[notebook] deduplicate completion results

[Carreau]

remove context completion that are duplicates from introspection
completion

fixes #1840

[Carreau]

made a mistake, close, will reopen.

[Carreau]

@fperez , this should remove the "duplicates" completions by favoring introspection completion from context completion.

There might (really not sure) be some false positive of 'duplicates', so some context completion might not show up in rare case, but no completion coming from live object introspection will be discarded.

[fperez]

The functionality seems to work correctly, thanks! I only have a couple of comments:

1. You are using code in this style of brace handling:

    function _existing_completion(item, completion_array)
    {
        if(item.substr(0,1) == '.')
        {
            for( var c in completion_array )
            {

but in general I see our JS files use the more traditional placing of the opening brace in the same line:

    function _existing_completion(item, completion_array) {
        if(item.substr(0,1) == '.') {
            for( var c in completion_array ) {

so it should probably conform to that.

1. I don't know enough JS to say anything about the actual implementation of the filtering. It looks a bit complicated to me, in Python I would have simply done right before returning something like:

return sorted(set(matches))

thus ensuring uniqueness, rather than doing all the per-item check you are doing. But again, I don't know much JS, so perhaps there's no cleaner way to do it. @ellisonbg, @minrk, thoughts on this one?

[minrk]

There doesn't seem to be something like set in javascript, but a quick Googling reveals a pretty simple sort/unique transform, that could probably be even simpler if you know that everything is a string:

http://stackoverflow.com/questions/4833651/javascript-array-sort-and-unique

[Carreau]

Well, issue is that both introspection and context completion does not return the same thing.
If you try to complete a.b.<tab> kernel will return a.b.something whereas coompletion based on codemirror token will only return .something , so as the 2 kinds of completion are not the same length you have to keep the replace range tightly linked to the completion, then i'm sorting and comparing dict based on their keys.

And as the 2 keys are different I have the deduplication try to match only the ends of the strings, but only when one of them start with a dot, because obviously, you don't want to find that width is the same as set_width.

Still I'll see If I can shorten the logic.

As for the brace, i'll never get used to it, will fix it.

[fperez]

Great, thanks. Ping us when you've updated the logic and we'll revisit, I'd really like to see this merged.

[Carreau]

Well, I tried, there is one check that I can omit, but I don't see how I can deduplicate faster. Sorting does not help, uniqueness is not well enough related to comparing items.

assuming I want methAfrom object object. Then object.<tab> gives me :

• object.methA # from kernel
• object.methB # from kernel
• .methA # from codemirror.
• .methB # from codemirror.

Which is already sorted. and obviously the duplicates are not next to each other... so you can't test for equality, or use hash, set ....

If you have any more ideas ... or i'm good to go

[ellisonbg]

I think this can go in as is.

[fperez]

I'm OK too if there's no better option, I just wanted to look into it. Thanks for taking the time to investigate, @Carreau. I'm merging now b/c this bug is really annoying me.