.lintr defaults are not respected

[pat-s]

I have a .lintr file in my repo root and running lintr::lint_package() works fine.

However, the disabled linters are not honored from languageserver and still appear in the problems pane.

.lintr

linters: with_defaults(assignment_linter = NULL, object_name_linter = NULL)

[randy3k]

What editor was it?

[pat-s]

Ups sorry, forgot that you do editor-agnostic work here -> vscode.

[randy3k]

By any chance, did you ever specified languageserver.default_linters in .Rprofile? That setting is now deprecated.

[renkun-ken]

I could reproduce this.

It's actually a bug on our side. Since we use lintr::lint(text) at https://github.com/REditorSupport/languageserver/blob/master/R/diagnostics.R#L94, it writes the content to a temp file and no longer finds the .lintr config as the original file.

Since there's no exported functions for us to get the linters specified in the right config file, fixing it would be a little tricky. Any good idea, @randy3k?

[randy3k]

I guess we could hack into https://github.com/jimhester/lintr/blob/1a7114b87c4d672772c74ffc30c932dcdb3ce017/R/settings.R#L64 to specify lintr.linter_file

[renkun-ken]

It looks we could safely specify options(lintr.linter_file) in diagnose_file since it is run in a background process that does not affect the main process.

[renkun-ken]

I test with setting options(lintr.linter_file) in diagnose_file as implemented in #236 and it works nicely.

[pat-s]

Thanks for taking this on so quickly.

Unfortunately, linting is now completely broken for me (with latest dev).
I get

Failed to run diagnostics

in the "Problems" pane.

I had to roll back to b5ce3a8.

[renkun-ken]

@pat-s Would you like to turn on the R lsp debug and see the output of the languageserver. For example, if I put a .lintr that is not ended with a newline, then the diagnostics could fail with the following log:

[2020-03-20 19:33:58.370] diagnostics_task: <callr_status_error: callr subprocess failed: Malformed config file, ensure it ends in a newline
  zero-length inputs cannot be mixed with those of non-zero length>
 in process 
-->
<callr_remote_error: Malformed config file, ensure it ends in a newline
  zero-length inputs cannot be mixed with those of non-zero length>
 in process 48797 

To ensure your .lintr actually work, please check if your .lintr works with your file:

lintr::lint("path/to/your_file.R")

[renkun-ken]

Looks like we should write the error message as the diagnostics message to make it easier to debug.

[renkun-ken]

d6dba1b shows the diagnostics error. Please try it and let us know the error shown in the diagnostics.

For example, if the .lintr if malformed, then the message will show like

Failed to run diagnostics: callr subprocess failed: <text>:2:0: unexpected end of input
1: with_defaults(assignment_linter = NULL, object_name_linter = NULL
   ^

[pat-s]

Thanks @renkun-ken for helping.

With latest dev, the diagnostic message I get is unfortunately not that much informative:

Failed to run diagnostics: callr subprocess failed: attempt to apply non-function

With languageserver diagnostics I get

<callr_remote_error in linters[[linter]](expr):
 attempt to apply non-function>
 in process 92379

Seems like something is wrong within the R script as I get

lintr::lint("~/git/mlr3learners/mlr3learners.fnn/R/LearnerClassifFNN.R")                                                    
Error in linters[[linter]](expr) : attempt to apply non-function

File

The call works with default linters, i.e. linters: with_defaults() in .lintr.

This is also fine linters: with_defaults(assignment_linter = NULL)

And the following causes the issue: linters: with_defaults(object_name_linter = c("snake_case", "CamelCase"))

---

I think it is an {lintr} issue: r-lib/lintr#471

Nevertheless the latest changes should help users tracking down issues in the future! Thanks.