setup.py loses the DESTDIR aka --root iff "/" is passed as prefix


http://seclists.org/nmap-dev/2012/q3/1025

I can reproduce this now, but only with a prefix of "/". For example,
this works:

$ python setup.py install --prefix "/a" --root "/home/david/destdir"
copying build/scripts-2.7/ndiff -> /home/david/destdir/a/bin

But this doesn't:

$ python setup.py install --prefix "/" --root "/home/david/destdir"
copying build/scripts-2.7/ndiff -> /bin

This looks like a Python/distutils bug.


the same seems to happen on python install:
when installing python 2.7.2, "2to3", "idle", "pydoc" and "smptd.py" are installed into /bin rather than into the chosen destdir

./python -E ./setup.py install \
        --prefix=/ \
        --install-scripts=//bin \
        --install-platlib=//lib/python2.7/lib-dynload \
        --root=//opt/python/
[...]
running install
running build
running build_ext
INFO: Can't locate Tcl/Tk libs and/or headers


running install_scripts
copying build/scripts-2.7/smtpd.py -> /bin
copying build/scripts-2.7/idle -> /bin
copying build/scripts-2.7/pydoc -> /bin
copying build/scripts-2.7/2to3 -> /bin
changing mode of /bin/smtpd.py to 755
changing mode of /bin/idle to 755
changing mode of /bin/pydoc to 755
changing mode of /bin/2to3 to 755
running install_egg_info
Writing /lib/python2.7/lib-dynload/Python-2.7.2-py2.7.egg-info
if test -f //opt/python//bin/python -o -h //opt/python//bin/python; \
then rm -f //opt/python//bin/python; \
else true; \
fi
(cd //opt/python//bin; ln python2.7 python)
rm -f //opt/python//bin/python-config
(cd //opt/python//bin; ln -s python2.7-config python-config)
test -d //opt/python//lib/pkgconfig || /bin/install -c -d -m 755 //opt/python//lib/pkgconfig
rm -f //opt/python//lib/pkgconfig/python.pc
[snip]

python 2.7.2 installation:
CFLAGS="-D_GNU_SOURCE -D_BSD_SOURCE" ./configure -C --prefix="/" || exit 1
make -j9  || exit 1
make DESTDIR="//opt/python" install || exit 1

what python generates out of 
       --install-scripts=//bin \
       --install-platlib=//lib/python2.7/lib-dynload \
is both wrong (i.e. the platlib stuff also ignores DESTDIR passed to make)

Thanks for the report.  This bug was already known, but I haven’t had the time yet to fix it.  I’m adding you to the nosy list on the other bug so that you can follow progress and maybe help testing the fix.