Conda packages (xeus-sql==0.2.1-hf52228f_2 + xeus==3.0.5-hac2b420_0) not working due to a (possible) ABI issue reg. nlohmann-json (Due to coinstallation with xeus-lua)

[camelusminimus]

Hi,

Short Version

The issue as such seems to be that the inclusion of xeus-lua leads to a solution in the conda solver that selects xeus build 3.0.5-hac2b420_0 (instead of the slightly newer 3.0.5-hac2b420_1) that is incompatible with xeus-sql build 0.2.1-hf52228f_2.

Would it be possible to update the xeus conda package definition to include this incompatibility?

If I may ask something a bit less-on-topic, but related (i.e. would be a workaround)
Is it possible to install both xeus-sql and xeus-lua in individual conda envs, but have them working in the same jupyter lab? (Provided the kernelspec is provided correctly?)
Or could the two different versions of libxeus generate problems? I must admit my knowledge of the ins-and-outs of dynamic libraries in linux are a bit rusty, but since the two kernels would not be in the same process, it should work, right?

I am sorry but the exact division of labor betweet jupyter-server and the kernels (and stuff like magics...) is still beyond me.

Longer Version

I tried to experiment with xeus-sql (and xeus-lua), and installed it using conda. This results in the following packages being installed

> conda list '^xeus'
# packages in environment at /opt/ai-skills/python/conda/envs/base:
#
# Name                    Version                   Build  Channel
xeus                      3.0.5                hac2b420_0    conda-forge
xeus-lua                  0.7.5           lua54h00ab1b0_0    conda-forge
xeus-sql                  0.2.1                hf52228f_2    conda-forge
xeus-zmq                  1.0.2                h0541b36_1    conda-forge

Sadly it seams the thus installed xeus-sql and xeus version are not ABI compatible, due to the influence of nlohmann-json. On trying to use the kernel, the jupyter lab complains:

/opt/ai-skills/jupyter/conda/bin/xsql: symbol lookup error: /opt/ai-skills/jupyter/conda/bin/../lib/libxeus-sql.so.0: undefined symbol: _ZN4xeus12xinterpreter21internal_request_implERKN8nlohmann16json_abi_v3_11_210basic_jsonISt3mapSt6vectorNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEEblmdSaNS2_14adl_serializerES5_IhSaIhEEEE

After Demangling, this symbol ist

> c++filt "_ZN4xeus12xinterpreter21internal_request_implERKN8nlohmann16json_abi_v3_11_210basic_jsonISt3mapSt6vectorNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEEblmdSaNS2_14adl_serializerES5_IhSaIhEEEE"
xeus::xinterpreter::internal_request_impl(nlohmann::json_abi_v3_11_2::basic_json<std::map, std::vector, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, bool, long, unsigned long, double, std::allocator, nlohmann::json_abi_v3_11_2::adl_serializer, std::vector<unsigned char, std::allocator<unsigned char> > > const&)

Checking for a similarly starting symbol in the installed xeus library, I find:

$ nm /opt/ai-skills/python/conda/envs/base/lib/libxeus.so | grep _ZN4xeus12xinterpreter21internal_request_implERKN8nlohmann
0000000000043d60 T _ZN4xeus12xinterpreter21internal_request_implERKN8nlohmann10basic_jsonISt3mapSt6vectorNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEEblmdSaNS1_14adl_serializerES4_IhSaIhEEEE
000000000001a7d2 t _ZN4xeus12xinterpreter21internal_request_implERKN8nlohmann10basic_jsonISt3mapSt6vectorNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEEblmdSaNS1_14adl_serializerES4_IhSaIhEEEE.cold

Demangled:

$ c++filt "_ZN4xeus12xinterpreter21internal_request_implERKN8nlohmann10basic_jsonISt3mapSt6vectorNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEEblmdSaNS1_14adl_serializerES4_IhSaIhEEEE"
xeus::xinterpreter::internal_request_impl(nlohmann::basic_json<std::map, std::vector, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, bool, long, unsigned long, double, std::allocator, nlohmann::adl_serializer, std::vector<unsigned char, std::allocator<unsigned char> > > const&)

Can one specify incompatibilities on the build level in conda?
I think it is less important that xeus-lua and xeus-sql are installed at the same time, but rather that conda will not let me install such a combination in the first place, right?