>> > And, btw, why not allow raw bytes here as well? is there some problem?
>>
>> Because of ambiguity, both these return the same value:
>>
>> (read (concat "?" (string 128)))     => 128
>> (read (concat "?" (string 4194176))) => 128
>
> And why is that a problem?

I don't know, Andreas remarked that it creates ambiguous output,
and I fixed the reported problem.

> Alternatively, we could print raw bytes in some special way.  But not
> treating them as characters sounds some subtlety that will be hard to
> explain.

The existing 'prin1-char' used as a reference implementation
doesn't print integers like 4194176 as characters, so the patch
does the same.

Anyway, here is a complete patch with tests and documentation: