Remove installed packages before pip install

The release of pip10 has shown up a few issues here

Firstly, pip10 now refuses to overwrite distutils installed packages,
which includes "python-virtualenv" on centos.  History has shown us
that we want the packages installed and overwritten, to avoid the
packages coming back and messing things up.

Pre-install all the packages, then list the files in the packages with
"rpm" directly and remove them.  This way pip is happy to install.

We need to take better account of the package names for this; on
Fedora things have switch to "python2-virtualenv" instead of
"python-virtualenv" and we can't use an alias to list the package
contents.

This also highlighted that python2-pip is in EPEL for centos, so
enable that when we install it.  Make the epel element a no-op for non
centos/rhe distros.

There is a related change in recent fedora that python3 now installs
binaries into /usr/local/bin.  There are commented swizzles in here to
ensure we retain the status quo of "pip" and "virtualenv" both being
python2 based, with the python3 versions being called explicitly
"pip3" and "virtualenv3" respectively.

Change-Id: I2ffdd9f615ae6b00428c17249e4f216774991b99

diskimage_builder/elements/epel/README.rst

@@ -5,7 +5,7 @@ epel
 This element installs the Extra Packages for Enterprise Linux (EPEL)
 repository GPG key as well as configuration for yum.
 
-Note this element only works with platforms that have EPEL support
+Note this element is only useful with platforms that have EPEL support
 such as CentOS and RHEL
 
 DIB_EPEL_MIRROR:

diskimage_builder/elements/epel/pkg-map

@@ -0,0 +1,16 @@
+{
+ "distro": {
+    "centos7": {
+       "wget": "wget",
+       "yum-utils": "yum-utils"
+    },
+    "rhel7": {
+       "wget": "wget",
+       "yum-utils": "yum-utils"
+    }
+ },
+ "default": {
+   "wget": "",
+   "yum-utils": ""
+ }
+}

diskimage_builder/elements/epel/pre-install.d/05-rpm-epel-release

@@ -25,7 +25,8 @@ else
             ;;
         *)
             echo "$DISTRO_NAME is not supported"
-            exit 1
+            # Not really a failure; we just don't do anything
+            exit 0
             ;;
     esac
     PKG_NAME=$(wget -q $URL -O - |grep -oE "(href=\"epel-release-$RELEASE-[0-9,.].*)" | cut -d'"' -f2)

diskimage_builder/elements/pip-and-virtualenv/element-deps

@@ -1,4 +1,5 @@
 dib-python
+epel
 install-types
 package-installs
 source-repositories

diskimage_builder/elements/pip-and-virtualenv/install.d/pip-and-virtualenv-source-install/04-install-pip

@@ -8,12 +8,31 @@ set -o pipefail
 
 if [[ $DISTRO_NAME =~ (opensuse|fedora|centos|centos7|rhel7) ]]; then
 
+    # Default packages
     _do_py3=0
-    packages="python-virtualenv python-pip python-setuptools"
+    _extra_repo=''
-    if [[ $DISTRO_NAME =~ (fedora) ]]; then
+    case "$DISTRO_NAME" in
+        centos*|rhel7)
+            # note python2-pip in epel
+            _extra_repo="--enablerepo=epel"
+            packages="python-virtualenv python2-pip python-setuptools"
+            ;;
+        fedora)
             _do_py3=1
+            packages="python2-virtualenv python2-pip python2-setuptools"
             packages+=" python3-virtualenv python3-pip python3-setuptools"
-    fi
+            ;;
+        opensuse)
+            case "$DIB_RELEASE" in
+                42*)
+                    packages="python-virtualenv python-pip python-setuptools"
+                    ;;
+                tumbleweed)
+                    # XXX: python3?
+                    packages="python2-virtualenv python2-pip python2-setuptools"
+                    ;;
+            esac
+    esac
 
     # force things to happen so our assumptions hold
     pip_args="-U --force-reinstall"
@@ -30,36 +49,49 @@ if [[ $DISTRO_NAME =~ (opensuse|fedora|centos|centos7|rhel7) ]]; then
     if [[ $DISTRO_NAME = opensuse ]]; then
         zypper -n install $packages
     else
-        ${YUM:-yum} install -y $packages
+        ${YUM:-yum} ${_extra_repo} install -y $packages
     fi
 
