After receiving a
bug report and doing some research I found a
Debian bug report. (If you read it and notice something, I already talked to people about it.)
Anyway, it seems that calling CryptoPP::AutoSeededRandomPool will cause problems with a drained /dev/urandom (how this can happen is a mistery to me, but apparently it does) and this can be solved by using another random number generator (Debian uses CryptoPP::AutoSeededX917RNG<CryptoPP::DES_EDE3>).
Are there any reasons not to switch to another random number generator? The patch doesn't apply anymore, but we still call it in EncryptedStreamSocket.cpp and abstracted in RandomFunctions.cpp, where it seems to cause the problem reported by Par@noid.
