| LuisVilla | |
| Sep 18 2015, 11:51 PM |
| F4417218: Screen Shot 2016-08-30 at 11.58.53.png | |
| Aug 30 2016, 6:59 PM |
Just added a link for Queen Esther Marrow; had misspelled it as Queen Ester Marrow (missing "h").
A search on-wiki for "Queen Ester Marrow" (misspelled, no h) correctly suggests "did you mean Queen Esther Marrow"?
Could the link dialog do the same auto-suggest, at least when there are no results?
Could the link dialog do the same auto-suggest, at least when there are no results?
Yes, if we threw a huge amount of money at the search cluster, but that's really a question for Discovery.
Hard to say. In principle I think it should be fine from a performance standpoint, but I'd suggest doing some kind of load test.
I should point out that the last time I did an audit comparing prefixsearch to fulltext, it came up pretty inconclusive which one is better. This bug points out one of the cases where fulltext is better, but there are numerous cases where prefixsearch is better. So I'd also suggest doing some kind of test to validate the assumption that fulltext is actually better than prefix for the link dialogue.
What if this was just switched to the new completion suggestion api instead?
https://en.wikipedia.org/w/api.php?action=cirrus-suggest&text=Queen+Ester+Marrow
This api is speifically setup to take thousands of api calls per second to replace the existing type-as-you-search.
note i don't think VE should call cirrus-suggest directly, we are integrating that into the core opensearch. If necessary we can add a new api call in core specifically for suggestions, the cirrus-suggest api is *experimental* and will be deleted.
Definitely a possibility too, although realistically that API is not yet validated by Discovery as being production-quality from a product perspective.
All to what Erik said: full-text search is not designed for search as you type and should be ideally triggered only when the user has done typing his query.
Completion suggester is certainly the right solution to do this as it is designed for search as you type and can handle small typos.
I'd suggest not to implement a solution based on full-text search.
This task is not actionable by Discovery at the stage, so I removed it from our sprint and put it back in the backlog.
The solution in future would probably be:
Realistically, this is not a super high priority issue for either Editing or Discovery at present, so I wouldn't expect much movement on either end.
Now that the completion suggester has been released to production, this should fix the kind of issues that this task was aimed to fix. This significantly improves the search in VisualEditor (and for other search clients). As such, closing this task as "declined" since we have another solution that works in production.
FWIW, the autosuggester fixes the problem I had when I initially filed this bug, so thanks! and agree with the close.
Proof:
I'd actually say this is effectively Resolved by Discovery, but that's Deskana's call. :-)