+    # pip10 (unlike earlier versions) will not uninstall distutils
+    # installed packages (note this is only a subset of packages that
+    # don't use setuptools for various reasons).  We give it a little
+    # help by clearing out the files from the packages we are about to
+    # re-install so pip doesn't think anything is installed.  See:
+    # https://github.com/pypa/pip/issues/4805
+    for pkg in $packages; do
+        rpm -ql $pkg | xargs rm -rf
+    done
+
     # install the latest python2 pip; this overwrites packaged pip
     python /tmp/get-pip.py ${pip_args}
 
-    # pip and setuptools are closely related; we want to ensure the
+    # Install latest setuptools; there is a slight chicken-egg issue in
-    # latest for sanity.  Because distro packages don't include enough
+    # that pip requires setuptools for some operations like building a
-    # info in the egg for pip to be certain it has fully uninstalled
+    # wheel.  But this simple install should be fine.
-    # the old package, for safety we clear it out by hand (this seems
-    # to have been a problem with very old to new updates,
-    # e.g. centos6 to current-era, but less so for smaller jumps).
-    # There is a bit of chicken-and-egg problem with pip in that it
-    # requires setuptools for some operations, such as wheel creation.
-    # But just installing setuptools shouldn't require setuptools
-    # itself, so we are safe for this small section.
-    rm -rf /usr/lib/python2.7/site-packages/setuptools*
     pip install ${pip_args} setuptools
 
     if [[ $_do_py3 -eq 1 ]]; then
         # Repeat above for python3
+
+        # python2 on fedora always installs into /usr/bin.  Move pip2
+        # binary out, as we want "pip" in the final image to be
+        # python2 for historical reasons.
+        mv /usr/bin/pip /usr/bin/pip2
+
         # You would think that installing python3 bits first, then
         # python2 would work -- alas get-pip.py doesn't seem to leave
         # python3 alone:
         #  https://github.com/pypa/pip/issues/4435
         python3 /tmp/get-pip.py ${pip_args}
-        rm -rf /usr/lib/python3.?/site-packages/setuptools*
         pip3 install ${pip_args} setuptools
-        # reclaim /usr/bin/pip back to pip2
+
+        # on < 27, this installed pip3 to /usr/bin/pip.  On >=27 it's
+        # /usr/local/bin/pip.  reclaim /usr/bin/pip back to pip2 and
+        # remove the /usr/local/bin/pip (i.e. python3 version) if it
+        # exists, so that "pip" calls pip2 always.  if we want pip3 we
+        # call it explicitly.
         ln -sf /usr/bin/pip2 /usr/bin/pip
+        rm -f /usr/local/bin/pip
     fi
 
     # now install latest virtualenv.  it vendors stuff it needs so
@@ -67,14 +99,27 @@ if [[ $DISTRO_NAME =~ (opensuse|fedora|centos|centos7|rhel7) ]]; then
 
     # python[2|3]-virtualenv package has installed versioned scripts
     # (/usr/bin/virtualenv-[2|3]) but upstream does not!  (see [2]).
-    # For consistency, clear them out and then reinstall so we're just
+    # For consistency, reinstall so we're just left with python2's
-    # left with python2's version
+    # version.  Note this is a rather moot point, the usual way we get
-    # [2] http://pkgs.fedoraproject.org/cgit/rpms/python-virtualenv.git/tree/python-virtualenv.spec#n116)
+    # a python3 environment is to call "virtualenv -p python3 foo" and
-    rm /usr/bin/virtualenv*
+    # that works to create a python3 virtualenv, even if using
+    # python2's version.  Thus we probably don't *really* need to
+    # "pip3 install virtualenv".  What we don't want is "virtualenv
+    # foo" creating a python3 virtualenv by default, because that
+    # confuses a lot of legacy code.
+    #
+    #[2] http://pkgs.fedoraproject.org/cgit/rpms/python-virtualenv.git/tree/python-virtualenv.spec#n116)
+    pip install ${pip_args} virtualenv
+    mv /usr/bin/virtualenv /usr/bin/virtualenv2
     if [[ $_do_py3 -eq 1 ]]; then
         pip3 install ${pip_args} virtualenv
     fi
-    pip install ${pip_args} virtualenv
+
+    # Reclaim virtualenv to virtualenv2; similar to above, on fedora
+    # >27 the pip3 version has gone into /usr/local/bin; remove it so
+    # only /usr/bin/virtualenv exists
+    ln -sf /usr/bin/virtualenv2 /usr/bin/virtualenv
+    rm -f /usr/local/bin/virtualenv
 
     # at this point, we should have the latest
     # pip/setuptools/virtualenv packages for python2 & 3, and