fixed: make -O3 option private

See merge request !4

The special wireguard interface has its ifa_addr pointer set to
null, leading to a crash in NameService::publish() which iterates
over all interfaces, checking ifa_addr->family.

Protection against ifa_addr == nullptr were added in various places.

License

See merge request !2

This is an artitifical dependency on a barely used feature of
'transport'. Better to move it here and cut the ties.

Compilation of all direct clients in work area works.

Master

See merge request !1

In addition, if there is already an entry for this MC address, but with
a different partition name, the configuration will fail.

The OKS value of the multicast address is allowed to be * instead of
a real IP address. In this case the following procedure is used to allocate a
unique MC address across all partitions. The network/netmask part of the string
should still be defined, e.g. */10.149.0.0/255.255.0.0.

The allocate_multicast_address() method of NameService should be called by a
single application in the partition (typically the HLT supervisor). It will
first iterate over all existing IS objects of type 'MultiCastGroup' in the
RunParams IS server of the initial partition. If it finds an entry which
matches its own partition, the multicast address from this object will be
used. This means that MC groups will be stable as long as the initial partition
is up and running.

If the method does not find an entry, it will try to insert a new object
which contains the multicast address as a string in its name. This guarantees
the uniqueness of the object and therefore address. The method will loop over
a range of pre-defined multicast address until insertion succeeds. If the range
is exhausted an exception CannotAllocateMulticast is thrown.

As a final step either the static or dynamically allocated address and network
are published in the local partition in all DFConfig IS servers as an object
with the fixed name 'MultiCastAddress'.

The lookup_multicast_address() method simply looks up the latter object in its
appropriate DFConfig IS server and returns it.

Note that the IS object contains both the partition name and the network part, but
the latter should always be taken straight from OKS for purposes of configuration.
The IS object simply contains it as auxilliary information.

This is a hack to solve a P1-specific routing problem. See comment for
more details.

TCP keep-alive probes are globally controlled by an environment variable
called
TDAQ_ASYNCMSG_KEEPALIVE, with the the following format:
<enable>:<timeout_s>:<interval_s>:<max_failures>
where:
- <enable> (boolean, default: false)
  enables or disables TCP keep-alive probes
- <idle_s> (integer, default: /proc/sys/net/ipv4/tcp_keepalive_time)
  is the number of seconds a connection needs to be idle before TCP begins
  sending out keep-alive probes
- <interval_s> (integer, default: /proc/sys/net/ipv4/tcp_keepalive_intvl)
  is the number of seconds between TCP keep-alive probes
- <max_failures> (integer, default: /proc/sys/net/ipv4/tcp_keepalive_probes)
  is the maximum number of TCP keep-alive probes to send before giving up
  and killing the connection if no response is obtained

This lets exceptions thrown in derived classes to fall through to the
io_service::run() method invokation.

WARNING: asyncmsg objects one of whose completion methods has thrown an
exception may be in an inconsistent state afterwards.

All completion methods must be invoked in a manner equivalent to using
boost::asio::io_service::post(), instead of executing them
immediately. asyncClose() violated this guarantee if the Session had no
outstanding operations.

change default name server; add two more lookup functions for retrieving the name of the application as well as the endpoint

Fixes ATDAQDF-47

Fixes ATDAQDF-46