Oniguruma docs simply state: "Do not pass invalid byte string in the regex character encoding."

Based on my testing, following are the behavior details for invalid standalone encoded bytes:

• Standalone \x80 to \xBF throw error "invalid code point value".
• Standalone \xC0 to \xF4 throw error "too short multibyte code string".
• Standalone \xF5 to \xFF fail to match anything, but don't throw. This feels like a bug.

The behavior changes if an invalid standalone encoded byte is used as the end value of a character class range (due to option ONIG_SYN_ALLOW_INVALID_CODE_END_OF_RANGE_IN_CC):

• Standalone \x80 to \xBF are treated as \x7F.
• Standalone \xC0 to \xF4 throw error "too short multibyte code string". This feels like a bug.
• Standalone \xF5 to \xFF are treated as \x7F.
  • If the range is within a negated, non-nested character class (ex: [^\0-\xFF]), \xF5 to \xFF are handled like \x{10FFFF}. This feels like a bug. For example, [^\x01-\xFF] matches only \x00. The bug can be worked around by nesting the class. For example, [[^\0-\xFF]] matches \x80 and higher.

Disclosure: I'm testing using Oniguruma 6.9.8 via vscode-oniguruma. However, the release notes for subsequent versions don't mention any related changes.