Hello Paul, First, thanks for the time you spend on this and sorry for the delay in my response. Paul Eggert writes: > I can't reproduce the problem on my Fedora 29 x86-64 platform. My > build log is attached (compressed). One difference is that I didn't > use the configure argument "--host=$(rpm -E > '%{_arch}-fc%{fedora}-%{_host_os}%{_gnu}')", which caused ar, ranlib, > and pkg-config to not be found on your platform. What happens if you > omit the --host argument? Nothing change. I've tried many variant : - with only --prefix=/usr/local - without anything - without --host=... After each try, this fail with the same error. > > Also, the first command discrepancy is that my build log says "GEN > ../../../etc/charsets/8859-2.map" and similarly for other *.map files, > whereas yours does not, and there are other places where it appears > that your directory contains files that are already built and may have > the wrong contents. I suggest looking into that. I've also done some experiments : - in a directory on the side of my emacs-26 source: everything work as expected. - after deleting the build-26 directory and re-applyed the following commands: 1) git clean -dfx 2) ./autogen.sh 3) mkdir build-26 && cd build-26 3) ../configure with the same options as yours and this doesn't build. FYI I've attached my traces and also 2 diffs: - one after the configure step between my two build-26 dirs (the one beside of the emacs-26 directory and the other inside). -- emacs-26.outoftree.diff - one between your traces and mine for the build inside emacs-26/build-26 (for this I've stripped off the paths before emacs-26. eg. removed /home/...) -- emacs-26.outoftree.diff.eggbert.diff For the second diff I've found that the my temacs dump has lesser size than yours and of course the "Nothing to be done for 'compile-first'." for 'emacs-26/build-26/lisp' directory I'm felling like if I've jumped inside the rabbit's hole of Alice. If you see something that make sense to you ...