- Extend the multipart API to allow sending vectors of messages or helper
thin wrapper FairMQParts. See example in examples/MQ/8-multipart.
- NewMessage() can be used in devices instead of
fTransportFactory->CreateMessage().
Possible arguments remain unchanged (no args, size or data+size).
- Send()/Receive() methods can be used in devices instead of
fChannels.at("chan").at(i).Send()/Receive():
Send(msg, "chan", i = 0), Receive(msg, "chan", i = 0).
- Use the new methods in MQ examples and tests.
- No breaking changes, but FAIRMQ_INTERFACE_VERSION is incremented to 3
to allow to check for new methods.
- Remove the compile time check of the transport implementation.
The transport (zeromq/nanomsg) can be chosen at run time with:
`device.SetTransport("zeromq"); // possible values are "zeromq" and "nanomsg"`.
For devices that use FairMQProgOptions, the transport can be configured via cmd option:
`--transport zeromq` or `--transport nanomsg`. Default values is "zeromq".
The device receives the configured value with:
`device.SetTransport(config.GetValue<std::string>("transport"));`
Old method of setting transport still works. But the NANOMSG constant is not defined.
- Remove old `fairmq/prototest` directory. It was only used as a test for protobuf.
The protobuf part of Tutorial3 does the same (with different values).
- Fix a bug in FairMQPollerNN, where the `revents` value was not initialized.
This caused the `poller->CheckOutput()` to trigger when it should not.
Both FairMQLogger and FairMQStateMachine use Boost Fusion internaly.
Because the state machine needs FUSION_MAX_VECTOR_SIZE 20 defined,
same value has to be defined in the logger, otherwise Fusion is not
properly compiled.
a) move the XML parser into the FairMQ/options/FairMQParser.h
b) add a routine in FairMQProgOption to check whether the necessary XML or JSON input files are there, and send an error message if not there
- Policy based devices:
a) rename GenericSampler to base_GenericSampler and use an alias template named GenericSampler
b) in base_GenericSampler, rename template parameter to simple variables <T,U,… > and use typedef for clarity
c) introduce an anonymous function container in the base_GenericSampler host class with a register task template member function and an Executetasks()
d) add two new template parameters in base_GenericSampler for the anonymous function container map. parameter is K for the key type (default=int) and L for the value type (default=std::function<void()>)
- Tutorial7:
a) use FairMQProgOption to configure devices in tutorial7
b) introduce several template functions helper in tutorial7 to reduce code redundancy
c) show examples in tutorial7 of task registration with callback and lambda expression for the sampler devices
d) separate the executable build of the tutorial7 data generator to remove the Roofit banner when executing the MQdevices
