Quote from: azeteki on December 01, 2014, 01:04:06 AM

There are a few different approaches to this whole business that I can think of.

The first is to rely on bitcoind to perform the wallet functionality. Tell bitcoind to send some coins, and bitcoind consults its' wallet.dat and so forth.
Least likely to go wrong, but limited in scope.

The second and more ambitious is to effectively produce a standalone wallet that uses bitcoind to connect to the p2p network only. In this case bitcoind-ncurses would have its' own 'wallet.dat' format, produce its own keys, do its' own coin control, and so on and so forth. Much more 'fun' for me, and also achieving the secondary aim of producing a viable split between the 'node' that is bitcoind and the 'wallet' that is bitcoind, but requiring that I really know what I'm doing.

There would be a middle ground here: using the new watch-only addresses of bitcoind, you could delegate all the tracking to bitcoind. Private keys and coin control could then be handled by bitcoind-ncurses. Would be a nice compromise as it would allow you to get all the potential while getting around having to implement your own rescanning, indexing, and tracking code.

Quote

What I would need to do to feel comfortable with either solution is to spend a lot of time studying how bitcoind achieves its' wallet functionality and also ensure that Python is even capable of producing a reasonably secure wallet (which would entail a fair amount of crypto-study).

To elaborate on the latter point - take a basic function like 'walletpassphrase'. This needs to be handled very carefully. You can end up in a situation where the passphrase sits around in RAM for the rest of all eternity after being used. Or perhaps the size of the program's RAM usage increases by some predefined way based on passphrase content. Maybe the query takes a different amount of time depending on the length of the password, etcetera. Some of these probably matter as side channel attacks and others may not. If I am to produce a wallet tool for general consumption then I have a moral obligation to do everything within my power to ensure that it is safe because people can and will use it to store lots of money.

I think bitpop's suggestion of taking a look at armory is a good one. It's also written in python and solved most of these problems already. An ncurses interface to some of armory's functions (while relying only on bitcoind) would be an awesome combination!

Feature wise, what I would love is to see a ncurses interface to coincontrol: constructing transactions by selecting UTXOs from the list of available ones and computing fees following different schemes (free, pre 0.8.2, current, manual).

Quote

It makes me very happy that people have found my work useful so far. Finding some solution to fund the roof over my head whilst also pursuing this sort of project remains the challenge we all struggle with.

Indeed.

I can only say: keep up the good work! Hope you'll find some income in the space that allows you to continue this work. If this ever happens to myself, I would happily donate as well...