Re: [Bitcoin-development] BIP70: why Google Protocol Buffers for encoding?

[Nicolas DORIER <nicolas.dorier@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 6684 bytes --]

Sure,

But the mobile targets, it is still easier to use Json + HTTPS, especially
when you want one code base for everything.
And as you said, developers need to think about fetching mozilla store time
to time, and check revocations themselves. This is not obvious thing to do,
and hard to test correctly.

If your use case was the primary utility of BIP70, then I'd say it fit the
bill. But for cross plateform client development an atlernative would be
easier.

> why not allow both serializations and keep serialization format a
parameter, keep everyone happy.

It would be another BIP, because if we use JSON with HTTPS, the difference
is also in the semantic (no embedded certificates)

I will likely provide this option for a product I am developing. I will
only use another Content Type. We'll see then how it goes.

2015-01-28 18:14 GMT+01:00 Mike Hearn <mike@plan99.net>:

> I think we'll just have to agree to disagree on this one. I've implemented
> BIP70 a couple of times now and didn't find it to be difficult. I know you
> had odd problems with the C# protobuf implementation you were using but
> library bugs can happen for any kind of programming.
>
> I forgot to mention the other reason it's done this way. One of the
> driving goals of BIP70 was to support the TREZOR and similar devices. For
> hardware wallets, it's critical to keep the amount of code they need to run
> as small as possible. Any bugs in the code there can cause security holes
> and lead to the device being hacked.
>
> Doing it the way you suggest would mean the secure code would have to
> contain complex and bug-prone text parsing logic as well as a full blown
> HTTP and SSL stack, that requires not only X.509 handling but also lots of
> other stuff on top. It'd increase cost, complexity and decrease security
> quite a bit.
>
> Whilst I appreciate if your platform provides a scripting-like API and
> nothing low level it might seem easier to use JSON+HTTPS, that isn't the
> case for one of the primary design targets.
>
>
>
> On Wed, Jan 28, 2015 at 6:04 PM, Nicolas Dorier <nicolas.dorier@gmail.com>
> wrote:
>
>> Mike, I am not denying it is impossible to do all of that.
>> Just that it is not a trivial stuff to do to make it works everywhere,
>> and I think that it is not a good thing for a client side technology.
>> BIP70 has its use, and I understand why there is case where it is good to
>> ship the certs in the message and not depends on the transport.
>>
>> But a standard that just use JSON and HTTPS, even if less flexible that
>> BIP70, would make it easier and sufficient for today's use case.
>>
>> On Wed, Jan 28, 2015 at 5:55 PM, Mike Hearn <mike@plan99.net> wrote:
>>
>>> My point is not that there is a limitation in BIP70. My point is that
>>>> you put the burden of certificate verification on developer's shoulder when
>>>> we can just leverage built in HTTPS support of the platform.
>>>>
>>>
>>> Platforms that support HTTPS but not certificate handling are rare - I
>>> know HTML5 is such a platform but such apps are inherently dependent on the
>>> server anyway and the server can just do the parsing and validation work
>>> itself. If WinRT is such a platform, OK, too bad.
>>>
>>> The embedding of the certificates is not arbitrary or pointless, by the
>>> way. It's there for a very good reason - it makes the signed payment
>>> request verifiable by third parties. Effectively you can store the signed
>>> message and present it later to someone else, it's undeniable. Combined
>>> with the transactions and merkle branches linking them to the block chain,
>>> what you have is a form of digital receipt ... a proof of purchase that can
>>> be automatically verified as legitimate. This has all kinds of use cases.
>>>
>>> Because of how HTTPS works, you can't easily prove to a third party that
>>> a server gave you a piece of data. Doing so requires staggeringly complex
>>> hacks (see tls notary) and when we designed BIP70, those hacks didn't even
>>> exist. So we'd lose the benefit of having a digitally signed request.
>>>
>>> Additionally, doing things this way means BIP70 requests can be signed
>>> by things which are not HTTPS servers. For example you can sign with an
>>> email address cert, an EV certificate i.e. a company, a certificate issued
>>> by some user forum, whatever else we end up wanting. Not every payment
>>> recipient can be identified by a domain name + dynamic session.
>>>
>>>
>>>> However, if you want to use your plateform's store, then you are toasted
>>>>
>>>
>>> That's a bit melodramatic. BitcoinJ is able to use the Android, JRE,
>>> Windows and Mac certificate stores all using the same code or very minor
>>> variants on it (e.g. on Mac you have to specify you want the system store
>>> but it's a one-liner).
>>>
>>> Yes, that's not *every* platform. Some will require custom binding glue
>>> and it depends what abstractions and languages you are using.
>>>
>>>
>>>> Have you tried to do that on windows RT and IOS ? I tried, and I
>>>> quickly stopped doing that since it is not worth the effort. (Frankly I am
>>>> not even sure you can on win rt, since the API is a stripped down version
>>>> of windows)
>>>>
>>>
>>> There is code to do iOS using the Apple APIs here:
>>>
>>>
>>> https://github.com/voisine/breadwallet/blob/master/BreadWallet/BRPaymentProtocol.m#L391
>>>
>>>
>>>> Why have you not heard about the problem ? (until now, because I have
>>>> this problem because I need to have the same codebase on
>>>> winrt/win/android/ios/tablets)
>>>>
>>>
>>> WinRT is a minority platform in the extreme, and all the other platforms
>>> you mentioned have the necessary APIs. Java abstracts you from them. So I
>>> think you are encountering this problem because you desire to target WinRT
>>> and other platforms with a single codebase. That's an unusual constraint.
>>>
>>> AFAIK the only other people who encountered this are BitPay, because
>>> they want to do everything in Javascript which doesn't really provide any
>>> major APIs.
>>>
>>>
>>>> Also, you bundle mozilla's store in bitcoinj, what happen when the
>>>> store change and your customer have not intent to use bitcoinj new version
>>>> ? by leveraging the plateform you benefit from automatic updates.
>>>>
>>>
>>> Yes, there are pros and cons to bundling a custom root store.
>>>
>>>
>>>> Also, does java stores deals with certificate revocations ? sure you
>>>> can theorically code that too... or just let the plateform deals with it.
>>>>
>>>
>>> It can do OCSP checks, yes, although I believe no wallets currently do
>>> so. A better solution would be to implement an OCSP stapling extension to
>>> BIP70 though.
>>>
>>
>>
>

[-- Attachment #2: Type: text/html, Size: 9596 bytes --]