Le 23/08/2016 à 22:12, Luke Dashjr via bitcoin-dev a écrit :
> BIP 39: Mnemonic code for generating deterministic keys
> - Used by many wallets and hundreds of thousands of users.
> 
> BIP 44: Multi-Account Hierarchy for Deterministic Wallets
> - Appears to be implemented by multiple wallets.
> 

I personally believe that BIP39/BIP44 is a bad design. There is limited
support for these BIPs in Electrum, in order to provide compatibility
with hardware wallets. However, I do not plan to use BIP39/BIP44 for
default Electrum wallets, for the following reasons.

(Note that it does not make sense to consider BIP39 and BIP44
independently. Any wallet that decides to implement one without the
other would be considered as broken.)

Here is why I rejected this design:

1 - BIP44 uses multiple accounts. This means that in order to be
compatible with the standard, a wallet *must* implement multiple
accounts. A wallet that decides to keep things simple and use only one
account, will not allow users to recover all their funds when they
restore from a BIP39 seed, and will be considered as broken.

2 - An appealing feature of deterministic wallets is that you can use
the same instance of your wallet on different devices. Two instances of
your wallet can automatically synchronize their Bitcoin addresses, and
display the same balance. The problem is that hardened derivations break
this property. Indeed, with hardened derivations, software wallets need
to ask the user's password in order to derive new accounts. Therefore,
in order to implement automated detection of newly created accounts, a
BIP44-compatible software wallets would need to ask the user's password
whenever a new account is detected. This means that the wallet would ask
the password without the user initiating any action. This seems to be an
avenue for malware.

Of course, hardware wallets do not have that issue, because they can
derive new accounts without requesting a password from the user. BIP44
is a standard that has been designed for hardware wallets, but that
makes things really difficult for software wallets.

3 - Unneeded complexity. From an end user perspective, the multiple
accounts in BIP44 achieve the same result as using different derivation
passphrases with the same BIP39 seed phrase. The only real difference is
that BIP44 accounts can be enumerated deterministically, while
passphrases in general cannot. However, this property is of limited
interest, because automatic synchronization of multiple accounts cannot
be guaranteed for bip44 software wallets, as explained in 2.

4 - BIP39 is inconsistent. It uses a hash of the utf8 encoded 'seed
phrase' in order to derive the BIP32 seed. This hash-based derivation
was added on my suggestion, in order to make the BIP independent from
the particular wordlist used to generate the seed phrases. However,
BIP39 also requires the implementation of a checksum, in order to verify
that a seed phrase is valid. Suprisingly, the specification of the
checksum involves wordlist indices. This means the checksum (and thus
the BIP) requires a fixed wordlist. This defeats the purpose of using a
hash for the derivation of the seed.

The authors of the BIP should either have used hash functions for both
the seed AND the checksum (that is what Electrum does), or for none of
them (in that case case, you can have a bidirectional function between
seed phrases and entropy, which is nice if you want to perform Shamir
secret sharing of seed phrases, at the expenses of a fixed wordlist). In
its current state, BIP39 takes the worst of both worlds.

5 - The fact that the wordlist must be part of BIP39, and cannot be
changed in the future, seems a terrible idea to me. I believe that a
specification should always try to be minimal. In that case, the
specification includes a 2000+ words dictionary, when it could have
avoided that.

Even if you decide that BIP39 is final, there will always be users
requiring the addition of wordlists for new languages. So, in practice,
this BIP will never be final.

6 - Finally, and most importantly, BIP39 seed phrases do not have a
version number. Without a version number, how are you going to derive
addresses from a BIP39 seed phrase, when wallets start to use to new
derivation methods (such as SegWit, or Schnorr signatures)? Does it mean
that a BIP39 compatible wallet will have to check addresses from all the
derivation methods that ever existed in the past, in order to ensure
that all coins are correctly retrieved? Or will there be users that
cannot access their coins because their BIP39 seed phrase is too old for
newer software?