How to use Discourse regexes with watched words?

[rizka]

I’ve worked on censoring the most vulgar swear words with regular expressions today. Why regexes? Well, in Finnish and other Uralic languages like Hungarian and Estonian words are inflected. A single swear word could have maybe thousands of mutations, which is why it is awesome to have the ability to use regex patterns. It is also no coincidence that it was another Finn who proposed this originally.

Support for wildcards in word censoring

I don’t know if anyone actually uses this for other than fun but due to inflection it would be nice to be able to use wildcards when adding censored words.

I need some quick advice about which regex flavor Discourse uses. I experience some unexpected behavior with non-alphanumeric characters which is awkward especially because of the common letter ä in the Finnish alphabet. I got the regex into pretty good shape by basic knowledge about regexes and the method of trial and error, but for an even better result, I would need documentation or something.

[Falco]

You can read about it in the source code.

[elijah (Eli the Bearded)]

Reading that, I don’t see much about them except to see that they are Javascript regular expressions. (I would have assumed Ruby without that link.) So a Javascript reference would be in order.

RegExp

Which has internal links to specifications, if you want to go deeper.

[Mittineague]

which regex flavor Discourse uses.

AFAIK, Ruby and Postgres support POSiX

https://ruby-doc.org/core-2.2.0/Regexp.html

https://www.postgresql.org/docs/9.3/static/functions-matching.html

[rizka]

Cool, thank you for your replies all. I’ll look into them and return if I still can’t figure it out.

[justin (Justin DiRose)]

Discourse Regexes (Watched Words)

To use regular expressions (regex) in watched words you must first turn on the watched words regular expressions site setting.

Discourse by default matches all uppercase and lowercase forms of a word entered as a regular expression. That is,

thread

This will match thread, THREAD, and thReAd.

(t|7)hr(3|e)(4|a)d

This will match all of the cases above, plus thr3ad, 7hread, and thr34d.

threads?\S+

This will match thread and threads but not threaded or threading.

However, there’s a glaring error in ALL the above examples! The words threadlike and unthreading are matched (un▪️▪️▪️▪️▪️ing), even though they’re not referring to thread. How do we fix that?

We’d have to amend our regex to handle word boundaries.

\bthreads?\b

This looks for boundaries around the word so that unthreading or threadlike aren’t caught by the filter, but thread and threads still are.

For handling Unicode characters

gr(ü|ue)(ß|ss)e

This matches all commonly spelled forms of the word grüße — including gruesse and GRÜSSE

Say we want to block the word Über, but not Übersicht. Using word boundaries like \b(ü|ue)ber\b doesn’t work because some of the JavaScript regex word flags don’t handle Unicode characters.

Instead we have to make our own boundaries.

(?:^|\s)(ü|ue)ber\b

This will now appropriately match Über and ueber, but not Übersicht or uebersicht.

A final warning

Regex is extremely powerful and thus dangerous. An incorrectly written regex statement can cause issues for your users. Test your regex statements on non-production instances before going live.

[supermathie (Michael Brown)]

@justin

Test your regex statements on non-production instances before going live.

If you want to get more serious (or silly) about this kind of thing, you can introduce formal test cases. For example, I’ve put @justin’s über-excellent example onto regex101.com: https://regex101.com/r/4ano0r/1/tests

If you do so, ensure you switch the regex flavour to ECMAScript:

hey, they spelled flavour wrong