NEURON fingerprint + features-enabled API

[alexsavulescu]

Overview of the feature

Would be handy to have (in addition to nrnversion for starters) a sort of fingerprint function that gives out full compilation info that we could enrich( i.e. Python version and build info) :

• quicker to debug issues (ask users to provide fingerprint)
• facilitate scripting and testing

Something along the lines of:

$> nrniv -fingerprint
-- --------------+--------------------------------------------------------------
--  Fingerprint  | Version info
-- --------------+--------------------------------------------------------------
-- NEURON -- VERSION 8.0a-654-g6cdbef8d1 HEAD (6cdbef8d1) 2021-09-15
-- --------------+--------------------------------------------------------------
--  Fingerprint  | Build options
-- --------------+--------------------------------------------------------------
-- C COMPILER    | /usr/bin/cc -> (Ubuntu 9.3.0-17ubuntu1~20.04) 9.3.0
-- CXX COMPILER  | /usr/bin/c++ -> (Ubuntu 9.3.0-17ubuntu1~20.04) 9.3.0
-- BUILD_TYPE    | RelWithDebInfo (allowed: Custom;Debug;Release;RelWithDebInfo;Fast)
-- COMPILE FLAGS | -g  -O2 -Wno-write-strings
-- Shared        | ON
-- Default units | modern units (2019 nist constants)
-- MPI           | ON
--   INC         | /usr/lib/x86_64-linux-gnu/openmpi/include/openmpi;/usr/lib/x86_64-linux-gnu/openmpi/include
--   LIB         | /usr/lib/x86_64-linux-gnu/openmpi/lib/libmpi_cxx.so
--   DYNAMIC     | OFF
-- Python        | ON | Python 3.6.8 (default, Aug 13 2020, 07:46:32) [GCC 4.8.5 20150623 (Red Hat 4.8.5-39)]
--   EXE         | /home/savulesc/Workspace/nrn/venv/bin/python3
--   INC         | /usr/include/python3.8
--   LIB         | /usr/lib/x86_64-linux-gnu/libpython3.8.so
--   MODULE      | ON
--   DYNAMIC     | OFF
-- Readline      | /usr/lib/x86_64-linux-gnu/libreadline.so
-- Curses        | /usr/lib/x86_64-linux-gnu/libcurses.so;/usr/lib/x86_64-linux-gnu/libform.so
-- RX3D          | ON
--   OptLevel    | 0
-- Interviews    | ON
--   PATH        | /home/savulesc/Workspace/nrn/external/iv
--   INC         | /home/savulesc/Workspace/nrn/external/iv/src/include
--   X11 (INC)   | /usr/include
--       (LIBDIR)| /usr/lib/x86_64-linux-gnu
-- CoreNEURON    | OFF

We could also introduce a sort of API that would easily tell us if a feature is enabled, something like fingerprint('CoreNEURON^GPU')

In principle we can extrapolate enough of this info from nrnversion but that would take a few manual operations, i.e:

# Get git commit
$> nrniv -nobanner -nogui -c 'nrnversion(1)'
NEURON -- VERSION 8.0a-654-g6cdbef8d1 HEAD (6cdbef8d1) 2021-09-15

# Get non-default CMake options
$> nrniv -nobanner -nogui -c 'nrnversion(6)'
cmake option default differences: 'NRN_ENABLE_RX3D=OFF' 'NRN_ENABLE_MODULE_INSTALL=OFF' 'NRN_ENABLE_PYTHON_DYNAMIC=ON' 'NRN_MPI_DYNAMIC=/nrnwheel/openmpi/include;/nrnwheel/mpich/include' 'CMAKE_BUILD_TYPE=Release' 'CMAKE_INSTALL_PREFIX=/root/nrn/build/cmake_install' 'CMAKE_C_COMPILER=/opt/rh/devtoolset-9/root/usr/bin/cc' 'PYTHON_EXECUTABLE=/root/nrn/nrn_build_venv38_72428107/bin/python'

Foreseeable Impact

No impact on current usage. We may choose to deprecate nrnversion sometime in the future.

