Cannot find definition for function 'Mod':module_info/1

[carlotm]

Describe the bug

Got Cannot find definition for function 'Mod':module_info/1 from this snippet

attr(Mod, K) when is_atom(Mod) and is_atom(K) ->
    Attrs = Mod:module_info(attributes),
    proplists:get_value(K, Attrs).

To Reproduce

Just use the module_info/1 function

Expected behavior

Don't get the error message

Actual behavior

Got Cannot find definition for function 'Mod':module_info/1

Context

• erlang_ls version (tag/sha): 0.48.0
• Editor used: vim
• LSP client used: https://github.com/yegappan/lsp

[fridayy]

Can confirm. I reckon a possible solution could be to skip the index lookup if Mod is a variable?

POI  = #{data =>
           #{mod_range => #{from => {14,13},to => {14,16}},
                 name_range => #{from => {14,17},to => {14,28}},
                 fun_is_variable => false,mod_is_variable => true},
           id => {'Mod',module_info,1},
           range => #{from => {14,13},to => {14,28}},
           kind => application}.

%% els_crossref_diagnostics.erl
has_definition(#{kind := application, id := {_Module, module_info, _Arity}, data := #{mod_is_variable := true}}) -> 
    true;

[robertoaloi]

@fridayy the has_definition clause you mention is used for the "XRef-like" diagnostics. It basically means that, if the module name is a variable, the language server assumes that a definition exists and does not complain about not existing functions.

The issue that @carlotm is reporting refers to the fact that no definition is found for the module_info/1 function. I'd say this is expected, since that function is a special function inserted by the compiler for each module. There's no place we can "jump to", so the "Cannot find definition" message seems legit. We could of course rephrase the message for those special functions.

[fridayy]

@robertoaloi thanks for the feedback!

I quickly skimmed through the code and found the has_definition clause that apparently handles module_info for "conventional" (not passed as function args) modules and thought this could be extended to handle this case as well. Happy to learn more about erlang-ls :)

But I see your point, in this particular case Mod might not even be an actual module so I think the error is reasonable and legit or maybe a warning would be more appropriate (if worth the hassle).

[carlotm]

@robertoaloi Thank you. Getting an error for valid code is a bit odd IMHO, but whatever... I'll live with it

[robertoaloi]

@carlotm Could you show where this surfaces as an error (e.g. via a screenshot)? I would expect the server, when no definition is found, to just return a [] or null according to the spec. I wonder if it's the client (yegapeppan) doing something weird here.

[fridayy]

[carlotm]

Mine is similar, just a little less fancy :)

[robertoaloi]

Ah, I see what's happening. So this is in fact coming from the XRef diagnostics. From the message I assumed it was a regular go-to-definition. Then yes, we can tweak that case.

[carlotm]

And how does it compare with the case where, for example, you invoke a function which does not exist?

If the function does not exist in Mod, no errors. In case of a non existing function, same exact error.

[plux]

Yep, this is a bug in CrossRef diagnostics.

[carlotm]

Thas was quick! Thank you!