Leaving insert-mode (ESC) inserts an underscore one char left of point

[PorcelainMouse]

This happens now when editing a .tex file. I type a word and then hit ESC to exit incert-mode, which takes me out of insert-mode, but first it inserts an underscore in between the last and second-to-last character of the word I just typed.

I just updated my plugins with PlugUpdate, and now I have this strange behavior when editing latex files. I know that it is related to vim-latex because, when I comment out this plugin, this erroneous behavior stops. But, it could obviously be an interaction.

And, it seems to be related to completion. It doesn't happen when completion has finished searching for suggestions. So, if I finish a word, then put a space, then hit ESC, it doesn't happen. Also, if I wait after typing a word until completion gives me a list of suggested completions (or none), then when I hit ESC, it doesn't happen. It's like I'm interrupting completion and that's what causes it, but only if vim-latex is active.

Oh, I'm using deoplete completion.

[gerw]

What happens if you disable 'deoplete completion'? I think this is a strange interaction between both plugins.

[PorcelainMouse]

I don't fully know how to do it correctly, but I temporarily disabled the deoplete plugin, itself, by removing it from my config file so that it wouldn't get started. That caused some errors, but it worked, meaning the problem symptom was not observed in this condition.

Looking over my configuration, though, I noticed I had an option for deoplete completion set: auto_complete_delay. With this value set to 100 (ms) or greater, the symptom is observed, with it set to 80 or less, the symptom does not occur. So, a workaround is to reduce this delay.

The delay prevents nvim from trying to automatically complete everything I type constantly. This is especially useful in LaTeX, since it can get a little bit in the way when I'm composing regular prose. So, it would be cool to have this conflict resolved. It was originally 300 ms, but I don't know what is the default value; 0 I suspect.

[PorcelainMouse]

Okay, this is actually a bigger problem that I noticed at first.

'cw' is also broken. Try 'cwESC' and it get 'newwor_d', even with the previous workaround of setting the deoplete delay to a low value (even 0). Do you not see this too? I'm confused. Obviously this is not usable with cw broken.

[PorcelainMouse]

Oh wait; it is still related to the same thing. it's just that I'm really fast with 'cw' and I beat deoplete finding the replacement word even with delay set to 0. Okay, so it's still the same issue.