get_offset crashed

[davidanthoff]

From crash reporting.

Full error message there, not here, because it reveals too much about the client system.

Stack trace:

LanguageServer.LSOffsetError:
   at get_offset(::LanguageServer.Document, ::Int64, ::Int64) (./julialang.language-julia-0.16.7/scripts/packages/LanguageServer/src/document.jl93)
   at get_offset (./julialang.language-julia-0.16.7/scripts/packages/LanguageServer/src/document.jl96)
   at julia_getModuleAt_request(::LanguageServer.TextDocumentPositionParams, ::LanguageServerInstance, ::JSONRPC.JSONRPCEndpoint) (./julialang.language-julia-0.16.7/scripts/packages/LanguageServer/src/requests/features.jl305)
   at (::LanguageServer.var"#16#47"{LanguageServerInstance})(::JSONRPC.JSONRPCEndpoint, ::LanguageServer.TextDocumentPositionParams) (./julialang.language-julia-0.16.7/scripts/packages/LanguageServer/src/languageserverinstance.jl278)
   at dispatch_msg(::JSONRPC.JSONRPCEndpoint, ::JSONRPC.MsgDispatcher, ::Dict{String,Any}) (./julialang.language-julia-0.16.7/scripts/packages/JSONRPC/src/typed.jl63)
   at run(::LanguageServerInstance) (./julialang.language-julia-0.16.7/scripts/packages/LanguageServer/src/languageserverinstance.jl305)
   at top-level scope (./julialang.language-julia-0.16.7/scripts/languageserver/main.jl59)
   at include (boot.jl328)
   at include_relative(::Module, ::String) (loading.jl1105)
   at include(::Module, ::String) (Base.jl31)
   at exec_options(::Base.JLOptions) (client.jl287)
   at _start() (client.jl460)

This suggests that we are asking for an invalid position.

I can think of two explanations right now: a) we just have some simple bug in the client code, b) there are some tricky timing issues where for example the didChange notifications that are sent to the LS might get buffered while folks type, but on the client we get instant notifications when the selection changes, and then we are operating on different versions of the document. If the latter is the problem, we could probably add some diagnostics where we send the version field of the doc with the request, that would allow us to identify that problem in the server.

We should probably treat this bug as a fairly high priority one, this is the one that I saw relatively frequently in the insider build already, and where we said we would try to fix it quickly.