I use Keepass, KeepassX, KeepassDroid and Dropbox to sync them all together. I usually generate my passwords with `pwgen` rather than the purely-random strings that come out of Keepass, as sometimes I'll want to type the password by hand -- pwgen`s passwords are much easier to copy as they avoid ambiguous characters.
I recently went looking for a passphrase generator that supports the http://xkcd.com/936/ scheme, and failed to find one I thought I could trust. So I created [1], which is about as simple as it can get, is loaded (from my own server) over SSL and doesn't rely on anything server-side or use any frameworks, so you can easily verify it by hand. It does rely on Firefox at the moment as Chrome doesn't have the crypto functions it needs yet. I intend to clean it up and publish it properly, but I'm all out of tuits, let alone round ones.
I don't see any valid issues raised in that discussion beyond the question if 44ish bits of entropy is enough. And it is quite easy to bump the complexity if necessary. Wordlist of 10000 words (instead of 2000) and 5 (instead of 4) words and you get 66 bits of entropy (log2(10000^5)), which should be enough for most uses.
edit: some numbers for scale: oclHashCat claims 2500M tries per second for plain SHA1 with single HD7970 GPU. That would mean that 4 of 2000 word passphrase would be cracked in 1.5 hours and 5 of 10000 word passphrase would take 1268 years.
It's normally not too bad -- KeePassDroid creates notifications that copy the username or password when selected, so you find the entry then switch applications to where you want to use it and use those notifications.
More often, though, I'll be looking up credentials on my phone to enter into a different system. That's where having not-completely-random passwords becomes useful, as the output of `pwgen` is much easier to type.
I recently went looking for a passphrase generator that supports the http://xkcd.com/936/ scheme, and failed to find one I thought I could trust. So I created [1], which is about as simple as it can get, is loaded (from my own server) over SSL and doesn't rely on anything server-side or use any frameworks, so you can easily verify it by hand. It does rely on Firefox at the moment as Chrome doesn't have the crypto functions it needs yet. I intend to clean it up and publish it properly, but I'm all out of tuits, let alone round ones.
[1]: http://ares.aylett.co.uk/pw/