support a "don't care" for cwd in the runInTerminal request

[weinand]

Since the cwd property of the runInTerminal request is not optional, it is not possible to express the "don't care" case, e.g. if a DA knows that the "cwd" is already correct and that there is no need to generate a cd ... command.

[puremourning]

how about just sending . ?

A special "don't care" value is no less a breaking change than allowing the key to be omitted; both require a client capability, right?

[weinand]

A cwd . will still result in a cd .;
Please see dotnet/vscode-csharp#4203. Gregg wants to completely suppress the cd ...

[puremourning]

not in my client. I don't really understand where the cd . is coming from... ? There's no requirement for the client to issue that as a command in the terminal.

[weinand]

Depending on the implementation of runInTerminal in the client, it might be necessary to ensure that the debuggee is launched in the specified cwd. For example if the terminal is reused across debug launches, and two projects have different working directories.

[gregg-miskelly]

Both VS Code (so far I have tested Linux and Windows) and VS are happy with empty string. Anyone have a client which isn't?

[puremourning]

In case it helps...

• up until yesterday, vimspector would give an obnoxious error if server sent an invalid path for cwd. I happened to find a buggy server that happened to misbehave, so I changed it yesterday https://github.com/puremourning/vimspector/blame/master/python3/vimspector/terminal.py#L21
• vim-dap would probably break if I read this correctly https://github.com/dradtke/vim-dap/blob/f15a8b869bca5951936dcdddd4a262c34d7f518d/autoload/dap.vim#L574
• nvim-dap apparently ignores cwd https://github.com/mfussenegger/nvim-dap/blob/926b049d7200ec7f0c69e5bcb6cbfa737207cafb/lua/dap.lua#L243-L253
• I checked dap-mode but I can't elisp enough to be sure one way or another https://github.com/emacs-lsp/dap-mode/blob/8b7cbc68904036c211f29420c24cac82e74d817a/dap-mode.el#L974.

I suspect that any other client would reasonably expect the value supplied to be a valid path.

[weinand]

Since I don't want to introduce a new capability for this ancillary feature, I propose to change the cwd comment to the following:

  /**
   * Working directory for the command. For non-empty, valid paths this typically results in execution of a change directory command ('cd cwd'). 
   */
  cwd: string;

[gregg-miskelly]

LGTM