Inline responses

On Fri, Apr 23, 2021, 11:18 AM David A. Harding <dave@dtrt.org> wrote:
On Tue, Apr 20, 2021 at 08:46:07AM -0700, Jeremy via bitcoin-dev wrote:
> Script is technically "too wide" a type as what I really want is to
> only return coins with known output types.

I don't understand this concern.  If script is too wide a type, then
OP_RETURN being a scriptPubKey of arbitrary length up to almost a
million bytes is also going to be too wide, right?


I meant the type itself is too wide, not the length of the value. As in Script can represent things we know nothing about. There's a bit of leaky abstraction since the values self describe the type they are. For addresses it's just representations IMO for the standard output types one might expect from standard software. 

Btw: According to... Oh wait... You?
https://bitcoin.stackexchange.com/questions/35878/is-there-a-maximum-size-of-a-scriptsig-scriptpubkey the max size is 10k bytes. Still probably too big for an address, but I'd be ok with making op_return addresses only defined for a small size (e.g. 128 bytes?)



> 1) Should it be human readable & checksummed or encoded?

It should absolutely not be human readable in the sense of being
meaningful to humans.  We've seen in the past that tools and sites that
display OP_RETURN data as ASCII encourage people to put text in the
block chain that is offensive and illegal.  This puts people running
nodes at risk of social and legal intervention.  Bitcoin's
premissionless nature means we can't stop people from creating such
problems, but we can lower the risk by having our tools default to
meaningless representations of OP_RETURN data.

The best advice I've seen is to display OP_RETURN data in hex.  It's
still possible to say things like "dead beef" with that, but significant
abuse is hard.  This will, of course, make even 80 byte OP_RETURN
"addresses" very long.

Is it possible/easy to, say, using bech32m make an inappropriate message in the address? You'd have to write the message, then see what it decodes to without checking, and then re encode? I guess this is worse than hex?

But it seems this is a general thing... If you wanted an inappropriate message you could therefore just use bech32m addressed outputs.


> 2) Should it have a fixed length of max 40-80 bytes or should we support
> arbitrary length strings?

If it doesn't support the fell range, somebody's just going to complain
later and there will have to be a v2 address.

So 10,000 bytes? Or do we care to represent outputs that would be consensus invalid?



> 3) Should it be possible (i.e., from core) to pay into such an OP_RETURN or
> should we categorize OP_RETURNS as a non-payable address type (and just use
> it for parsing blockdata)

I don't think including arbitrary data in the block chain is something
that's currently useful for typical end users, and applications that
want to use OP_RETURN with Bitcoin Core can already call
create(psbt|rawtransaction) with the `data` field, so I'd be mildly
opposed in including such a feature in Bitcoin Core's wallet.  If at
least a few other wallets add the feature to pay OP_RETURN "addresses"
and it seems popular, then I'm wrong and so I would probably then change
my position.

One of the nice things is that the current psbt interface uses a blind union type whereby the entires in an array are either [address, amount] or ["data", hex]. Having an address type would allow more uniform handling, which is convenient for strongly typed RPC bindings (e.g. rust bitcoin uses a hashmap of address to amount so without a patch you can't create op returns).


Regarding "parsing block data", I don't think there's any need to change
Bitcoin Core's current representation of OP_RETURN outputs (which is
just showing the hex-encoded script in RPC output).  For any program
needing OP_RETURN output, hex format is going to be a the next best
thing to getting it in raw binary.  Any other address format is going to
be equal or more work.

Thats a fair point. I'm mostly thinking about this in the context of strongly typed languages/frameworks where you'll get an address object or enum out, rather than something *stringly* typed. But yes in terms of stringy languages I don't think any changes are needed. 

Additionally, as mentioned in the other thread about OP_RETURN this
week, increasing transaction fees should increasingly push uses of
OP_RETURN off the network or into more efficient constructions, so it
doesn't seem warranted to me to spend a lot of time trying to optimize
how we use it when we'll be using it less and less over time.


Hmm. I agree it should get priced out over time. However there are some uses for this kind of stuff. E.g. stealth addresses, or a single instance of open time stamps. 

The main reason I think they merit some sort of std address type is that I'm writing software that can handle things that we might reasonably see on the network. And it's relatively annoying (without a custom type) to represent OP_RETURN as a not-exceptional type of thing.

In my code what I have done is added the following type:

pub enum ExtendedAddress {
/// A regular standard address type
Address(bitcoin::Address),
/// An OP_RETURN
OpReturn(OpReturn),
/// Unknown
Unknown(bitcoin::Script),
}
Which works more or less fine, but I would much prefer to not have to do this in a custom way, as opposed to a way which is defined in a standard manner across all software (after all, that's the point of standards).

Best,

Jeremy