bug: Unable to ban numbers about obscenity

[jo3-l]

The issue you're seeing here is caused by the collapse duplicates transformer, which is included in the set of recommended transformers. After this transformer is applied, the original input 666 becomes 6 which no longer matches the pattern 666; there's a warning to this effect in the documentation of the transformer:

This transformer should be used with caution, as while it can make certain
patterns match text that wouldn't have been matched before, it can also go
the other way. For example, the pattern hello clearly matches hello, but
with this transformer, by default, hello would become helo which does
not match. In this cases, the customThresholds option can be used to
allow two ls in a row, making it leave hello unchanged.

Thus to avoid this behavior you can either do as is suggested above and add a custom threshold that avoids transforming 6, or disable the transformer entirely.

As the behavior is documented I'm inclined to mark this as not a bug, though I absolutely see how the library behaves unintuitively in this case and would be open to considering suggestions for improvement.

[jo3-l]

This one is really quite unfortunate. The issue is again due to one of the default transformers, which skips non-alphabetic characters; this allows patterns such as foo to match f_o_o. Hence 8562 is transformed to the empty string which matches nothing.

While this is not a bug per se I think this is extremely unintuitive behavior and I'm inclined to remove this transformer from the set of recommended ones for English text, or perhaps even removing "recommended" transformers in general. The original intent was to provide something that works nicely out of the box, but as your report shows once anything custom is added it's easy for it to become confusing. WDYT?

[jo3-l]

Also, on an unrelated note — would you be willing to comment on which parts of the Obscenity API you use in your project? In particular, I'm curious as to whether you use any of the more complex features (custom transformers, censors, etc.) The reason I'm asking is now that I have a bit more time to work on open source, I'm thinking about cleaning up the API a little and releasing a v1.0.0, but if users are tied to the existing API I'll hold off on breaking changes.

[jo3-l]

Thanks for the response, that's helpful!

Back to the issue: I anticipate that I'll have time to work on releasing a fix next weekend, but for now you should be able to work around it in your application by passing a custom set of transformers instead of englishRecommendedTransformers.

[jwoertink]

Thanks for responding! That makes sense for 666, but I think this also fails for any number. For example 8562.

To add a bit more context to this, I'm using this in a live chat filter for my site. The chat exists on a live-stream page where the streamer can set which words they'd like to ban. This came up because a streamer had set the different parts of their address as "banned words" to avoid being doxxed in real-time. So for example, "8562" is banned

const customDataset = new DataSet();
customDataset.addPhrase((phrase) => {
  return phrase.setMetadata({ originalWord: "8562" })
    .addPattern(parseRawPattern("8562"))
    .addPattern(pattern`|8562|`)
});
const matcher = new RegExpMatcher({
        ...customDataset.build(),
        ...englishRecommendedTransformers
      });

matcher.getAllMatches("8562").length //=> 0

Though, this may just be because I have my patterns wrong... You've done a really great job with adding docs, so I may just be missing something.

[jwoertink]

I think your idea makes sense.

As for the API, I don't get too crazy with it. There's probably a lot more I should be doing with it that I'm not currently.
For the most part, that code I pasted above is essentially it. We have 2 separate functions where the first one is our default banned words.

import {
  DataSet,
  RegExpMatcher,
  englishRecommendedTransformers,
  pattern,
} from "obscenity";

const joystickDataset = new DataSet()
  .addPhrase((phrase) => {
                // they generally follow this pattern
		return phrase
			.setMetadata({ originalWord: 'badword' })
			.addPattern(pattern`badword`)
			.addWhitelistedTerm('...') // some close related word
	})
  .setMetadata({ originalWord: 'otherword' })
			.addPattern(pattern`oth[e]rword`)
			.addPattern(pattern`|oth|`)
			.addPattern(pattern`|word|`)
			.addPattern(pattern`?therword`)
const matcher = new RegExpMatcher({
  ...joystickDataset.build(),
  ...englishRecommendedTransformers,
});
export default {
  methods: {
    isTextInViolation(text) {
      return matcher.getAllMatches(text).length > 0;
    },
  }
}

Then we have the second function where a streamer can specify which words or phrases they want banned

bannedChatWordsDataset() {
  const customDataset = new DataSet();
  this.streamer.bannedChatWords.forEach((item, idx) => {
    const word = item.toLowerCase();
    customDataset.addPhrase((phrase) => {
      return phrase
        .setMetadata({ originalWord: word })
        .addPattern(parseRawPattern(word))
        .addPattern(pattern`|${word}`)
        .addPattern(pattern`${word}|`)
    });
  });

  return customDataset;
},

bannedChatWordsMatcher() {
  return new RegExpMatcher({
    ...this.bannedChatWordsDataset.build(),
    ...englishRecommendedTransformers
  });
},
isMessageInViolationOfBannedChatWords(message) {
      return this.bannedChatWordsMatcher.getAllMatches(message).length > 0;
    },

We do still have a few bugs around this implementation. I believe there's some words where you can't say just the word, but if you use it in a sentence, then it goes through... I have to come back to that portion later.

Stoked to hear you'll have more time on this though! It's a great project ❤️