I believed that it's generally expected that a password generator would produce passwords of a certain minimal length. At least I considered that nobody would write a generator (intended for a real-world usage) that'd produce, say, 3-character password for some edge case.
However, you sound reasonable. This leads us right to the extreme case - should empty passwords be allowed? (Considering that the user will be bugged like hell before letting him to do so.)
I should think more about this.
sigh
:|
The interesting thing to note is that if they had any significant problem with this scheme, they would have changed it. Maybe we worry too much about the strength of passwords. The password verification process may be hardened enough, even for the needs of a bank.
Awesome. Not like I use Hotmail, but... So now if someone's password generator just happen to generate "weak" password not containing, for example, a digit (uh, even `openssl rand -base64 12` provides such outputs from time to time) user'll have to step away from usual password generation scheme and create special password just for hotmail.com.
Please, for the love of sanity, never ever forbid any passwords (except for too short ones, with a reasonable minimal length). Just freak user out so he'll think twice before using possibly weak password. You'll educate users this way instead of frustrating them.
(And never limit maximum length or set of possible characters, except for rare cases where there are technical obstacles requiring to do so - like non-8-bit-safe protocols. If user wants to authenticate with a passpoem, written in runic alphabet — let him have it.)