I tested this in 
an artificial Fedora environment where I removed O_PATH, and it passed 
'make check', so this should port to OS X 10.9 as well. Please give it a 
try.

PS. Could we somehow arrange for port-to-old-macOS testing to be done 
before an Emacs release, rather than after?

0001-Don-t-assume-openat.patch
Description: Text Data