Includes:
- refactor build_lib and build_lib_example macros
- unify src and contrib macros
- replace macro with function not to leak definitions
- parse list of arguments
- different cmake-format file for modules to list one item per line
Includes:
- fix python warnings and wrap pkgconfig_interface_include_directories in quotes to treat it as a string
- prevent failure when Python is not found
- refactor python bindings checks and error messages
- export 3rd-party include directories and linked libraries
- Adding a NS3_REEXPORT_THIRD_PARTY_LIBRARIES option to always export include directories and libraries used by a module, making it automatically available to 3rd-party modules
- Making NS3_REEXPORT_THIRD_PARTY_LIBRARIES option enabled by default
- replace colliding shortcuts with list of non-ambiguous shortcuts
- remove unnecessary code blocks from click
- warn users that doxygen target requires examples and tests
- These are used to generate introspected documentation and guarantee proper dependency checking and build order
Includes:
- scan python scripts
- run python scripts from ns3
- replace visualizer file copy with configure_file to prevent cmake refreshes
- replace ns__init__.py file copy with configure_file to prevent cmake refreshes
- fix bindings scanning with cmake
- pass include directories to modulegen for castxml consumption
- add missing parameters of Recv in python-unit-tests.py
- change apiscan targets from apiscan-module to libmodule-apiscan
- change bindings targets from module-bingings to libmodule-bindings
- scanning and bindings build tests
- scan scratch python scripts
- replace FindPython3 with FindPython to be compatible with CMake 3.10
- do not export private visual-simulator-impl.h
- do not export udp-socket-impl.h
- use .so suffix for bindings on Mac instead of .dylib
Includes:
- inform ns3 user cmake was not found, or an unsupported version was found
- replace open-mpi environment variables with command-line arguments
- mark more variables as advanced
- add -v and --verbose option to run and only output the programs output by default
- fix scratch subdir prefixes and add tests
- prevent cmake crash when scratch sources do not have a main function
- disable MPI module by default
- wrap cmakelists_content in quotes to prevent failure of filter_libraries
- install pkgconfig files and add tests
- pkg-config generation (still missing installation)
- forward PYTHONPATH to modulegen
- fix dependency search for brite, click and openflow
Co-authored-by: Sandra Lagen <sandra.lagen@cttc.es>
This commit also fixes an issue that was discovered during this implementation,
and that is that the stream was never assigned for the uniform random variable used for
the term D and alpha generation.
-Store term D and alpha into the channel params structure.
-Make sure to use correct angles and zenit of departure and arrival:
Save the direction of the generation of the channel params and channel matrix,
in order to know when regenerating channel matrix if the saved angles aoa, aod,
zoa and zod are of of s-to-u or u-to-s, so to properly use them.
- Simplify the type of the key of channel matrix map
- Remove unecessary if statements and abort replace by assert
- Add a more detailed description of the ThreeGppSpectrumPropagationLossModel changes
It seems that is not the best way to save in the structure the pointer
to the channel condition model because the underlying condition,
and later when we compare the new condition with the old one, we
are basically comparing the underlying object with itself, because
we saved the pointer to it in the structure.
Instead we should save the concrete values of LOS and O2I condition
in the 3gpp structure, and then once the ChannelCondition state changes,
we can compare these saved values against the new once.
This could not be done when the pointer was being changed because pointer
was pointing to the same object.
This fixes basically the condition in 3gpp whether th channel needs to be
updated when condition changes.
-Create separate base structures for matrix and params in MatrixBasedChannelModel
-Change the key of m_channelParamsMap to be based on the pair of nodeIds instead of pair of antenna arrays:
Since the parameters in the m_channelParamsMap are common for all the channels among the same pair of nodes,
these parameters should be saved per the pair of node ids, instead per the pair of panels.
On the other hand, the map m_channelMatrixMap saves the channel matrixes per the pair of antenna arrays.
- Implement the update of the channel params and channel matrix of ThreeGppChannelModel
- Fix doxygen and remove unused structures in ThreeGppChannelModel
- Save 3gpp channel matrix and long term per antenna pair instead of node pair
- Split GetNewChannel function in two parts: parameter generation and matrix generation
- Split 3gpp channel matrix and parameters into two different structures
This possibility is needed for the types of spectrum propagation models that need to have a pointer
to the TX and RX antenna arrays to perform calculations, such as e.g.,
ThreeGppSpectrumPropagationLossModel, which currently has hard limitation of having a maximum of 1
antenna array instance per device. ThreeGppSpectrumPropagationLossModel now inherits a new class
PhasedArraySpectrumPropagationLossModel. Its DoCalcRxPowerSpectralDensity function has now two
additional parameters: TX and RX antenna arrays. Also, AddDevice function is removed from
ThreeGppSpectrumPropagationLossModel, because it is not anymore needed to specify for each
device its TX/RX antenna, since these are now passed as parameters. Hence, to use
ThreeGppSpectrumPropagationLossModel, when implementing Spectrum PHY features,
modules should implement the function of SpectrumPhy GetAntenna to return the instance of
PhasedArrayModel of that SpectrumPhy. These instances will be passed by MultiModelSpectrumChannel
(one of the core classes in the spectrum module) to propagation classes, children of
PhasedArraySpectrumPropagationLossModel, when it is necessary to DoCalcRxPowerSpectralDensity to
callculate the propagation loss.
Additionally: Fixed pair key generation for 3gpp channel model: Previous implementation was
using a cantor function of two integer of 32 bits and as a results was giving a unique
representatio of 32 bits. This is wrong, because such function needed even more than 64
bits to represent that value. Even if in ns-3 simulation we are not going to have such large
node numbers, we should fix this. Instead of just changing 64 bits from cantor function, I propose to
replace the cantor function with just simple concatenation of two 32 bits.
I also propose moving min and max inside of the function, to prevent some eventual future errors
in using GetKey, because with the previous implementation there was assumption that who calls GetKey
will previously sort its parameters, and first provide the min value, and then the max value,
and this was not even documented.
Including:
- add missing command for introspected doxygen
- run get_version.sh before running doxygen
- use find_package(Doxygen) to get the doxygen executable
- silence python and sqlite find_package warnings
- return cmake returncode if configuration fails
- require GTK3 3.22
- link all libraries to print-introspected-doxygen
- replace shell with use_shell for variable name
- revert wrong changes to propagation of return codes and add test
- disable pch when ccache is found
- make --enable-sudo a post-build step and a runtime option
- add docs subparser
- add enable-sudo option
- refactor positional argument values
- fix --check option and add shell option
- replace --no-task-lines with --quiet
- replace --nowaf with --no-build
- replace --run --run-no-build with run (--no-build)
- replace ns3 documentation related arguments with targets
- document test-ns3.py
- export include directories used by ns3 libraries
- refactor CMake documentation dependency checking and behavior
- add --allow-run-as-root for running MPI examples on the CI
NOTE 1: the purpose of calling the PHY capabilities changed callback in
WifiPhy::DoChannelSwitch was to have a station sending a Reassociation
Request after changing the channel width. However, changing channel width
implies a channel switching. Therefore, the station requires to associate
again and hence it is useless to invoke such callback.
NOTE 2: MAC queues were flushed upon channel switching. This is now removed
because station and AP may move to the same channel; hence, why dropping all
pending frames? Also, in case of (future) 11be multi-link devices, a channel
switching on a single link does not prevent to continue using the other links.
