x/tools/gopls: "pattern too long" error with gopls@master + k8s + coc.nvim

[findleyr]

Noticed that when using gopls@master in kubernetes, I am starting to get a new exception in coc.nvim:

I believe this is related to the (enormous) file watching pattern we serve for Kubernetes directories. But it is surprising that I don't get this error with gopls@v0.11.0. Needs investigation.

More generally, I really hope we can switch to simply requesting all file watching notifications, and avoid the enormous pattern entirely. But this must be carefully tested in various LSP clients.

[hyangah]

can we have an experimental flag to use a simplified file watching patterns instead and let people test (in particular those involved in the issue that resulted in the current way of file watch registration)?

[findleyr]

So, it turns out that we do get this error with gopls@v0.11.0, so I have moved this to 'okay-after-beta1'.

Here is the upstream discussion about file watching: microsoft/vscode#109754

In that discussion, it is suggested that we could simply watch **. I think we should test this in various LSP clients to confirm it works.

If we do so, I don't think we need to guard it behind a flag. Or if we do, the default should be to use the simplified watching logic and we can can have a temporary flag to revert to the old behavior.

[findleyr]

Unfortunately I don't think this is going to get prioritized for v0.12.0, as much as I'd like to fix it.

[findleyr]

Note, so that I don't forget: we could write a conformance test for LSP clients:

1. set up a test repo.
2. write a program that performs certain file operations on disk.
3. configure clients to start gopls with -debug=... and -logfile=...
4. pass the debug server and logfile location to the program from step 2
5. have the program tail the gopls logs and query gopls state via the debug server, to assert that disk operations were processed correctly

[gopherbot]

Change https://go.dev/cl/496835 mentions this issue: gopls: add a new "subdirWatchPatterns" setting