BGP remote-as for a peer-group is now configured before adding peers
(as this would otherwise generate an error and the neighbor not added)

this was the case for peers inside a VRF but not in the global routing
table; this fix ensures that the remote-as is now configured before
adding neighbors, in the same way

one file per protocol or command family as the single file was getting
too long

the unit tests depend on deepops>=1.8 to pass, so this updates that as
a dependency

also, the conversion debug messages have been tidied a little

version 1.8 changes what happens when identical data structures are
compared, returning an empty object of the same type, rather than None

this probably doesn't affect anything that actually uses asimtote but,
without this, some of the unit tests fail and that's confusing!

removed 'match' to make messages shorter

rearranged trigger message to put match first and block name afterwards
in square brackets (to make more consistent with other messages)

this should have no effect on functionality

tidies the names of lots of internal classes and makes the call of
inherited context enter() methods better

better highlights abstract converters (_AbsCvt_xxx) and contexts

better use of split arguments (not assuming parent/context class
argument lists) and calls

context converters are now called Context_xxx

abstract command converters are now called _AbsCvt_xxx

still some tiyding to do against route-maps

this includes the name of abstract context classes based on it, which
will be tidied in a future commit

internal reorganisations that should have no effect on actual
functionality but significant enough to warrant a version change

converter methods now receive the context wildcard arguments as a
single leading list, rather than being supplied discretely; this
argument is usually just passed to an enter() method and avoids needing
to specify them explicitly

it also avoids the need to explicitly call split_args() (which has been,
removed), if the boundary between the context and local arguments has
adjusted or is otherwise dynamic (such as when a converter class is
reused) - the split_args_offset attribute has been renamed
context_offset

get_sort_key() replaced by sort_key attribute

some correcting of class names

this was previously used by the split_args() method but that has been
removed, in favour of it being done by get_args() and the context being
supplied to all the converters by default, so the name was not
particularly accurate

merged into the get_args() method as there is no reason for converters
to call split_args() themselves manually, now (as the context arguments
are supplied as a separate parameter to them)

the call of methods like add(), update(), delete(), truncate() and
remove(), as well as the filter() methods, now have an extra argument
'context_args' supplied, which is a list of all the wildcard arguments
from the context portion

the individual arguments from the command and extension portions are
still supplied individually, if present

this avoids the need to explicitly name the context arguments and
(usually) supply them to the inherited enter() method individually, as
the simple argument can just be passed through (typically by expanding
it with *c)

it also avoids the need the call split_args() explicitly in a method as
this will be handled by the caller; split_args_offset is still
respected in this split

one downside is the need to receive a context argument even if it's
empty (at the top level)

there needs to be some tidying of comments and other neatening changes
(various TODOs have been added) but this commit marks the state when
the test suite runs successfully

these had 'Track_' at the beginning, erroneously

there doesn't seem any need for this to be determined dynamically
(requiring a function)

renamed internal method _get_full_sort_key() method to
_get_sort_key()

added 'router bgp ...' converters and unit tests

at the point where these were added, the live configuration on the
network compared with the generated configuration results in no
differences

renamed get_sorting_key() to get_sort_key() and made it only return the
command + extension fields

the actual sorting key is now retrieved using a new
_get_full_sort_key() method, which prepends the context and replaces
the None fields with the empty string

this makes it easier to override the sorting key (avoiding the need to
specify the full path)

previously only 'update' and 'remove' were shown but now the test for
'add' and 'truncate' is run first and the message adjusted, if it's
actually one of those, giving a clearer message

and added an 'all prefixes' rule test for each

in command parsing and converter classes

no bugs or other corrections required for the converters so no need for
a version bump

better explain the consequences of this handling

there aren't standard and extended ACLs for IPv6

as with v0.16.2, this version fixes a lot of bugs and improves things
discovered whilst writing unit tests for the converters, as well as
adding some missing commands and internal functionality

fix: 'router ospf[v3] / area ... nssa' option removal

fix: 'route-map / set ip next-hop verify-availability' would crash if
removed

fix: 'route-map / set community' did not handle the 'additive' option

fix: 'route-map / set community' did not correctly include the set of
communities but the literal word 'communities' instead

fix: 'route-map / match ...' converters now return changes in the order
of the things in the list (e.g. if using tags, these will now be in
numeric order) - this doesn't break anything but made the output
indeterminate for writing tests

fix: 'interface / ipv6 nd prefix ...' now re-added after an interface
'vrf forwarding' change

fix: 'interface / standby ... ip ...' now correctly handles promoting a
second address to the primary (the secondary must be removed first) or
demoting the primary to a secondary (the primary must be removed first)

fix: 'ip[v6] route ...' now correctly handle if either of the interface
or router/gateway addresses are missing (these are now not included in
the parsing configuration dictionary, instead of being present but
None, which was inconsistent with other omitted options)

add: tests for VLAN and subinterface creation

removed: 'set ipv6 next-hop verify-availability [...]' - this command
was handled but doesn't exist!

improved update to 'set vrf' in route-map: the old VRF is no longer
removed (unless changing to and from 'global')

these test all the conversions done in the Cisco IOS converters
'router' module

same fix as above for OSPF for IPv4 but for OSPFv3

if options like 'no-summary' were removed, this would not be handled
but now the command is re-entered without them to do this

no longer clears the old VRF but just sets the new one (if shifting
between 'set vrf' and 'set global', it still clears the old setting,
though)

this command (and the tracking version of it) don't exist (at least in
the versions of IOS we're using) so have removed them: the parsers,
converters and parsing unit tests

previously the converter did not match the '...-track' areas of the
configuration differences, causing changes to be ignored

previously set_debug() accepted positional arguments but this meant
sometimes the wrong debugging option was enabled - this forces all
arguments to be keyword-only so we can make sure the correct debugging
option is enabled

when removing 'set ip next-hop verify-availability' option, incorrectly
used 'new' state to determine sequence number rather than 'old' state

did not handle 'additive' option (although a TODO on handling adding
and removing it from a list that's otherwise the same)

also, the community list was not correctly fetched but the literal
string of 'communities' used as the list

these work from sets of community lists, access lists, prefix lists,
etc. and so the order was indeterminate - the converters now sort the
sets before working on them to give a consistent order

(this didn't typically break conversion but does make unit test output
more predictable)

expanded the unit test for VRF add/remove/update to confirm that this,
along with other commands like HSRP configuration, are re-applied after
the VRF change

unlike changing the 'ip address' on an interface, configuring the
primary address for HSRP does not remove it as a secondary, if it is
already configured there, but generates an error

the same thing applies in reverse (you can't configure a secondary HSRP
address if it's currently in use as the primary, although the ordering
didn't tend to trigger that)

this change makes sure that the old addresses are removed before the
changes are made

added unit tests to expose this issue (and confirm it's fixed)