Re: [Bitcoin-development] Missing fRelayTxes in version

Re: [Bitcoin-development] Missing fRelayTxes in version

[Tamas Blummer]

[-- Attachment #1: Type: text/plain, Size: 388 bytes --]

Hi Mike,

The issue with the current parser is that those fields are conditionally optional on that there will be no subsequent fields added.
If there will be further fields they will become manadory. 
 
Why not bump the version and parse the fields as mandatory from then on? This would be backward compatible and cleaner
going forward.

Tamas Blummer
http://bitsofproof.com


[-- Attachment #2: Type: text/html, Size: 5308 bytes --]

Re: [Bitcoin-development] Missing fRelayTxes in version

[Mike Hearn]

[-- Attachment #1: Type: text/plain, Size: 1679 bytes --]

Sure but why not do that when there's an actual new field to add? Does
anyone have a proposal for a feature that needs a new version field at the
moment? There's no point changing the protocol now unless there's actually
a new field to add.

Anyway I still don't see why anyone cares about this issue. The Bitcoin
protocol does not and never has required that all messages have a fixed
number of fields per version. Any parser written on the assumption it did
was just buggy. Look at how tx messages are relayed for the most obvious
example of that pattern in action - it's actually the raw byte stream
that's stored and relayed to ensure that fields added in new versions
aren't dropped during round-tripping. Old versions are supposed to preserve
fields from the future.



On Thu, Jun 20, 2013 at 9:30 AM, Tamas Blummer <tamas@bitsofproof.com>wrote:

> Hi Mike,
>
> The issue with the current parser is that those fields are conditionally
> optional on that there will be no subsequent fields added.
> If there will be further fields they will become manadory.
>
> Why not bump the version and parse the fields as mandatory from then on?
> This would be backward compatible and cleaner
> going forward.
>
> Tamas Blummer
> http://bitsofproof.com
>  <http://bitsofproof.com/>
>
>
>
> ------------------------------------------------------------------------------
> This SF.net email is sponsored by Windows:
>
> Build for Windows Store.
>
> http://p.sf.net/sfu/windows-dev2dev
> _______________________________________________
> Bitcoin-development mailing list
> Bitcoin-development@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/bitcoin-development
>
>

[-- Attachment #2: Type: text/html, Size: 5506 bytes --]

Re: [Bitcoin-development] Missing fRelayTxes in version

[Tamas Blummer]

[-- Attachment #1: Type: text/plain, Size: 2169 bytes --]

I agree that this can be deferred until there is an actual new field without any harm. But then remember to update the BIP37 too saying that it is optional only if flag added in BIPXX is not present.

Your argument is that this complexity is already there so why not preserve it. I think eliminating complexity (that has no benefit) strengthens the system.

Tamás Blummer
http://bitsofproof.com

On 20.06.2013, at 09:36, Mike Hearn <mike@plan99.net> wrote:

> Sure but why not do that when there's an actual new field to add? Does anyone have a proposal for a feature that needs a new version field at the moment? There's no point changing the protocol now unless there's actually a new field to add.
> 
> Anyway I still don't see why anyone cares about this issue. The Bitcoin protocol does not and never has required that all messages have a fixed number of fields per version. Any parser written on the assumption it did was just buggy. Look at how tx messages are relayed for the most obvious example of that pattern in action - it's actually the raw byte stream that's stored and relayed to ensure that fields added in new versions aren't dropped during round-tripping. Old versions are supposed to preserve fields from the future.
> 
> 
> 
> On Thu, Jun 20, 2013 at 9:30 AM, Tamas Blummer <tamas@bitsofproof.com> wrote:
> Hi Mike,
> 
> The issue with the current parser is that those fields are conditionally optional on that there will be no subsequent fields added.
> If there will be further fields they will become manadory. 
>  
> Why not bump the version and parse the fields as mandatory from then on? This would be backward compatible and cleaner
> going forward.
> 
> Tamas Blummer
> http://bitsofproof.com
> 
> 
> ------------------------------------------------------------------------------
> This SF.net email is sponsored by Windows:
> 
> Build for Windows Store.
> 
> http://p.sf.net/sfu/windows-dev2dev
> _______________________________________________
> Bitcoin-development mailing list
> Bitcoin-development@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/bitcoin-development
> 
> 


