curiosity81, it seems (at a glance) that you are deploying a word generator generated from a decent (i.e. non-human) source of randomness. Not a syntactically valid phrase, not something the user comes up with, not something from a book you cross your fingers and hope to be really obscure. Not what most people call a “brainwallet”.
True. That's why I mentioned above that "brainwallet" might be the wrong term.
I have to stress again, that I wanted simple code for generating key pairs from strings, either from a concatenated complex list of words (kind of brainwallet) or a complex set of characters (kind of passphrase). And I want to understand the code. A kind of finger exercise. And maybe some of the code is simple enough to be useful for others.
For instance, to derive the public key of a private key is the only piece of code, which is not a bash script. I found this code somewhere at stackoverflow (the link is given in the source). It is c, but relatively easy to compile. If I remember correctly only libssl-dev is necessary for compilation. This is the only simple (!) piece of code I found! Another example is the bash script for computing base58. The code you can find online uses a lot of bash magic, so that it is virtually impossible for a normal guy to understand. At least I learned what bc is. So it was easier to write the code by myself with the benefit, that I now know how it works.
That raises an obvious question, which I must ask out of—curiosity: Why don’t you simply use BIP 39? It was developed by the same experts whose security acumen you trust when you use Bitcoin anyway. Its wordlists were developed with human use in mind, e.g., all words on the English wordlist are unique within the first four characters. And it will perfectly encode 128–256 bits of randomness in 12–24 words, without any of the pitfalls of trying to develop your own word randomization scheme.
With your wordlist, I presume not tuned to a power of 2, did you avoid the common mistake of introducing modulo bias? (I did not review your code.) Does your wordlist exclude similar, confusable words? (I am guessing not.) Etc.
Honestly, I did not take care of the random generator's entropy yet. But I can assure you, that I am paranoid enough to invest (and already invested) a lot of energy to get the maximum entropy into my systems. In the current version, one bash script simply shuffles (shuf) a complete aspell dictionary and takes the head (the n first words) of the output. That's it. Clearly I have to figure out in detail how shuf works and where it gets its entropy. Entropy pool could be filled with additional entropy via randomsound or similar. But how to connect shuf to /dev/random or /dev/urandom is a question for the future. And should be as simple as possible.
I did not wrote my own BIP39 generator, even though I am aware of the proposal. Maybe I even stole some ideas (checksum). Maybe I used them wrongly.
Thanks for the suggestion. I will take a look.
