Recent changes break whole word search

[nebulade]

Description

The latest release (1.7.5) breaks some search patterns for whole words. In this case typing restor in the search field correctly shows matches for the whole word restore while when the whole word is typed into the search field, the search results are empty.

This can be tested with a small md file with the following contents:

## Restore

Apps can be restored to a previous backup by clicking on the `Restore` button.

The previous version (1.7.4) does not have this issue.

[squidfunk]

This has to do with the default stemming by lunr, fixed in master which disabled stemming, and about to be released.

[nebulade]

Thats awesome. Thanks!

[squidfunk]

Fixed in 1.8.0

[kevung]

Hello everybody,
I encounter the same problem as @nebulade with mkdocs-material-5.2.2.
Are there other people noticing this regression?

[kevung]

The problem I encounter seems to be a problem from Mkdocs-Material.
The search is fully working when the language is in english, but fails in french, for example.

The search is fine with a mkdocs.yml:

site_name: test

theme:
  language: en
  name: material

plugins:
  - search

use_directory_urls: false
nav:
  - Intro: index.md

but does not work with this mkdocs.yml

site_name: test

theme:
  language: fr
  name: material

plugins:
  - search

nav:
  - Intro: index.md

[squidfunk]

Possibly related to #1603 and #1675

[connerxyz]

Hello, I'm experiencing a very similar issue with 7.0.4. I've read through this issue and #1603 and #1605.

I get what appears to be no search results for certain whole words which appears to be because of stemming, (e.g. "refer" returns results for partial matches of "reference", but "reference" itself returns no results).

I noticed the comment that stemming was disabled... and I saw [the suggestion] to extend the theme and override en.html partial... But I don't see any stemming there. I experimented with removing the trimmer from the pipeline without any affect

I understand there's a difference of mkdocs vs. the material theme, and complexity of accommodating search use-cases across many languages.

What I don't understand is why whole words wouldn't be indexed in addition to stemmed words? From a user POV you want to be able to search for whole words (regardless of stemming) and aren't these whole words defined by the separator?

[squidfunk]

Thanks for reporting. If you can narrow it down to a small reproducible case and two versions that work / don't, we can debug it. The official docs work with that very use case you mentioned, i.e. "reference" returns results.

[connerxyz]

I've found the problem has something to do with prebuild_index: true.

Not working example.

Search configuration in mkdocs.yml looks like...

plugins:  # https://github.com/mkdocs/mkdocs/wiki/MkDocs-Plugins
  - search:  # https://squidfunk.github.io/mkdocs-material/plugins/search/
      lang: en
      separator: '[\s\-\.]+'
      prebuild_index: true
  - macros

Working example

plugins:  # https://github.com/mkdocs/mkdocs/wiki/MkDocs-Plugins
  - search:  # https://squidfunk.github.io/mkdocs-material/plugins/search/
      lang: en
      separator: '[\s\-\.]+'
  - macros

[connerxyz]

PS – How do you enable the query string /?q=reference? Thanks.

[squidfunk]

@connerxyz ah yeah, prebuilding the index does not honor the pipeline, which excludes stemming by default. I think I also mentioned this on some issue on the MkDocs issue tracker some time ago.

PS – How do you enable the query string /?q=reference? Thanks.

Deep linking of search results is one part of sharing search results, currently only available via Insiders 😉

[connerxyz]

Good stuff. I love your funding model, with the goals etc. This is a great project/community. Thanks.