Now, in a testenv, if you forget to use '-s st/ad_dc/etc/smb.conf',
you only see this:

$ bin/samba-tool user rename  dsadsa
ldb: Unable to open tdb '$HERE/st/client/private/secrets.ldb': No such file or directory
ldb: Failed to connect to '$HERE/st/client/private/secrets.ldb' with backend 'tdb': Unable to open tdb '$HERE/st/client/private/secrets.ldb': No such file or directory
Could not find machine account in secrets database: Failed to fetch machine account password from secrets.ldb: Could not open secrets.ldb and failed to open $HERE/st/client/private/secrets.tdb: NT_STATUS_CANT_ACCESS_DOMAIN_INFO
ltdb: tdb($HERE/st/client/private/sam.ldb): tdb_open_ex: could not open file $HERE/st/client/private/sam.ldb: No such file or directory

Unable to open tdb '$HERE/st/client/private/sam.ldb': No such file or directory
Failed to connect to 'tdb://$HERE/st/client/private/sam.ldb' with backend 'tdb': Unable to open tdb '$HERE/st/client/private/sam.ldb': No such file or directory
ERROR(ldb): uncaught exception - Unable to open tdb '$HERE/st/client/private/sam.ldb': No such file or directory

rather than all that AND a stack trace.

Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>

Often we'll write something like

   ERROR: Unable to find user "potato"

which can get lost in the jumble of other output. With this patch, we
colour the word "ERROR" red but not the rest of the string, unless it is
determined that colour is not wanted (due to one of --color=never,
NO_COLOR=1, output is not a tty).

We choose to redden the word "ERROR" only to maintain legibility in the
actual message, while hopefully increasing the noticeability of the line.

Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>

More often than not we are using colour in stderr, but are deciding
based on stdout's tty-ness. This patch changes to use both, and will
affect the following situation:

 samba-tool  2>/tmp/errors   # used to be colour, now not.

of course, if you want colour, you can always

 samba-tool --color=yes 2>/tmp/errors

Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>

In particular, visualize needs it to decide colour for an output
file that may or may not be stdout, so it needs to make its own
decision for that file.

Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>

We don't put --color into options.SambaOptions because we can't handle
the 'auto' case in the options module without knowing whether or not
self.outf is a tty, and a) this might not be resolved and b) is fiddly
to pass through.

The .use_colour class flag allows samba-tool subcommands to avoid having
--color, and is *also* useful in the short term for visualise and drs
commands to avoid having this --color clobber their own bespoke versions
(temporarily, during the transition).

Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>

Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>

We catch output in outf and errf for testing, which we currently do
with

    cmd.outf = self.stringIO()
    cmd.errf = self.stringIO()

on the final resolved commands. But this does not catch the output of
the super-commands, of which we normally expect none. Using

    supercmd._resolve(*args, outf=self.stringIO(), errf=self.stringIO())

will redirect output all the way up the chain.

Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>

By default parse_args will use sys.argv[1:], which is to say the
command-line without the command name. We have always fed it the
equivalent of sys.argv, then trimmed the command off the result. That
was a bit silly.

Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>

Prior to this commit, in super-commands, the first half of the _run()
is resolving what sub-command to run, and the second half is working
out what to print if that failed. Some issues with that are:

 * it looks a little bit complicated.

 * the tests can't use the tool's resolution code, because it runs
   immediately, while the tests first want to fiddle with self.outf
   and so on.

 * it makes it harder to subclass and override the resolution code, so
   instead we do strange things like where we subclass dict as in
   main.py.

So we split it into ._resolve() and ._run().

There are a few tests that break. We mark these as flapping, rather
than knownfail, so as to avoid going into extremely fine-grain filters
for tests that will be fixed within a few commits.

Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>

The run() method is always called with arguments, so it crashes before
the NotImplementedError() is ever reached. That's OK, but this is better.

Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>