@alexandrudima Can you point to me the registration?

@jrieken Maybe this is caused by using the registries and no longer using suggestSupport. Should we add a weight property on registration?

Here is the default textual completion: https://github.com/Microsoft/vscode/blob/master/src/vs/editor/common/modes/abstractMode.ts#L233

I also suggest using find all references on SuggestRegistry, maybe other places need to have a lower weight?

I don't like the weight option. I would register for the * language which results in a lower score: https://github.com/Microsoft/vscode/blob/master/src/vs/editor/common/modes/languageSelector.ts#L34

Thank you @aeschli !

Hm, I kind of expected an easier fix. The problem of the higher score is that the TextualSuggestSupport is contributed for every mode id.

modes.SuggestRegistry.register(this.getId(), new TextualSuggestSupport(this.getId(), editorWorkerService, configurationService));

If you would just contribute a single time for '*', it would get the score it had before
Also it seems the TextualSuggestSupport doesn't really need to know the mode id.

Actually, the problem is bigger than textual suggestion. Basically, all language features that don't work with merging results (suggest, formatting, etc) were affected cos before they were known to be coming from a built-in language and getting the lowest score. For instance, an extension adding/replacing formatting for html would be broken by that change. That's why we restored the old implicit knowledge.

Once, everything moves to the extension host, the problem goes away because you top the score by coming later which you do by taking an ext-dependency