(unless init-file-name
		    (setq init-file-name alt-file))
is added
in https://git.savannah.gnu.org/cgit/emacs.git/commit/?id=4118297ae2fab4886b20d193ba511a229637aea3.

Before this commit, I think ‘.emacs’ is always used  for ‘user-init-file’ value if
init file is missing.

                 (and (equal (file-name-extension alt-file) "el")
                      (setq alt-file (file-name-sans-extension alt-file)))
...
                 (load alt-file 'noerror 'nomessage)))

should be also changed in the commit to use “init.el” not “init".


> 2020/12/16 1:23、Eli Zaretskii <eliz@gnu.org>:
> 
>> From: Lars Ingebrigtsen <larsi@gnus.org>
>> Cc: tsuucat@icloud.com,  45197@debbugs.gnu.org
>> Date: Tue, 15 Dec 2020 07:14:58 +0100
>> 
>> alternate-filename-function is the lambda up there, so it's
>> "/tmp/test-home/.config/emacs/init".  So this always sets user-init-file
>> to the .el-less file name.  (Well, it would have even if the alternate
>> file name had ended in ".el".)
>> 
>> That makes sense for making a file name for `load' (so that it'll load
>> the .elc file, if it exists), but setting `user-init-file' to that
>> .el-less name sounds like the problem here?
> 
> I'm not arguing against the fact that there's a bug, I'm just saying
> that there's code which was supposed to handle this, and I think
> that's where this should be fixed.


I can’t find such a code.

I made a patch to set 'user-init-file' value to init”.el" for XDG convention users 
who do not have an init file.

(But alternate-filename-function is only used for "init.el” so 
                  ...(if (equal (file-name-extension alt-file)
				 "el")
			    (file-name-sans-extension alt-file)
                          alt-file)...
may be redundant.)

--
tsuucat