Topic on User talk:Magnus Manske

Jump to: navigation, search
Summary by ArthurPSmith

Magnus has fixed searching, hurray!

ArthurPSmith (talkcontribs)

Hi Magnus - something seems to have changed in the search function in Mix n match within the last week or so:

  • previously it would match partial words, now it seems to match whole words only
  • short words like ''of', 'for', 'the', or initials (A. , etc.) NEVER match now - I have to remove them from the search string before I can get matches

There was a change slightly earlier that seemed to start matching more generously, so it looks like you've been playing around with it for a few weeks - any details or updates you can share? And the 'of' thing is a little annoying...

ArthurPSmith (talkcontribs)

I guess the whole UI has changed over the last few days - it looks nice, but have you discussed the changes anywhere? Thanks!

Nono314 (talkcontribs)

The UI change has been advertised through a blog post at http://magnusmanske.de/wordpress/?p=441

However there was also a less advertised move to using full text search that was probably motivated by the legacy search becoming really slow. However this may be a bit hasty and not so thoroughly tested as needed, as I experienced the same downsides as you lately.

Restrictions are described at https://mariadb.com/kb/en/mariadb/fulltext-index-overview/.

The most blocking one is the fact that words with length <= 3 are excluded from index. This may be a good thing since they were sometime causing issue with the previous search, but obviously they should also be excluded from search string.

Note that all other words are now mandatory, which should suppress the flood of false positives that where overwhelming at some point.

I also found another issue: when name parts are separated by dashes in external names, they will not be found if the query string has spaces instead. This is really blocking for French names where both characters are commonly used between given names.

ArthurPSmith (talkcontribs)

Hmm, while people are working on search (if somebody is) - is there any chance to allow the Mix n match search results to be paginated, so it's not limited to the first 20 or however many it shows? That would be a big help for some purposes...

Magnus Manske (talkcontribs)

I didn't realise this would cause so many issues...

  • Short (and overly long) words are now removed from search automatically
  • Words with dashes ("Jean-Antoine") are now split ("Jean" AND "Antoine") before search (even though this should really be a bug report for the French language...)
  • I have raised the search result limit from 25 to 100. No pagination yet.
ArthurPSmith (talkcontribs)

Excellent on all counts, thanks!!!