The pg_upgrade binary is not in postgresql-upgrade package now.
In some upgrade scenarios, postgresql-upgrade is redundant package
(SCLs/PGRPMSs).  Hack the code so postgresql-setup is able to
correctly advice which package should be installed to make the
"old" engine available.

* postgresql-setup.in (upgrade): Don't check for pg_upgrade binary
existence but rather for postgres binary, and warn user with
proper package name if missing.
* share/postgresql-setup/library.sh.in (parse_upgrade_setup):
White-list also 'package' constant name.
* configure.ac: Bump version.

* configure.ac: Version bump.
* NEWS: Document.

Resolves: rhbz#1247477

* configure.ac (MOUNTPOINT): New configure-time variable.
* postgresql-setup.in: Warn if $pgdata or `dirname $pgdata` is
a mountpoint.
* NEWS: Document.
* README: Document that its better to have util-linux in BR.

We can't always use '$upgradefrom_id' as "default" previous
service name.  This broken in-place upgrades with
'--unit postgresqlPORT' (old style service files layout) or
'postgresql@port'.  Now, by default, use the same service name
for both older and newer PG stack.

While we are on it, make it a bit more automatic and pick
'postgresql92@second' as "older stack" if upgrading into unit
'postgresql93@second'.

Add new option --upgrade-from-unit which allows us to override
default decision.

* postgresql-setup.in: Set 'option_upgradefrom_unit' to hopefully
correct default.  Add new info message in case we are not
upgrading from differently named unit.
(USAGE_STRING): Document new --upgrade-from-unit option.
(service_configuration): Don't fork based on $mode and simply use
fourth parameter as 'service'.  Fix the debug message a bit.
(long_opts): Add new option upgrade-from-unit.
* NEWS: Document.

This commit should resolve complaints against pg_ctl -w usage from
initscript (see rhbz#800534 for more info).

The problem is that pg_ctl -w can block whole system startup too
long - until the server is fully ready to accept connections.  The
commit 8c7b2cd5f6d9efb59568382cd8b6e88d9be517bb thus invented
basically bad approach for initscript.

Now, the "-w" option is used only if the PGSTARTWAIT is explicitly
set to 1 by administrator.  In this case, the PGSTARTTIMEOUT
numeric value (in seconds) is respected (or limit 30 seconds is
set by default).

Otherwise, if PGSTARTWAIT is unset (default), initscript keeps
checking only for the pidfile existence (at most PGSTARTTIMEOUT
seconds).  The hardwired 'sleep 2' command (activated before
8c7b2cd5f6 commit) is still not used - regular/default system
startup should be a bit faster than before.

* postgresql.init.in (start): Fix handling of PGDATA, PGPORT,
PGOPTS and PGSCLS - to respect special quoting characters inside,
we should escape the strings before passing those to command
evaluation.  Use printf %q as we use bash and we don't care too
much about portability.  Also, don't call postgresql-ctl with -w
option if PGSTARTWAIT is empty or undefined.  Implement pidfile
polling for the cases PGSTARTWAIT is unset.  Return success faster
if server is already running.
($PGSTARTWAIT): New env variable.
* postgresql-ctl.in: Don't parse $PGPORT into -o option as it is
not needed.  Unset PGPORT if the variable is empty - postgres
server would fail with empty value.
* configure.ac: Bump version to 3.3.
* NEWS: Document bugfixes.

Do not left created dropin directory in case of some initdb
failure, or so.  This allow user run the same --new-systemd-unit
repeatedly without changing the command-line.

* postgresql-setup.in (exit_handler): New exit callback wrapper.
($cleanup_dropin): Code to be executed to clean drop-in directory.
(generate_systemd_dropin): Fill $cleanup_dropin callback.
(initdb): Clean $cleanup_dropin in case of successful initdb.
This is not needed for upgrade() because we never create drop-ins
there.

* NEWS: Document changes.

* postgresql-setup.in (service_configuration): Move the setting of
pgdata into sysvinit-only branch.
* NEWS: Document fixed bug.
* configure.ac: Bump version to 3.2.

* postgresql-setup.in (upgrade): Call old_data_in_use to ensure
that the old database server is not running.
(old_data_in_use): New function.
($cleaup): Rename to $failure_cleanup.
* NEWS: Document.