why do you expect base64 strings to be detected as whatever multibyte encoding?

it's a plain string - not more and not less and "BASE64" is not a multibyte encoding at all

From reading the documentation I reached this assumption

In  the docs it says for mb_check_encoding

Checks if the specified byte stream is valid for the specified encoding.

A list of supported encodings I can get from

mb_list_encodings which will also include base64 

So I would assume mb_check_encoding would be able to handle all encodings that mb_list_encodings shows

I believe the issue is that mb_check_encoding() works on the assumption the input comes from a stream, meaning there may be more input to read after what was passed to it. "test" encodes to "dGVzdA==" which is a valid base-64 string, however =s are strictly used as padding at the end of the output and so cannot appear in the middle of a stream.

As such, if you rtrim the =s off then mb_check_encoding will report the "dGVzdA" is valid. But then that puts the responsibility of validating the number of =s on the user.

However, I don't think that makes this behavior correct. If it were reading a UTF-8 stream and the string ended in the middle of a multibyte sequence then it should report as valid: there may be more to read that would complete the sequence. Valid "so far". The case of a padded base-64 string is different as the string ended in a valid state - it would be invalid if more bytes followed, but failing because there *might* be more isn't good.


Would be nice if ext/mbstring provided a true stateful validator. Like a class that you feed stream input as it arrives and can query for the current validation state, or a procedural function that returns some state identifier that you pass back to it with subsequent calls.

> However, I don't think that makes this behavior correct. If it were reading a UTF-8 stream and the string ended in the middle of a multibyte sequence then it should report as valid.

No, this is not how mb_check_encoding() is supposed to work. The string as a whole must be valid, not just a valid prefix. (Unfortunately many flush implementations in mbfl are somewhat broken, so theory and practice may not align very well.)

Ah shoot, you're right. It's mb_detect_encoding that works on partial strings, not mb_check_encoding.

I'd like to mention that For word "δοκιμή" the encoded string is zrTOv866zrnOvM6u and the result is false as well. It's not the padded = signs only.

@info: Yes, I was wrong about the nature of the bug, and the code I checked my response with also happened to be wrong (and in a way that made it look like mb_check_encoding was working). You can disregard most of what I said :(

No problem. I just saw that the title of the bug changed and it's about padded strings while it's failing for all base64 strings. A bit misleading

I suppose it could use another small update.