Length for AES256 symmetric encryption passphrase?

[Robert J. Hansen]

> Interesting math.  However, I believe the OP mentioned they're generating
> the password and storing so human readable, i.e. English, isn't an issue.
> What would be the recommended length for completely random characters
> generated, for example, by a password manager such as keepassx?

Your questions are not clear enough to be answered.

"What would the recommended length for completely random characters  
generated, for example, by a password manager such as keepassx?  If  
one were using the password as the symmetric key in libgcrypt?  Or  
perhaps even just using openssl tools?"

1.  Well, which password managers?  Just because a character is  
completely random tells me nothing about how much entropy is contained  
in each symbol.  "TTHTHHTTH" is a completely random sequence  
(generated it just now by flipping a fair coin), but it only has one  
bit of entropy per symbol.  "fBTvC" is a completely non-random  
sequence, but it has a lot more entropy per symbol.  Without knowing  
how a random password is generated I can't answer this.

2.  Recommended for what purpose?  256 bits of entropy is wild  
overkill for almost all purposes.  128 bits of entropy is generally  
speaking plenty.

3.  Which toolkit?  libgcrypt and openssl are two completely different  
toolkits that work in completely different ways, and an answer  
appropriate for one might not be appropriate for the other.

4.  What is it you really want to know?  You already know: AES depends  
on having a 32-bit key which can support up to 256 bits of entropy.   
You've been told two good metrics for estimating entropy in a  
passphrase: 1.5 bits per glyph of English text, 5 bits per glyph of  
base-64ed random data.