Re: [Bitcoin-development] Multiwallet support

[Gregory Maxwell <gmaxwell@gmail.com>]

On Fri, Dec 21, 2012 at 3:53 AM, Eric Lombrozo <elombrozo@gmail.com> wrote:
> I started working on a new feature to allow for watch-only addresses in
> wallets. https://github.com/bitcoin/bitcoin/pull/2121
>
> In order to integrate this feature nicely into bitcoin / bitcoin, it will be
> necessary to disable signing and privkey export operations for watch-only
> addresses. Since disabling these things for only some of the keys in a
> wallet but not others is an API nightmare and complicates CreateTransaction
> logic, I propose adding multiple wallet capabilities and specifying upon
> creation whether a wallet is a:
>
> 1)  full signing wallet
> - importaddress is disabled.
>
> 2) watch-only wallet
> - signing and privkey export operations are disabled.
> - importprivkey only saves the associated address but not the private key.
> (behaves like importaddress <address for privkey>)
>
> In order to do the above, it will be necessary to add multiple wallet
> support. Anyhow, that was my initial motivation for multiple wallets - but
> obviously, there are a number of other reasons why people might want
> multiple wallet support.
>
> ----------------------
>
> Adding the ability to specify multiple wallets with associated names and
> passphrases in the config file should be fairly straightforward. However,
> exposing multiple wallets via RPC will be tricky as the existing RPC is not
> designed to support multiple wallets.
>
> As to not break compatibility with the existing RPC calls, we can have a
> main wallet which is always used as the default wallet. If the user wants to
> use a different wallet, the name of the wallet would have to be specified in
> the call. Unfortunately, it doesn't look like we can use many of the
> existing RPC calls (sendfrom, sendmany,sendtoaddress, etc...) since they all
> have optional parameters already and it would be awkward to just tack on the
> wallet name parameter at the end. Also, walletpassphrase is problematic as
> it is not stateless. So it looks like we need a whole separate set of calls
> which require a wallet name and passphrase (if the wallet is encrypted).
>
> For instance,
> walletsendtoaddress <walletname> <passphrase> <bitcoinaddress> <amount>
> [comment] [comment-to]
>
> I welcome any proposals or suggestions as to how this should be done.
>

How about a rpc like "usewallet <foo> <normal RPC>"  that simply
generalizes all the rpcs?

And instead of explicitly deactivating rpcs that don't make sense,
simply have them return an error.  Or, for example,  sendtoaddress on
a watching wallet should actually return an unsigned raw transaction
and a wallet specific message that tells you where to find the private
key.

I think it's desirable to not break compatibility but for this kind of
feature compatibility should not get in the way of doing it right.