Popup shows unwanted completions when triggered manual

[robinknaapen]

I'm currently using

let g:cm_auto_popup = 0
imap <C-Space> <Plug>(cm_force_refresh)

When I use this I notice that I get results in some of my completions that make no sense to me.

My expected result would be the same as with cm_auto_popup enabled

I've found a workaround for this issue. When looking through your source code I found out that changing cm_force_refresh fixed my issue.

func! cm#_force_refresh()
    " force=1
    call s:notify_core_channel('cm_refresh',g:_cm_sources,cm#context(),0)
    return ''
endfunc

What I did was change the 1 into 0.

I'm using nvim-cm-tern in my examples

[robinknaapen]

Concluded that this is a terrible workaround... This only worked in JS for me. In Golang this will result in NCM not finding any completion; In my experience.

[roxma]

That's line completion. https://github.com/roxma/nvim-completion-manager/issues/24

Line completion is a non auto popup source, it's only triggered by <Plug>(cm_force_refresh)

It has lower priority so it's at the bottom of the menu.

You could disable it with g:cm_sources_override

[roxma]

I'm wondering:

• Do you think we should add a mark to each item so that you know which source it's comming from, and then you could disable it?
• Do you think manual completion should popup the same results as auto-popup?
• How should manual completion behave while keeping the line completion source?

---

Here is a brief explanation of current behavior:

<Plug>(cm_force_refresh) activates all sources according to current file type. And it ignores the sources's cm_refresh_length. For example, with buffer keyword completion,

It auto pops up when you has typed four characters foob|, by default, But with manually completion, it's triggered for only one character. f|

Activates all souces could be annoying. But it seems it's needed in some use cases, for example, when you don't know what more to type to trigger the menu.

[roxma]

My proposal:

1. Add a mark into each completion item.

2. We could setup another config variable cm_manual_refresh_length, like cm_refresh_length, but only used for manual completion.

3. As for the line completion source, it should only be triggered by manually, and only when the cursor is at the beginning of a word, before typing a character. It should be hidden when suggestions with higher priority is available.

// cc @rafaeln

[roxma]

fixed in ncm2 <Plug>(ncm2_manual_trigger) has almost the same behavior of auto trigger