Re: master fails to build on FreeBSD when ACL support is on

[Glenn Morris <rgm@gnu.org>]

IMO the change in 4fd446e is obviously correct.
The current (reverted) code is wrong by inspection when target-file is
relative. In normal use, it never is, but it will be if
byte-compile-dest-file is changed to return a non-absolute value.

For example, this can be the case for automake, as we saw in

http://lists.gnu.org/archive/html/emacs-devel/2017-11/msg00551.html

since Makefiles normally use relative filenames, and automake changes
byte-compile-dest-file-function to return "$@".

Eg I predict the following will fail on a FreeBSD system like the one in
the original message in this thread:

./src/emacs --batch \
 --eval '(setq byte-compile-dest-file-function (lambda (x) "foo.elc"))'
  -f batch-byte-compile foo.el

This simulates how automake may call Emacs to compile a file.

(It fails on RHEL 7.4 with TMPDIR=/does/not/exist. Obviously this is not
sensible, but the point is that it should not be writing to TMPDIR at all.)


I am posting this analysis for completeness, with zero expectation of it
changing anything.