Problem: test_security_gssapi expects wrong event

Solution: do it to avoid leaks. Even though they are below 33 bytes so
it's all allocated on the stack, the small message size might change in
the future.

Solution: fix it

Problem: test_sockopt_hwm fails occasionally

Solution: relaxed assertion to accept 10% of the SNDHWM, and updated documentation accordingly

Solution: align assertions with specification of ZMQ_SNDHWM (actual number of enqueueable messages may be lower)

Problem: RPM -devel still ships libzmq.la

Solution: mention recent changes

Solution: don't. libtool's archive files are already being deprecated
in Debian-based distros. They are no longer necessary, and cause the
reverse dependencies to over-link.
EG: when built with sodium/pgm/krb5, all applications/libraries
linking against libzmq would be forced to link against sodium/pgm/krb5
despite not using any of those libraries symbols.

Problems: GSSAPI builds broken with picky compilers, Debian/RPM packages do not use GSSAPI, no CI jobs that test GSSAPI

Solution: enable option and install build-dependency for the jobs that
already install packages only, to minimise marginal cost in term of
runtime

Solution: enable option and install build-dependency for the jobs that
already install packages only, to minimise marginal cost in term of
runtime

Solution: fix it

Solution: add build-dependency and configure option

Solution: cast from void* to uint8_t* to match new definition

Solution: initialise session and peer_address

Solution: pass the session as well as the options

Solution: remove it from the gssapi_mechanism_base initialiser

Solution: remove local unused zap_client variable

Solution: move them together with the other DRAFT socket options, and
change value of DRAFT ZMQ_BINDTODEVICE from 90 to 92 to avoid clash

Solution: include mechanism_base.hpp instead of mechanism.hpp

Solution: call make_command_with_basic_properties

Problems: free of stack variable, TODO left to evaluate

Solution: add missing target_link_libraries
Fixes #2701

Solution: use LIBZMQ_UNUSED where necessary

Solution: add it to CMake's platform.hpp.in

Solution: as discussed, remove the deprecation notices, as many users
expressed the need for multi-part support.
Fixes #2699

Solution: monitor new events only if DRAFT APIs are enabled, and
convert to new event types. Same for DRAFT socket options.

Solution: remove the second one to fix build failure

Solution: use it before falling back to headers checks

Solution: remove it.
Looking at the code:

https://github.com/krb5/krb5/blob/master/src/lib/gssapi/mechglue/g_unseal.c#L55

gss_unwrap as the very first thing checks that plaintext is not a
null pointer, which in our case it's true given it's on the stack,
and then initialises its members to 0 length and null ptr.

https://github.com/krb5/krb5/blob/master/src/lib/gssapi/mechglue/g_rel_buffer.c#L36

So it should be safe to release it in all cases, and the release API
seems to check again if it's not a null pointer and then if the
members are 0 length and null pointer it's a no-op.

Solution: free wrapper.value instead of wrapper.

Problem: missing files from VS2008 solution

Problem: gssapi_mechanism_base_t aborts the application when it receives a message it cannot decode

Solution: generate list of sources and headers with:

 ls -1 src/*cpp | sort | sed 's|src/\(.*\)|      <File RelativePath="..\\..\\..\\..\\src\\\1" />|g'
 ls -1 src/*hpp | sort | sed 's|src/\(.*\)|      <File RelativePath="..\\..\\..\\..\\src\\\1" />|g'

and add them.

Solution: emit an ZMQ_EVENT_HANDSHAKE_FAILED_PROTOCOL event instead of using zmq_assert

Problem: Values in ZMQ_EVENT_HANDSHAKE_FAILED_* are not helpful

Solution: extracted common base class curve_mechanism_base_t

Solution: emit appropriate events

Solution: emit events at appropriate places