Hi Julien,
https://lists.gnu.org/archive/html/automake-patches/2018-03/index.html
Thanks for bringing them up again. I wasn't involved with automake back
then, but looking at them now, they seem mostly fine. The one that seems
too risky is making -c do something (preserve permissions) instead of
being a no-op. I think there could easily be ancient calls of install-sh
that use -c, and changing behavior now doesn't seem good. We could
invent a new option (-p I suppose) to do it.
The one that omits chown-ing of pre-existing directories also worries me
a bit (the one below). I can imagine a scenario where people installed
as root and chowned to some specific user. OTOH, I agree it seems
cleaner (not to touch existing directories.
Over the next days, I'll try installing these changes (one at a time)
and making sure the test suite still runs ok. Will keep you posted.
Thanks again. --karl
--- install-sh.orig 2018-03-17 14:18:08.701045461 +0100
+++ install-sh.useless-chown 2018-03-17 14:29:16.924781085 +0100
@@ -255,6 +255,10 @@
dstdir=$dst
test -d "$dstdir"
dstdir_status=$?
+ # Don't chown directories that already exist.
+ if test $dstdir_status = 0; then
+ chowncmd=""
+ fi
else
# Waiting for this to be detected by the "$cpprog $src $dsttmp" command
