The situation is somewhat simpler than that, but a full solution is a bit harder. Punctuation is currently intentionally stripped from search terms, which was implemented in 2015 in f132dfd and 815e021. This allows for certain fuzzy matching such as up-lighting (with punctuation) → uplighting (without).

Of course, this means that searching c^o&n$c@e#r!t yields results for concert, but it doesn't currently work the other way: searching rm (without punctuation) does not yield results for R&M (with) because Sphinx tokenizes the potential matches by splitting at any character not in charset_table:

Line 15 in e33b0f4

charset_table: "0..9, A..Z->a..z, _, a..z"

Therefore, R&M simply can't be directly matched as a single keyword. Likewise, a search for dean's finds a single event spelled Deans, while the only way to find the many events actually spelled Dean's is to tokenize the search term as something more like dean s.

Rather than explicitly allowing an arbitrary subset of punctuation, some combinations of charset_table and ignore_chars might be better than others here, but the most flexible scenario would be for the punctuation to be strictly optional on both ends. Unfortunately, ignore_chars only handles treating punctuation one way or the other:

For example, if '-' is simply not in the charset_table, "abc-def" text will be indexed as "abc" and "def" keywords. On the contrary, if '-' is added to ignore_chars list, the same text will be indexed as a single "abcdef" keyword.

A simpler solution that handles the single case of R&M (and its variants) might be to add exceptions.

As a work-around, searching r m finds R&M events whether they were spelled R&M or R & M.

Searching "r&m" has 0 results #417

Description

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions