python: Use PyBind11 instead of SWIG for Python wrappers
authorAndreas Sandberg <andreas.sandberg@arm.com>
Mon, 27 Feb 2017 13:17:51 +0000 (13:17 +0000)
committerAndreas Sandberg <andreas.sandberg@arm.com>
Tue, 2 May 2017 12:37:32 +0000 (12:37 +0000)
commit60e6e785f970578eba6dbee9330eaecf514b23c8
tree9c47c06d22a39b9467d3d4e35c11849a17df7815
parentba42457254cc362eddc099f22b60d469cc6369e0
python: Use PyBind11 instead of SWIG for Python wrappers

Use the PyBind11 wrapping infrastructure instead of SWIG to generate
wrappers for functionality that needs to be exported to Python. This
has several benefits:

  * PyBind11 can be redistributed with gem5, which means that we have
    full control of the version used. This avoid a large number of
    hard-to-debug SWIG issues we have seen in the past.

  * PyBind11 doesn't rely on a custom C++ parser, instead it relies on
    wrappers being explicitly declared in C++. The leads to slightly
    more boiler-plate code in manually created wrappers, but doesn't
    doesn't increase the overall code size. A big benefit is that this
    avoids strange compilation errors when SWIG doesn't understand
    modern language features.

  * Unlike SWIG, there is no risk that the wrapper code incorporates
    incorrect type casts (this has happened on numerous occasions in
    the past) since these will result in compile-time errors.

As a part of this change, the mechanism to define exported methods has
been redesigned slightly. New methods can be exported either by
declaring them in the SimObject declaration and decorating them with
the cxxMethod decorator or by adding an instance of
PyBindMethod/PyBindProperty to the cxx_exports class variable. The
decorator has the added benefit of making it possible to add a
docstring and naming the method's parameters.

The new wrappers have the following known issues:

  * Global events can't be memory managed correctly. This was the
    case in SWIG as well.

Change-Id: I88c5a95b6cf6c32fa9e1ad31dfc08b2e8199a763
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Andreas Hansson <andreas.hansson@arm.com>
Reviewed-by: Andrew Bardsley <andrew.bardsley@arm.com>
Reviewed-on: https://gem5-review.googlesource.com/2231
Reviewed-by: Tony Gutierrez <anthony.gutierrez@amd.com>
Reviewed-by: Pierre-Yves PĂ©neau <pierre-yves.peneau@lirmm.fr>
Reviewed-by: Jason Lowe-Power <jason@lowepower.com>
47 files changed:
SConstruct
src/SConscript
src/arch/arm/ArmPMU.py
src/arch/arm/ArmSystem.py
src/cpu/BaseCPU.py
src/cpu/kvm/BaseKvmCPU.py
src/cpu/kvm/X86KvmCPU.py
src/python/SConscript
src/python/_m5/__init__.py [deleted file]
src/python/m5/SimObject.py
src/python/m5/event.py
src/python/m5/internal/params.py
src/python/m5/params.py
src/python/m5/util/pybind.py [new file with mode: 0644]
src/python/pybind11/core.cc [new file with mode: 0644]
src/python/pybind11/debug.cc [new file with mode: 0644]
src/python/pybind11/event.cc [new file with mode: 0644]
src/python/pybind11/pybind.hh [new file with mode: 0644]
src/python/pybind11/pyobject.cc [new file with mode: 0644]
src/python/pybind11/stats.cc [new file with mode: 0644]
src/python/swig/core.i [deleted file]
src/python/swig/debug.i [deleted file]
src/python/swig/drain.i [deleted file]
src/python/swig/event.i [deleted file]
src/python/swig/inet.i [deleted file]
src/python/swig/pyevent.cc [deleted file]
src/python/swig/pyevent.hh [deleted file]
src/python/swig/pyobject.cc [deleted file]
src/python/swig/pyobject.hh [deleted file]
src/python/swig/pyobject.i [deleted file]
src/python/swig/range.i [deleted file]
src/python/swig/serialize.i [deleted file]
src/python/swig/stats.i [deleted file]
src/python/swig/time.i [deleted file]
src/python/swig/trace.i [deleted file]
src/sim/Process.py
src/sim/System.py
src/sim/init.cc
src/sim/init.hh
src/sim/power/PowerModel.py
src/sim/power/PowerModelState.py
src/sim/power/ThermalDomain.py
src/sim/power/ThermalModel.py
src/unittest/SConscript
src/unittest/stattest.cc
src/unittest/stattest.i [deleted file]
tests/configs/switcheroo.py