[-- Attachment #2: Type: text/html, Size: 9540 bytes --]

Re: [Bitcoin-development] Missing fRelayTxes in version

[Mike Hearn]

[-- Attachment #1: Type: text/plain, Size: 3043 bytes --]

You can't eliminate the complexity (yet), otherwise you wouldn't be able to
talk to old nodes. You'll have to wait until versions prior to a particular
version are hard-forked off and can be safely dropped at connect time.

That said the reason I'm being so grumpy about this is that compared to the
complexity in the rest of the system, this is such a trivial and minor
detail. It's hardly even worth thinking about. I mean, we have a scripting
language full of opcodes nobody ever figured out how to use and the
protocol uses a mixture of byte orders, so an optional field in the version
message is really not such a big deal :)


On Thu, Jun 20, 2013 at 10:17 AM, Tamas Blummer <tamas@bitsofproof.com>wrote:

> I agree that this can be deferred until there is an actual new field
> without any harm. But then remember to update the BIP37 too saying that it
> is optional only if flag added in BIPXX is not present.
>
> Your argument is that this complexity is already there so why not preserve
> it. I think eliminating complexity (that has no benefit) strengthens the
> system.
>
> *Tamás Blummer*
> http://bitsofproof.com
>  <http://bitsofproof.com/>
>
> On 20.06.2013, at 09:36, Mike Hearn <mike@plan99.net> wrote:
>
> Sure but why not do that when there's an actual new field to add? Does
> anyone have a proposal for a feature that needs a new version field at the
> moment? There's no point changing the protocol now unless there's actually
> a new field to add.
>
> Anyway I still don't see why anyone cares about this issue. The Bitcoin
> protocol does not and never has required that all messages have a fixed
> number of fields per version. Any parser written on the assumption it did
> was just buggy. Look at how tx messages are relayed for the most obvious
> example of that pattern in action - it's actually the raw byte stream
> that's stored and relayed to ensure that fields added in new versions
> aren't dropped during round-tripping. Old versions are supposed to preserve
> fields from the future.
>
>
>
> On Thu, Jun 20, 2013 at 9:30 AM, Tamas Blummer <tamas@bitsofproof.com>wrote:
>
>>  Hi Mike,
>>
>> The issue with the current parser is that those fields are conditionally
>> optional on that there will be no subsequent fields added.
>> If there will be further fields they will become manadory.
>>
>> Why not bump the version and parse the fields as mandatory from then on?
>> This would be backward compatible and cleaner
>> going forward.
>>
>> Tamas Blummer
>> http://bitsofproof.com
>>  <http://bitsofproof.com/>
>>
>>
>>
>> ------------------------------------------------------------------------------
>> This SF.net email is sponsored by Windows:
>>
>> Build for Windows Store.
>>
>> http://p.sf.net/sfu/windows-dev2dev
>> _______________________________________________
>> Bitcoin-development mailing list
>> Bitcoin-development@lists.sourceforge.net
>> https://lists.sourceforge.net/lists/listinfo/bitcoin-development
>>
>>
>
>

[-- Attachment #2: Type: text/html, Size: 9432 bytes --]

Re: [Bitcoin-development] Missing fRelayTxes in version

[Tamas Blummer]

[-- Attachment #1: Type: text/plain, Size: 3370 bytes --]

Yes it is trivial. I do not think greater complexity in the system should keep us from addressing low complexity issues.
You can't blame me or others not trying to simplify scripts, if there is such a headwind simplifying a version message.
You are right there is too much fuss about this.

