Topic: Why Core does not allow to use words to "spawn" your wallet? (Read 332 times)

I was not around when BIP 39 was first specified, so I cannot say why it was not implemented then, but I can speculate based on what I know. However I can say why it is still not implemented today.

Firstly, Bitcoin Core is an open source software where the developers work on things that they care about. There is no roadmap nor is there a project manager saying "go implement that". So if no developer wants to implement a feature, then the feature does not become implemented. So clearly, developers do not want to work on BIP 39. So we need to see why multiple independent developers do not want to implement BIP 39.

When BIP 39 first came out, Bitcoin Core had not yet implemented BIP 32, which is a prerequisite for using BIP 39. BIP 32 was first specified in 2012, but the Bitcoin Core wallet did not receive support for it until 4 years later in 2016. In that time, there had been multiple attempts to implement BIP 32, but those were much more complicated. What ended up being merged was the absolute bare minimum that could be considered BIP 32 support. The wallet was able to generate private keys deterministically, and that was about it. It did not allow users to set the BIP 32 seed, users could not import xprvs or xpubs, and users could not change the derivation path used. It ended up being this bare minimum and taking so long just for the bare minimum simply due to how the wallet is structured. It would take a lot of work to overhaul the wallet structure to actually support BIP 32 properly, so what we got was the absolute bare minimum, and that's what we have even till today, for the most part.

Only after BIP 32 was implemented could we start looking at BIP 39. The technical challenge post BIP 32 is not the big, although BIP 39 would still require significant changes to support. However by this time (in 2016), BIP 39 was already showing its weaknesses. Electrum used BIP 39 for some time, but they moved to their own mnemonic system (which is not really specified). One of the major issues of BIP 39 is that it is touted as a full wallet backup but fails to provide several key pieces of information. BIP 39 fails to provide the derivation paths used, and it fails to provide the type of scripts and addresses to use. The latter point was less evident prior to Segwit, but when Segwit was released in 2016, people began to realize the importance of specifying address types in backups.

People often point out the lack of version numbers in BIP 39, and usually this is referenced as a solution to the aforementioned derivation paths and address types. Personally, I don't think that versioning would help with that, but versioning would certainly help if there were improvements to BIP 39 such as using a different KDF.

Another frequently mentioned issue is the choice of KDF. The authors of BIP 39 chose to use PBKDF2, however this has long been considered to be a fairly weak KDF. I have heard some developers (of other wallets) indicate they don't want to use BIP 39 because of this, but I don't have a strong opinion on that.

There is also the fact that BIP 39 itself says that software should still accept mnemonics with invalid checksums, which is a really weird design decision to make. The checksum could definitely be better and the BIP should not be recommending that invalid checksums be basically ignored.

Lastly, the other main reason that Electrum moved away from BIP 39 was due to the requirement of fixed wordlists. In order for BIP 39 to generate the BIP 32 seed to use, it uses PBKDF2 on the mnemonic itself. This means that the mnemonic isn't really an encoding of the raw entropy used to generate it and that the same entropy used to create mnemonics with different wordlists (e.g. for other languages) would result in different wallets. This significantly reduces the portability and usability of BIP 39.

So with all of these shortcomings in mind, BIP 39 is not really the ideal single string wallet backup solution that people want, and so many developers feel like it is not worth it to invest the time to implement something that has been shown to be insufficient.

Now the question is what will we implement in lieu of BIP 39. This is something that I have been actively working on for the past several years. Pieter Wuille came up with the idea of strings of text called Output Script Descriptors which describe everything needed for a wallet to actually regenerate all of its keys and addresses. Unlike BIP 39, descriptors explicitly specifies the type of addresses (output scripts really, hence the name) to generate. They also explicitly provide the BIP 32 derivation paths. These two facts resolve the two major complains with BIP 39 - the lack of derivation paths , and the lack of address types. Furthermore, descriptors can be extended in the future with new types of descriptors. This can be done easily and was recently done to add taproot support. Bitcoin Core's wallet was completely overhauled so that we could use these descriptors as the basis of the wallet itself.