• Area(s) of change: CMake, HOC, Python

[pramodk]

I can't find issue/pr but I remember @nrnhines mentioning some API also needed to check if coreneuron / coreneuron-gpu support is available at runtime.
Edit: If we add API, we need a way to check different components. E.g. Is MPI/RXD/INTERVIEWS/CORENEURON/GPU/NMODL supported by this NEURON installation.

[ohm314]

You'd start with something like this: https://github.com/BlueBrain/nmodl/blob/master/cmake/GitRevision.cmake and add some info from CMakeCache.txt ?

[alexsavulescu]

Well a quick thought I had was to incorporate the status message we have at config time (and enrich it if possible/applicable with version/build info, some CMakeCache.txt as you mentioned). For the git revision we have something a bit complex at the moment, but maybe that can be updated. We need something extra for the API though, so more consideration required on how to implement this.

[nrnhines]

Perhaps not possible, but something I often desire is to know is the cmake ... configure line that would produce the same suite of executables.

[alexsavulescu]

Perhaps not possible, but something I often desire is to know is the cmake ... configure line

We can probably leverage the cmake cache and extract those:

(venv) savulesc@bbd-cjngk03:~/Workspace/nrn/bld$ cmake -LA . | grep NRN
Using Target in compiling and linking, you should take care of it if
                    it fail miserabily (CODE: 1234567890)
FETCHCONTENT_SOURCE_DIR_NRNTEST:PATH=
FETCHCONTENT_SOURCE_DIR_TESTCORENRN:PATH=
FETCHCONTENT_UPDATES_DISCONNECTED_NRNTEST:BOOL=OFF
FETCHCONTENT_UPDATES_DISCONNECTED_TESTCORENRN:BOOL=OFF
NRN_3RDPARTY_USE_RXD_TESTDATA:BOOL=ON
NRN_3RDPARTY_USE_TESTS_TESTCORENRN:BOOL=ON
NRN_DYNAMIC_UNITS_USE_LEGACY:BOOL=OFF
NRN_ENABLE_BACKTRACE:BOOL=OFF
NRN_ENABLE_CORENEURON:BOOL=OFF
NRN_ENABLE_COVERAGE:BOOL=OFF
NRN_ENABLE_DISCRETE_EVENT_OBSERVER:BOOL=ON
NRN_ENABLE_INTERVIEWS:BOOL=ON
NRN_ENABLE_MECH_DLL_STYLE:BOOL=ON
NRN_ENABLE_MODULE_INSTALL:BOOL=ON
NRN_ENABLE_MOD_COMPATIBILITY:BOOL=OFF
NRN_ENABLE_MPI:BOOL=ON
NRN_ENABLE_MPI_DYNAMIC:BOOL=OFF
NRN_ENABLE_PROFILING:BOOL=OFF
NRN_ENABLE_PYTHON:BOOL=ON
NRN_ENABLE_PYTHON_DYNAMIC:BOOL=OFF
NRN_ENABLE_REL_RPATH:BOOL=OFF
NRN_ENABLE_RX3D:BOOL=ON
NRN_ENABLE_SHARED:BOOL=ON
NRN_ENABLE_TESTS:BOOL=ON
NRN_ENABLE_THREADS:BOOL=ON
NRN_MODULE_INSTALL_OPTIONS:STRING=--prefix= --home=/usr/local
NRN_MPI_DYNAMIC:STRING=
NRN_NMODL_CXX_FLAGS:STRING=-O0
NRN_PROFILER:STRING=caliper
NRN_PYTHON_DYNAMIC:STRING=
NRN_RX3D_OPT_LEVEL:STRING=0
NRN_WHEEL_STATIC_READLINE:BOOL=OFF

that would produce the same suite of executables.

However, there are other factors (like compiler or python versions) that can play a hand in that regard.

[olupton]

Note that we now have a Python API for configuration-time settings:

from neuron.config import arguments
arguments["NRN_ENABLE_CORENEURON"] # etc.