bug: handle_leading_whitespace seems to delete all whitespace (?)

[benlubas]

with 0 indent at all, it's fine:

With 2 space indent on both lines it's fine:

But with 0 leading indent on the code block itself, I do not get correct completions when I'm in an if block.

[benlubas]

Hmmm. It seems like it works if there aren't any 0-indent blocks, and only fails when there are a mix of 0-indent and n-indent code blocks in the document

it's very inconsistent

[jmbuhr]

I assume the code is all there in the otter buffer, so our keeper.sync_raft code works?

Can you vim.print the params of the completion requests before and after we modify the positions here:

otter.nvim/lua/otter/lsp/init.lua

Lines 142 to 144 in 5f03444

	-- take care of potential indents
	keeper.modify_position(params, main_nr, true, true)
	-- send the request to the otter buffer

Maybe we are missing something. If this is not the case it is something harder to spot.

I previously had issues with the other lsp servers like pyright being out of sync, see

otter.nvim/lua/otter/init.lua

Lines 125 to 129 in 5f03444

	-- this has to happen again after the
	-- otter buffers got their own lsps
	-- to really make sure the clients are
	-- attached to their otter buffers
	keeper.sync_raft(main_nr)

What was happening there is as follows. Lsp servers don't actually look at the buffer (most of the time, for pyright we even disable it explicitly due to a bug https://github.com/jmbuhr/quarto-nvim-kickstarter/blob/6455ad9123041f11e3294508db8f216223e2ca8a/lua/plugins/lsp.lua#L288-L296) content. The client (neovim) sends a didOpen notification with the content of the buffer to the server and every time the content is changed it sends the changes over. In our case, we don't need to send these notifications, they get sent automatically by nvim when we change the otter buffer because pyright is attached to that buffer (that's mostly what being attached entails). But if for some reason one of these notifications gets lost the server gets confused. Just something to keep in mind in case nothing else makes sense.

[jmbuhr]

Also, I can not reproduce this with ./tests/debug_example.sh 05.norg

[jmbuhr]

What file do you experience the issue with and how do you activate otter?

[benlubas]

Here's the file:

file

@code python
y = []
y.pop()
if True:
  y.append(1)
@end

* Heading
  @code python
  print("hi there")
  x = []
  y_2 = x.copy()

  y.pop()

  while True:
    x.clear()
    for i in range(10):
      x.append(i)
    break
  @end

I activate it with a dev version of the Neorg otter integration, which is here. The relevant lines are:

otter.activate(
    module.config.public.languages, -- I have set to { "python", "lua" }
    module.config.public.completion.enabled, -- true
    module.config.public.diagnostics.enabled, -- true
    nil
)

The above integration is only providing user commands and making sure that handle_leading_whitespace is set to true.

[benlubas]

I can print and inspect in a little bit, I've got a bunch of random stuff to do today, you don't have to look into this for now, I'll try to get a consistent reproduction

[jmbuhr]

Ok. As another test, it is always interesting if activating otter manually after opening a file makes a difference compared to an autocommand or ftplugin, to rule out races with other plugins (like in this case: #154 (comment) ).

[jmbuhr]

And while we are at it, I want to eventually take the otter.keeper.modify_position code and split it up explicitly for each request method, such that we don't unnecessarily check params for all sorts of positions for every request, so if destructuring this happens to become part of your debugging process I'd be happy about a PR :)

[jmbuhr]

Something else must be going on here. If I take your file and and just add an autocommand to activate otter for norg

vim.api.nvim_create_autocmd({ "BufReadPost" }, {
    desc = "Activate Otter on Buf Enter",
    pattern = "*.norg",
    group = vim.api.nvim_create_augroup("NorgOtter", {}),
    callback = function(_)
      require'otter'.activate()
    end,
})

it just works. Same as activating manually.
However, I don't have anything else configured for Norg. Does the Norg plugin modify the buffer in any way after opening?

[jmbuhr]

Btw. as you can see in the autocommand, you don't have to necessarily pass languages to otter.activate, it creates otters for all languages found (and for which it knows file extensions) by default.

[benlubas]

Neorg doesn't alter the buffer after opening.

[benlubas]

I'm unable to reproduce it in ~5 minutes of playing around with it. I'll reopen if I find a reproduction