Currently, descriptors are not like BIP 39 in that they are much longer and aren't a mnemonic. I intend on proposing a generic mnemonic scheme which would allow for descriptors to be represented as a mnemonic, but that's still a ways off. There is still work to be done with descriptors themselves. BIP 39 is also insufficient for a generic mnemonic scheme since the way it is specified is really specific to entropy and generating a random seed for use in something like BIP 32. We would have to make a wholly new mnemonic scheme.

Im still not 100% sure as to why this idea was discarded in Core. I've read some other claims regarding how it was discarded because people may by accident and due lack of information use a non secure seed, for instance, removing words to make it easier to remember, thus compromising the security.

@achow101 posts here. Could you possibly explain this in detail?

Do you have more links with discussions on this? im sure this has been discussed and there is a lot of literature to read as to why this was discarded. I would like to know gmaxell, luke jr, laanjw, peter wiulle and more opinions on this to get a better idea.

No. Electrum generates a master key using a 12 word mnemonic, with the extended keys being in BIP32 format.

By creating a blank wallet and importing a HD seed in the format above, your addresses will be generated using that HD seed and thus you can restore your wallet using that HD seed without having any deviation in addresses. You will only get the same addresses and labels or any data stored locally are only able to be backed up using the wallet file.
The stackexchange link that you've referenced basically encapsulates the reason why Bitcoin Core isn't implementing it. Just because other wallets are able to implement a mnemonic system, it doesn't mean Bitcoin Core has to follow. I believe that it has been brought up several times but it wasn't implemented primarily because of the shortcomings of the mnemonic scheme.

That's more than 12 words. When it comes to 12 words it's pretty doable to memorize without being a genius. Imagine that there's a flood and you lose all of your backups or something. Or once again, you try to cross a border and you don't want to carry anything digital with you encrypted, or governments become increasingly tyrannical and they knock on your door looking for keys and so on. I see many use cases where you wouldn't want to have any encrypted data and being able to memorize 12 words and know your wallet is safe gives you a peace of mind in these extreme situations where carrying any data becomes a liability.


Wait this went over my head. So "cT2ozo5xcj6iCe4dTfywan3qJnib" is the "12 words" that Electrum uses so to speak? With "cT2ozo5xcj6iCe4dTfywan3qJnib" you could spawn it all? Including transaction history, name for each address, etc etc? What does it exactly spawn?

And I still dont understand why Core cannot generated 12 words instead of cT2ozo5xcj6iCe4dTfywan3qJnib assuming Electrum does it safely. At least make it optional.

I did not say that.
Both BIP39 and the algorithm Electrum uses are safe as long as the initial entropy was generated randomly using a strong RNG.

The "remember something" methods are always unsafe because you can't trust your memory after a very long time.
The point of mnemonic algorithms is NOT to let user "remember" the words but to simply write down a human readable form of the random octet string.
In other words it is easier for the user to without mistakes write down (with pen and paper)
than it is to write down

So basically using Electrum as a way to store BTC is not safe? the conclusion is that using any "remember these words and spawn your wallet" methods are not safe and that we must carry the wallet.dat file?

Damn, in this case, how do people deal with this when crossing borders? you never know when you are going to get stopped and forced to decrypt. And they make perpetual copies of your drives anyway and store them.

Actually, although it looks like a WIF, it really is the seed. My post was based on this answer:


The HD seed generates the master private key which cannot be set manually and changing it should generate the subsequent private keys. Because knowledge of what are the "subsequent private keys" is arcane - due to the fact there is one keypool for change and receiving addresses and that they use different paths - it's better to just make a new wallet file from scratch before you run sethdseed to avoid all this complexity.

Oh okay. I stand corrected. I thought both of you were talking about importing using an xpriv key, since the topic is about using BIP39 in Bitcoin Core. Was thinking of using a workaround to use the mnemonic for that.

Yes it is (regtest dump):


You might be talking about the "extended private masterkey"?

Why do you need to rescan after flushing the unused keys?