@alexcrichton pointed out that the error messages the current scheme will evoke will be unfortunate in ptr+length languages such as Rust in some cases where the Rust program has mutated the ARF string in an ARF-unaware way.
For an ARF translation like "hello\xffworld" => "\u{feff}hello�world\0hello\0\x7fworld", if Rust code appends a suffix such as ".txt", it'll produce "\u{feff}hello�world\0hello\0\x7fworld.txt". In this case, the embedded NUL will trigger ARF validation, and the validation will fail since the second half ends with ".txt" and the first half doesn't. A typical error message will include the name, which in this case will display as "hello�worldhelloworld.txt", which may confuse users.
Some brainstorming:
- Encode the entire second half in non-printing characters?
- Encode strings in the second half as something like base64 which would be printing but less likely to be mistaken for a corrupted string? We could get it to print as something like "hello�world-aGVsbG8gd29ybGQK.txt".
- Leave it as is?
@alexcrichton pointed out that the error messages the current scheme will evoke will be unfortunate in ptr+length languages such as Rust in some cases where the Rust program has mutated the ARF string in an ARF-unaware way.
For an ARF translation like "hello\xffworld" => "\u{feff}hello�world\0hello\0\x7fworld", if Rust code appends a suffix such as ".txt", it'll produce "\u{feff}hello�world\0hello\0\x7fworld.txt". In this case, the embedded NUL will trigger ARF validation, and the validation will fail since the second half ends with ".txt" and the first half doesn't. A typical error message will include the name, which in this case will display as "hello�worldhelloworld.txt", which may confuse users.
Some brainstorming: