Message-ID: <550057FD.6030402@electrum.org>
Date: Wed, 11 Mar 2015 15:58:05 +0100
From: Thomas Voegtlin <thomasv@electrum.org>
User-Agent: Mozilla/5.0 (X11; Linux x86_64;
	rv:31.0) Gecko/20100101 Thunderbird/31.4.0
MIME-Version: 1.0
To: Mike Hearn <mike@plan99.net>
References: <54F32EED.6040103@electrum.org>
	<CANEZrP23buJF0ENfrKGRuzpQ3Uod09s-kRcb3CBw1-OmUxEyZg@mail.gmail.com>
In-Reply-To: <CANEZrP23buJF0ENfrKGRuzpQ3Uod09s-kRcb3CBw1-OmUxEyZg@mail.gmail.com>
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: quoted-printable
Cc: Bitcoin Development <bitcoin-development@lists.sourceforge.net>
Subject: Re: [Bitcoin-development] Electrum 2.0 has been tagged
Precedence: list

Thanks Mike, and sorry to answer a bit late; it has been a busy couple
of weeks.

You are correct, a BIP39 seed phrase will not work in Electrum, and vice
versa. It is indeed unfortunate. However, I believe BIP39 should not be
followed, because it reproduces two mistakes I did when I designed the
older Electrum seed system. Let me explain.

The first problem I have with BIP39 is that the seed phrase does not
include a version number.

Wallet development is still in an exploratory phase, and we should
expect even more innovation in this domain. In this context, it is
unwise to make decisions that prevent future innovation.

However, when we give a seed phrase to users, we have a moral obligation
to keep supporting this seed phrase in future versions. We cannot simply
announce to Electrum users that their old seed phrase is not supported
anymore, because we created a new version of the software that uses a
different derivation. This could lead to financial losses for users who
are unaware of these technicalities. Well, at least, that is how I feel
about it.

BIP39 and Electrum v2 have a very different ways of handling future
innovation. Electrum v2 seed phrases include an explicit version number,
that indicates how the wallet addresses should be derived. In contrast,
BIP39 seed phrases do not include a version number at all. BIP39 is
meant to be combined with BIP43, which stipulates that the wallet
structure should depend on the BIP32 derivation path used for the wallet
(although BIP43 is not followed by all BIP39 compatible wallets). Thus,
innovation in BIP43 is allowed only within the framework of BIP32. In
addition, having to explore the branches of the BIP32 tree in order to
determine the type of wallet attached to a seed might be somewhat
inefficient.

The second problem I see with BIP39 is that it requires a fixed
wordlist. Of course, this forbids innovation in the wordlist itself, but
that's not the main problem. When you write a new standard, it is
important to keep this standard minimal, given the goal you want to
achieve. I believe BIP39 could (and should) have been written without
including the wordlist in the standard.

There are two ways to derive a master key from a mnemonic phrase:
 1. A bidirectional mapping between words and numbers, as in old
Electrum versions. Pros: bidirectional means that you can do Shamir
secret sharing of your seed. Cons: It requires a fixed wordlist.
 2. Use a hash of the seed phrase (pbkdf). Pros: a fixed wordlist is not
required. Cons: the mapping isn't bidirectional.

Electrum v1 uses (1). Electrum v2 uses (2).

Early versions of BIP39 used (1), and later they switched to (2).
However, BIP39 uses (2) only in order to derive the wallet keys, not for
its checksum. The BIP39 checksum uses (1), and it does requires a fixed
wordlist. This is just plainly inconsistent. As a result, you have
neither wordlist flexibility, nor Shamir secret sharing.

Having a fixed wordlist is very unfortunate. First, it means that BIP39
will probably never leave the 'draft' stage, until all languages of the
world have been added. Second, once you add a wordlist for a new
language, you cannot change it anymore, because it will break existing
seed phrases; therefore you have to be extremely careful in the way you
design these wordlists. Third, languages often have words in common.
When you add a new language to the list, you should not use words
already used by existing wordlists, in order to ensure that the language
can be detected. It leads to a first come first served situation, that
might not be sustainable in the future.

In order to support the old Electrum v1 seeds, all future versions of
Electrum will have to include the old wordlist. In addition, when
generating new seed phrases, Electrum now has to avoid collisions with
old seed phrases, because the old ones did not have a version number.
This is painful enough, I will not repeat the same errors twice.

Electrum v2 derives both its private keys and its checksum/version
number using a hash of the seed phrase. This means that wordlists can be
added and modified in the future, without breaking existing seed
phrases. It also means that it will be very easy for other wallets to
support Electrum seedphrases: it requires about 20 lines of code, and no
wordlist is required.


Thomas


Le 02/03/2015 16:37, Mike Hearn a =C3=A9crit :
> Congrats Thomas! Glad to see Electrum 2 finally launch.
>=20
>=20
>> * New seed derivation method (not compatible with BIP39).
>=20
>=20
> Does this mean a "12 words" wallet created by Electrum won't work if
> imported into some other wallet that supports BIP39? Vice versa? This s=
eems
> unfortunate. I guess if seeds are being represented with 12 words
> consistently, people will expect them to work everywhere.
>=20