ocaml-lsp reports bogus erros on large files when editor autosaving is enabled

[copy]

I can reproduce the following issue on OCaml 4.14 with ocaml-lsp-server 1.13.1 and 1.14.2, and OCaml 5.0 with ocaml-lsp-server pinned to current #500 (176d387).

It's a bit tricky to reproduce, but the following seems to work relatively reliably.

% opam source core
Successfully extracted to /tmp/core.v0.15.1
% dune build -w core/src/core.cmxs

Make some quick changes in large file with auto-saving enabled. I use the following vimrc:

command! AutoWrite autocmd TextChanged,InsertLeave <buffer> try | undojoin | silent! update | catch /^Vim\%((\a\+)\)\=:E790/ | finally | silent! update | endtry

augroup autowrite
    autocmd!
    autocmd Filetype ocaml AutoWrite
augroup END

Edit core/src/command.ml and repeatedly paste open Sys at the top. After a few pastes, lsp reports a bogus syntax error, while the dune watch command doesn't print anything of the sort.

Here is a recording: https://asciinema.org/a/gQqsK784tKJYwPxtB7wBn6I61
And the lsp logfile: lsp.log

[rgrinberg]

Are you sure you need watch mode to reproduce this? If lsp reports a false syntax error, the issue is likely due to document synchronization.

[copy]

Yeah, I'm pretty sure the issue only happens when watch mode is running.

[rgrinberg]

What about when you do the following:

1. Get to the state where the server is showing you a false syntax error
2. Save the document - :w

Will you still see the syntax error?

[copy]

Yes, the syntax error is still present after :w. The only way to get rid of it is to delete the offending line or restart the lsp.

[rgrinberg]

Okay, then do you mind trying this PR? #1004

This sounds like a syncing issue.

[copy]

Unfortunately, the issue still happens with #1004 applied.

[rgrinberg]

I'll have a closer look. Thanks for testing.

[rgrinberg]

I was unable to reproduce the issue on master after a lot of trying. @ulugbekna have you ran into issues like this?

[rgrinberg]

Fixed by #1004

[ulugbekna]

1. You probably meant refactor: simplify merlin diagnostics #1005
2. I can't reproduce the problem with refactor: simplify merlin diagnostics #1005 indeed, but let's also wait for @copy's feedback

[rgrinberg]

Yes, I meant the other PR. @copy please re-open if you can still repro the bug.

[copy]

I can confirm that this is fixed, cheers!