ROS backend depended on a mosquitto function, a stray copy-paste line ...

Unreliable delivery during handshake phase persists.

The major caveat for ROS is potential message loss right
after a Publisher is advertised. tcp_nodelay has been set
in the Subscribers and latching in the Publishers (resending
the last message on demand to late Subscribers).
These are not completely airtight fixes, though.

I combined the source files & added skipping whitespace & a C++ wrapper.
Original source: https://github.com/littlstar/b64.c

The C++ parser is very minimal and ignores section markers (in our
case there isn't really any need for different sections anyway).
Python expects one section [ipaaca] - the format of the files
is otherwise standard ConfigParser / ini:

[ipaaca]

transport.mqtt.host=localhost
transport.mqtt.port=1883

backend=ros

You can override any of those using an environment variable
e.g. IPAACA_TRANSPORT_MQTT_HOST for transport.mqtt.host

Requires some preparations of the lib dependencies;
prepackaged compiled ipaaca libs ought to be provided.

Mental note: MQTT backend (seen in Python) needs threaded event delivery

Some reusable refactoring into generic BackEnd possible

(still unconnected to actual MQTT in this commit!)

(Note: should be removed later / merged into ipaaca-tester-cpp).

Initial basic version for a 'multitool' console test battery.
For now, it implements a simple sniffer (with output similar
to ipaaca-iu-sniffer).
Planned: switching mode (sniffer, injector, modifier ...)
per command-line or interactively with keys; also implement
other language versions.

Caveat for sniffer function: resend_requests trigger ADDED event
twice (once for the sniffer "" scope, once for the hidden one).

(since RSB-Python only supports the unsigned version)

Recent addition of IntegerConverter in rsb0.14 seems flaky (it
seems Python will always send 64bit, and no conversion can be made),
-> now using 64bit types for transport everywhere revision numbers
are exchanged as an RPC result, i.e. all RemoteIU modifications
and ResendRequests (Also disabled our own legacy IntConverters).

Note: the ipaaca.proto Protobuf definition has not been changed,
for compatibility reasons. Java version contains explicit (int)
casts when protobuf objects are built. (TODO Discuss proto->64bit)

Tests between all three languages work (Netboot, rsb0.14,
protobuf2.6.1, spread4.4), tested with WOz+Cere+Billie.

Read access in locked payloads is performed on queued operations
before searching the old database. Threads other than the locking
thread consistently read the old information.

On POSIX, default Lock now also uses boost::thread (instead of pthread)
pthread linkage is still present for legacy wrapping.

ATTENTION: unlocked payload writing is (by design) not thread-safe
(won't crash but data might be inconsistent). Please use Locker armors
around all code that might modify a payload in more than one thread.
---> #1: Always use Locker in IU handlers that write to existing IUs!!

For string values, ONLY the empty string "" maps to false
(previously, "false", "False", and "0" also mapped to false)
For list and map values, zero-element ones now map to false
(previously, any list or map would map to true).

The new semantics are identical to Python behavior for your convenience,
since they were not generically prescribed by the STL anyway.

Also re-renamed the other two back to --rsb-host and --rsb-port ;)

!!! Please update the rsb package accordingly. !!!

End users need not change very much. (Please update rsc/rsb
version numbers in CMakeLists.txt in your own C++ projects).

- Some (internal) changes for rsb0.14 compatibility

Disabled removal of retracted IUs from InputBuffer to
match the Python version - since it is an item of discussion
that will be resolved in the next version anyway.

- OutputBuffer retracts all remaining live IUs on destruction
- sends correct event type now
- fixed reference error for calling handlers
- setters properly check for retracted flag
- also added some missing checks for committed flag

- OutputBuffer retracts all remaining live IUs on destruction
- sends correct event type now
- fixed reference error for calling handlers
- setters properly check for retracted flag
- also added some missing checks for committed flag