How to tell vscode language server to use julia executable path and DEPOT_PATH?

Tooling VS Code
, ,
1

I am using a setup on a shared cluster where I work with the vscode remote extension and where Julia is installed using a module system rather than juliaup.

In VSCode settings I set both Julia Executable path and Language Server Executable path to the 1.11.7 binary: /Net/Groups/Services/HPC_22/apps/julia/julia-1.11.7/bin/julia

Starting the Julia REPL correctly uses Julia 1.7:

julia> VERSION, DEPOT_PATH
(v"1.11.7", ["/User/homes/twutz/scratch/twutz/julia_gpu_depots", "/Net/Groups/Services/HPC_22/apps/julia/julia-1.11.7/local/share/julia", "/Net/Groups/Services/HPC_22/apps/julia/julia-1.11.7/share/julia"])

However the language server output pane reports:

 Info: Starting LS with Julia 1.11.6
  Activating project at `/Net/Groups/BGI/scratch/twutz/.vscode-server/extensions/julialang.language-julia-1.149.2/scripts/environments/languageserver/v1.11`
...
Failed to precompile Tokenize [0796e94c-ce3b-5d07-9a54-7f471281c624] to "/User/homes/twutz/.vscode-server/data/User/globalStorage/julialang.language-julia/lsdepot/v1/compiled/v1.11/Tokenize/jl_Qe9nz4".

ERROR: Unable to load dependent library /Net/Groups/BGI/scratch/twutz/julia_cluster_depots/juliaup/julia-1.11.6+0.x64.linux.gnu/bin/../lib/julia/libjulia-codegen.so.1.11

Message:libLLVM-16jl.so: failed to map segment from shared object

From that output I suspect that the LS (language server)

  • uses the wrong julia executable (the juliaup version from the shared file system rather than the module version)
  • uses a wrong JULIA_DEPOT_PATH (I set it to different value in my .profile when running julia from a gpu node:
if [ "$(hostname)" == "node-r6-he01.bgc-jena.mpg.de" ]; then
  export JULIA_DEPOT_PATH="/User/homes/twutz/scratch/twutz/julia_gpu_depots:"
else
  export JULIA_DEPOT_PATH="/User/homes/twutz/scratch/twutz/julia_cluster_depots:"
fi

I try to only use the module version and to use separate DEPOT_PATH depending on which node on the cluster julia is started (the nodes share the same file system). I suspect that the usage of the wrong julia executable and wrong DEPOT_PATH by the LS seems to mess up my julia installation, e.g. by compiling CUDA with available shared NVIDIA libraries or without them to the wrong DEPOT_PATH.

How do I tell vscode language server to use correct julia executable path and DEPOT_PATH?

2

In VS Code, click on the “Extensions” tab, and click on the gear symbol next to the Julia exentions and go to “Settings”. Just type in “path” into the search bar and all the path-related fields will appear.

You can definitely select the correct executable, but I don’t see a field for DEPOT_PATH explicitly.

3

Thanks for the reply. The suggested procedure results in two additional settings:

  • Julia: environment Path
  • Julia > Plots:Path
    Both seem to be of no significance.

I already configured

  • Julia Executable path
  • Language Server Executable path

but those are not picked up by the LS.