[zeromq-dev] czmqpp

[john skaller]

On 26/01/2012, at 6:26 AM, Yi Ding wrote:

> I remember Martin or Pieter saying somewhere that zmq.hpp (cppzmq) wasn't really being maintained these days and wasn't super useful to begin with.
> 
> It seems to me that czmq is a much more useful binding, and the only thing it really lacks for C++ programming is RAII.

That's a good thing. RAII (except locally) is a very bad paradigm.
In particular it does not play well with any kind of automated memory management,
especially garbage collection***

Local (on stack) RAII is ok (and perhaps even necessary do the presence
of evil exceptions).

** All advanced systems are necessarily garbage collected, even if they're written
in C or C++. That's because there is no other general strategy for managing
graphs. Of course all advanced networking applications have graph topologies
and involve external resources (sockets, nodes, OS stuff, even hardware).

RAII in these circumstances is utterly evil. Don't go there.
Management of the representative objects does NOT reflect the proper
management of resources: the sequence of assembly and release
of the representative and the actual resources not only may be distinct,
it almost invariable is necessarily distinct.

For example you may construct a topological representation of a network
in some order, and then activate it node by node in a quite different order.

If you have a look at ZMQ itself, which is written in C++, you'll note that
the primary classes are not initialised by constructors, but separate
init functions. After all .. a message, for example, can be initialised
multiple times in its lifetime (I actually think this is a bit over the top,
its very error prone .. I'm not allowing that in the Felix binding .. BUT
it has one advantage: maximal efficiency).

--
john skaller
skaller at users.sourceforge.net