search

mfussenegger / nvim-jdtls

Extensions for the built-in LSP support in Neovim for eclipse.jdt.ls
GNU General Public License v3.0
1k stars 62 forks source link

Running all tests with dap #563

Closed oysandvik94 closed 8 months ago

oysandvik94 commented 10 months ago

Problem Statement

I'm sure there's a way to solve this workflow issue, but I haven't figured it out myself.

I would like to run all the tests in my project, and identity which are failing.

With the test run commands with dap, I can run tests from a specific test class, but then I have to know which test class I want to check. Sometimes I dont know which tests could break.

I love how running a test class will populate my quickfix with errors, but I would love to get this from running all tests.

Ideas or possible solutions

If there isn't a way to do this now, is it possible to create a require'jdtls'.test_all() method, or is this difficult?

mfussenegger commented 8 months ago

Currently there is no function to run all tests. I had a look at what it would take to implement it and even have a prototype, but I'm not sure it's worth moving forward with it because it's a bit more complicated then I'd have liked.

The test-runner library used for junit/testng require at least class names to be specified and you can only run them per project. What this means is that nvim-jdtls would have to trigger 1 debug session per project and test-framework (or module, in multi-module projects).

This would then require merge-logic, and the test-result extraction logic would need to change, because it currently assumes that all results are for a single buffer.

But what makes this a bit of a deal breaker for me is that you cannot control concurrency. So for bigger projects this is likely going to be way slower than a mvn test or gradle test, where you can use forks.

I think a better option is to write a small tool or function that extracts the test failures from the xml files generated by mvn test or gradle test.

If it's a tool it could be used with :h -q. If it's a function it could add the errors to the quickfix list, or use the diagnostic api.

oysandvik94 commented 8 months ago

That makes sense. I agree a seperate tool/plugin would be the best. I tried implementing it in a compiler plugin a while ago so that I could pipe errors to QF with errorformat, but the filenames wasn't really deterministic enough (or at least I didn't manage to parse them properly).

I did start on a tool for myself to do this, so I'll see how far I get with that. Thanks for your input!