Topic: [Feature] Don't use Elasticsearch for BUR scripts

Posted under Site Bug Reports & Feature Requests

Requested feature overview description.

When using the "Script" field of the BUR search form, do a literal text search instead of using Elasticsearch.

Why would it be useful?

Elasticsearch is annoying enough when trying to find old tag discussions in forums, due to its refusal to recognise underscore characters and its attempts to broaden any query to the point of near-uselessness. There is no excuse for using it in the far more limited context of BUR search, where the only thing you will want to search is tag names - precisely the thing Elasticsearch fails horribly at handling.

For an example, I was recently trying to determine how unpleasant implementing the implication changes proposed in topic #36648 would be. I looked at the legwear wiki and spotted, to my despair, a long list of colour and pattern implications. Clicking on one of those revealed that e.g. black_legwear also had a list of implications, including from black_leggings. So at that point I thought it would be a good idea to search the BURs to find out when such an implication may have been created and if there was any relevant chatter at the time, or at least someone to blame, since a lot of people seem to be in agreement that leggings should be reclassified to bottomwear.

For my first attempt, I tried searching for leggings. This returned a list of every BUR that included the word "leg".

So then I narrowed it to black_leggings. This returned a slightly smaller list of every BUR that included the word "black" and the word "leg", with no requirement that they appear in the same line let alone the same tag.

Then I gave up and used the regular implication search instead, which actually does function correctly. No prizes for guessing who why I wasn't having much luck finding public discussion.

What part(s) of the site page(s) are affected?

BUR listing page

Only posts and post versions use elasticsearch, nothing else. What you are experiencing is called stemming.

Of course it doesn't make much sense for these searches. You can force the expected behaviour by using a wildcard search, like *leggings*. I will see about making this the default for BURs in the future.

  • 1