Move codebase to CPP

[alexsavulescu]

More details in #708.

[alexsavulescu]

For reasons that elude me at this point in time, 5 rxd tests are failing for the autotools jobs:

Passed:      ca_pump, multicompartment_mebrane_mismatch, multicompartment_reactions_with_v, nodes_update, cabuf, include_flux_cvode, cabuf_fixed_step, pure_diffusion_cvode, IraHH, react_region_specified, include_flux, verify_no_initialization_order_issue, reaction_test, multicompartment_reactions_del, reaction_param_test, multicompartment_reactions, pure_diffusion, reaction_null_dest, 3d/circadian_rhythm, 3d/pure_diffusion_3d_cvode, 3d/pure_diffusion_3d, 3d/include_flux3d_cvode, 3d/include_flux3d, 3d/multicompartment_reactions, wave1d/wave1d_125_1, wave1d/wave1d_25_5, wave1d/wave1d_taper_125_1, wave1d/wave1d_taper_25_5, wave1d/wave1d_tree_125_1, wave1d/wave1d_tree2_125_1, wave1d/wave1d_tree2_25_5, wave1d/wave1d_tree_25_5, ecs/ecs_example_cvode, ecs/ecs_example, ecs/ecs_multi_example_cvode, ecs/ecs_multi_example, ecs/ecs_before_sections, ecs/ecs_include_flux, ecs/ecs_include_flux_cvode, hybrid/bistable_hybrid_cvode, hybrid/bistable_hybrid_cvode_change_nthread, hybrid/bistable_hybrid, hybrid/bistable_hybrid_change_nthread, hybrid/pure_diffusion_hybrid_cvode, hybrid/pure_diffusion_hybrid
Failed:      hh_param, hh_cvode, hh_param_cvode, hh_morph, hh
Incomplete:  
---
Ran 50 tests:
Passed:     45
Failed:     5
Incomplete: 0

more here: https://travis-ci.com/github/neuronsimulator/nrn/jobs/414446958#L8937

[nrnhines]

How does one run one of the autotools tests. For cmake, make test is successful.

[alexsavulescu]

How does one run one of the autotools tests.

For autotools I use (on mac) :

cd nrn
git checkout cpp
./build.sh
./configure --without-x --with-paranrn=dynamic --with-nrnpython=`which python` --with-readline=no --prefix=`pwd`/install
make
make install
export PYTHONPATH=`pwd`/install/lib/python:$PYTHONPATH
export PATH=`pwd`/install/x86_64/bin:$PATH
python share/lib/python/neuron/rxdtests/run_all.py

For cmake, make test is successful.

I don't think those tests are run with cmake; by running python share/lib/python/neuron/rxdtests/run_all.py for a cmake install, all tests pass.

[pramodk]

@nrnhines : Just to be sync : @adamjhn just informed us following:

It looks like it's a difference in the currents coming from the hh mechanism. e.g. here is the difference in the ik for a simple model with no rxd on a CMake build - an autotools build

import numpy
from neuron import h
from matplotlib import pyplot
h.load_file('stdrun.hoc')
soma = h.Section('soma')
soma.insert('hh')
stim = h.IClamp(soma(0.5))
stim.delay = 50
stim.amp = 1
stim.dur = 50
tvec = h.Vector().record(h._ref_t)
kvec = h.Vector().record(soma(0.5)._ref_ik)
h.finitialize(-70)
h.continuerun(100)
numpy.save("/tmp/kvec.npy",kvec.as_numpy())
#kvec2 = numpy.load("/tmp/kvec.npy")
#pyplot.plot(tvec, kvec - kvec2)
#pyplot.xlabel("time (ms)")
#pyplot.ylabel("diff (mA/cm$^2$)")

So, if simple hh results are different between autotools and CMake then I suspect we have missed something fundamental. We haven't checked details yet. (Note that this different happens only this pull request branch; master works correctly)

[pramodk]

@nrnhines : I have fixed the issue in 26b4a77. (hh.mod modified by CMake was accidentally got committed)

[alexsavulescu]

Thank you @adamjhn & @pramodk.
This PR is ready for review.

[pramodk]

I will block some dedicated hours to go though all those 458 files 👀. All changes are straightforward but just to be sure we don't miss anything.

• @alexsavulescu @jorblancoa : we should run full BBP simulation test suite with this branch.
• We should also build this on Cray Piz-Daint system (just to be sure everything is consistent)

And then we should be ready to go 🚀

[alexsavulescu]

I'll also

• re-run nrnivmodl on all ModelDB repos for linux and macos.

[pramodk]

@alexsavulescu : Having bit curated commit message would be helpful for this mega merge? May be you can take a look at the log and copy message here?

@nrnhines @alexsavulescu : another question - should we merge all (or most of the) commits from current master to release/8.0 before this merge?

[pramodk]

Still WIP

[alexsavulescu]

@alexsavulescu : Having bit curated commit message would be helpful for this mega merge? May be you can take a look at the log and copy message here?

This sums it up:
move codebase to cpp (except nmodl + 3rd party old libs) - #708

@nrnhines @alexsavulescu : another question - should we merge all (or most of the) commits from current master to release/8.0 before this merge?

Sure, save ourselves some cherry-picking

[pramodk]

Half way through...!

[pramodk]

Few minor comments/questions otherwise LGTM

[alexsavulescu]

@pramodk so I'm happy if you're happy. CI providers have been erratic these past days (GHA doubling jobs, failed actions for no reason, Azure reporting failed jobs but they weren't...) but everything is a-OK.

[pramodk]

@pramodk so I'm happy if you're happy.

😃 ✅ 🚀