Work with proposed updates to julia-mode LaTeX completion

[non-Jedi]

JuliaEditorSupport/julia-emacs#100 changes
LaTeX completion in julia-mode to work by a combination of
completion-at-point-function and abbrev-mode. This is the necessary
change to give full-featured LaTeX completion with emacs-jupyter julia
buffers with the PR in its current form.

Unlike #201, this is actually working. Obviously shouldn't be merged until/if JuliaEditorSupport/julia-emacs#100 is merged, but since I've already done the work, I thought it worthwhile to publish here.

[non-Jedi]

The LaTeX symbol completion rework has been merged into julia-mode, so this needs to be fixed now. Unfortunately, this PR isn't working as is. It seems like something is clearing out completion-at-point-functions because neither of julia-mode-latexsub-completion-at-point-around nor julia-mode-latexsub-completion-at-point-before are even getting called.

Quickly skimming through the code in jupyter-repl.el, I'm not seeing why this would be the case though. @nnicandro, any guidance on why adding things to completion-at-point-functions to run before jupyter-completion-at-point wouldn't work? I'm fairly confident this PR was working when I first created it, so something has probably changed in how completion works with emacs-jupyter since then.

[non-Jedi]

Further investigation. Normally if I describe-variable on completion-at-point-functions in jupyter-repl-mode running the julia kernel, I get what I want with the latexsub functions running first:

(julia-mode-latexsub-completion-at-point-around
 julia-mode-latexsub-completion-at-point-before
 jupyter-completion-at-point
 t)

But if I add advice to completion-at-point to see the value of completion-at-point-functions immediately before completion:

(advice-add 'completion-at-point :before
            (lambda ()
              (message "completion-at-point-functions before completion-at-point called: %s"
                       completion-at-point-functions)))

I see that jupyter-completion-at-point got moved to the top of the list:

(jupyter-completion-at-point julia-mode-latexsub-completion-at-point-around julia-mode-latexsub-completion-at-point-before t)

Interestingly, sometimes invoking indent-for-tab-command will cause completion-at-point to be called twice, and the second time completion-at-point-functions has the latexsub functions first.

I still don't see what in the code would be causing this rearrangement of completion-at-point-functions.

[nnicandro]

I'm not sure what could be going on. The completion code hasn't changed in a long time. Have you tried using something like add-variable-watcher on completion-at-point-functions to check what function is changing the value of that variable, e.g. you can put a call to backtrace in the watch-function. Note there is also remove-variable-watcher to undo the watcher.