high95%
CharacterTextSplitter/RecursiveCharacterTextSplitter corrupt chunk content when merging with non-lookaround regex separator and keep_separator=False
Using a consuming regex separator (e.g. \s+) with is_separator_regex=True and keep_separator=False causes both CharacterTextSplitter and RecursiveCharacterTextSplitter to insert the literal regex pattern string as a merge separator, producing chunks that contain visible regex syntax and are not substrings of the source text.