I have found a similar behavior in Spanish. The problem here is that characters with diacritics are removed, but no error is triggered.

For example, the text:

# Note the second letter starting from the left
Sí, efectivamente, el Servicio de Vigilancia

It is transformed into:

S, efectivamente, el Servicio de Vigilancia

In my case, this is the code I have used

df['tweet'] = df['tweet'].apply (lambda x: p.clean (x))

The dataframe is a readed from a CSV file in UTF8 (without BOM)

I think the offending option that destroys everything that is not an English character is:
def preprocess_emojis(self, tweet_string, repl):
processed = Patterns.EMOJIS_PATTERN.sub(repl, tweet_string)
return processed.encode('ascii', 'ignore').decode('ascii')
There should be a better way to clean emojis
And there is:
https://github.com/carpedm20/emoji
Maybe this library should be in charge of demoji-fying, although it stubbornly adds aliases, like: :flexed_biceps:

The same issue appears with Arabic texts using windows 10.
if the emoji option is on it will delete all characters.

Can confirm the same for Hindi and Codemixed Hindi-English as well.

Also removes all Korean text!