Initial attempt at implementing VERBATIM.

[1uc]

Edit the idea is to temporarily print the NOCMODL macros, e.g. the MOD file:

VERBATIM
_lget_gbl = gbl;
ENDVERBATIM

generates:

        // Setup for VERBATIM
        #define gbl inst.global->gbl
        #define get_gbl get_gbl_globals
        #define _lget_gbl ret_get_gbl
        #define t nt->_t
        // Begin VERBATIM

        _lget_gbl = gbl;
        // End VERBATIM
        #undef gbl
        #undef get_gbl
        #undef _lget_gbl
        #undef t
        // End of cleanup for VERBATIM

One can see the following happening:

       // Translate `gbl` to the correct C++ code:
       #define gbl inst.global->gbl

       // Because the substring `get_gbl` is present we print the
       // full name of the function `get_gbl` (even though it's not
       // needed).
       #define get_gbl get_gbl_globals

       // Same translation logic for the return type.
       #define _lget_gbl ret_get_gbl

There's one more twist relates to threadargs in both NOCMODL and NMODL there's two variations of prototypes: one internalthreadargsproto and threadargsproto. In NOCMODL the number and order of the arguments are the same (their types differ slightly). Therefore, in NOCMODL a single macro is sufficient to call both types, i.e. threadargs works for both types of functions. There is no internalthreadargs. In NMODL the two signatures differ significantly and for this reason we need to massage VERBATIM code to determine which string to substitute for threadargs.

[codecov]

Codecov Report

Attention: Patch coverage is 25.00000% with 84 lines in your changes missing coverage. Please review.

Project coverage is 86.23%. Comparing base (c5052b9) to head (63ff7e0).
Report is 2 commits behind head on master.

Files	Patch %	Lines
src/codegen/codegen_neuron_cpp_visitor.cpp	1.23%	80 Missing ⚠️
src/codegen/codegen_coreneuron_cpp_visitor.cpp	87.50%	2 Missing ⚠️
src/codegen/codegen_cpp_visitor.hpp	0.00%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1364      +/-   ##
==========================================
- Coverage   86.75%   86.23%   -0.52%     
==========================================
  Files         180      180              
  Lines       13669    13760      +91     
==========================================
+ Hits        11858    11866       +8     
- Misses       1811     1894      +83     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[bbpbuildbot]

Logfiles from GitLab pipeline #223048 (:white_check_mark:) have been uploaded here!

Status and direct links:

• ✅ spack_setup
• ✅ build:intel
• ✅ build:nvhpc
• ✅ test:intel
• ✅ test:nvhpc

[JCGoran]

Overall looks good, though I haven't checked how this is used in NEURON tests or modelDB. Is there a reference somewhere how users can access the variables in VERBATIM (like _lget_gbl)? If not, some docs on how NMODL does it would be great as the main NEURON docs don't really list any rules for the replacement mechanism.

[1uc]

Is there a reference somewhere how users can ...

I don't know but I'm not writing one. Users should not use and ideally not know about VERBATIM. It constantly causes severe issues that limit our ability to generate code freely.

[nrnhines]

Is there a reference somewhere how users can ...

I don't know but I'm not writing one. Users should not use and ideally not know about VERBATIM. It constantly causes severe issues that limit our ability to generate code freely.

There has never been a promise that use of NEURON internals within mod file VERBATIM blocks would be stable. The assumption is that occasionally VERBATIM can be useful even if intimately tied to a specific version NEURON. An example
is the VERBATIM block in ri.mod of the Traub model which makes use of the details of the internal segment data structure.
https://github.com/ModelDBRepository/82894/blob/master/mod/ri.mod . As in this case, we developers have had some success in supporting published modeldb models. Since NEURON is used in part in the domain of Computational Neuroscience research, it is sometimes necessary for users to extend the program. In practice this sometimes means that the researcher should ask a developer for advice on how to implement the extension needed. Forking and modifying NEURON source is a last resort and in most cases dealing with the issue with a mod file with a VERBATIM block that has knowledge of NEURON internals sufficed.

[JCGoran]

There has never been a promise that use of NEURON internals within mod file VERBATIM blocks would be stable.

In that case, I think we need to put a huge warning label in the NEURON docs, currently there is some mention of compatibility issues, but it's not nearly prominent enough.

Regarding the mod file linked: it doesn't work with this NMODL PR as the #defines from the VERBATIM block do not appear anywhere in the cpp file.

[1uc]

it doesn't work

Fixed & testing that now. It's related to SUFFIX nothing.

[1uc]

Regarding how well it's tested, not very well. It's enough to compile everything we need for the nrn repository. Without it's slowly a bit tedious to make progress towards getting nmodl working for all nrn unit-test.

[JCGoran]

Seems to work (w/ some non-verbatim-related tweaks) with the built-in NEURON files, great job!