Tamás Blummer
Founder, CEO
http://bitsofproof.com

On 20.06.2013, at 10:31, Mike Hearn <mike@plan99.net> wrote:

> You can't eliminate the complexity (yet), otherwise you wouldn't be able to talk to old nodes. You'll have to wait until versions prior to a particular version are hard-forked off and can be safely dropped at connect time.
> 
> That said the reason I'm being so grumpy about this is that compared to the complexity in the rest of the system, this is such a trivial and minor detail. It's hardly even worth thinking about. I mean, we have a scripting language full of opcodes nobody ever figured out how to use and the protocol uses a mixture of byte orders, so an optional field in the version message is really not such a big deal :)
> 
> 
> On Thu, Jun 20, 2013 at 10:17 AM, Tamas Blummer <tamas@bitsofproof.com> wrote:
> I agree that this can be deferred until there is an actual new field without any harm. But then remember to update the BIP37 too saying that it is optional only if flag added in BIPXX is not present.
> 
> Your argument is that this complexity is already there so why not preserve it. I think eliminating complexity (that has no benefit) strengthens the system.
> 
> Tamás Blummer
> http://bitsofproof.com
> 
> On 20.06.2013, at 09:36, Mike Hearn <mike@plan99.net> wrote:
> 
>> Sure but why not do that when there's an actual new field to add? Does anyone have a proposal for a feature that needs a new version field at the moment? There's no point changing the protocol now unless there's actually a new field to add.
>> 
>> Anyway I still don't see why anyone cares about this issue. The Bitcoin protocol does not and never has required that all messages have a fixed number of fields per version. Any parser written on the assumption it did was just buggy. Look at how tx messages are relayed for the most obvious example of that pattern in action - it's actually the raw byte stream that's stored and relayed to ensure that fields added in new versions aren't dropped during round-tripping. Old versions are supposed to preserve fields from the future.
>> 
>> 
>> 
>> On Thu, Jun 20, 2013 at 9:30 AM, Tamas Blummer <tamas@bitsofproof.com> wrote:
>> Hi Mike,
>> 
>> The issue with the current parser is that those fields are conditionally optional on that there will be no subsequent fields added.
>> If there will be further fields they will become manadory. 
>>  
>> Why not bump the version and parse the fields as mandatory from then on? This would be backward compatible and cleaner
>> going forward.
>> 
>> Tamas Blummer
>> http://bitsofproof.com
>> 
>> 
>> ------------------------------------------------------------------------------
>> This SF.net email is sponsored by Windows:
>> 
>> Build for Windows Store.
>> 
>> http://p.sf.net/sfu/windows-dev2dev
>> _______________________________________________
>> Bitcoin-development mailing list
>> Bitcoin-development@lists.sourceforge.net
>> https://lists.sourceforge.net/lists/listinfo/bitcoin-development
>> 
>> 
> 
> 


[-- Attachment #2: Type: text/html, Size: 14457 bytes --]

Re: [Bitcoin-development] Missing fRelayTxes in version

[Pieter Wuille]

On Thu, Jun 20, 2013 at 09:36:40AM +0200, Mike Hearn wrote:
> Sure but why not do that when there's an actual new field to add? Does
> anyone have a proposal for a feature that needs a new version field at the
> moment? There's no point changing the protocol now unless there's actually
> a new field to add.
> 
> Anyway I still don't see why anyone cares about this issue. The Bitcoin
> protocol does not and never has required that all messages have a fixed
> number of fields per version. Any parser written on the assumption it did
> was just buggy. Look at how tx messages are relayed for the most obvious
> example of that pattern in action - it's actually the raw byte stream
> that's stored and relayed to ensure that fields added in new versions
> aren't dropped during round-tripping. Old versions are supposed to preserve
> fields from the future.

Actually, that is not the same issue. What is being argued for here is that
the version in the version message itself should indicate which fields are
present, so a parser doesn't need to look at the length of the message. That
seems like a minor but very reasonable request to me, and it's trivial to do.
That doesn't mean that you may receive versions higher than what you know of,
and thus messages with fields you don't know about. That doesn't matter, you
can just ignore them.

I see no problem with raising the protocol version number to indicate
"all fields up to fRelayTxes are required, if the announced nVersion is above N".
In fact, I believe (though haven't checked) all previous additions to the version
message were accompanied with a protocol version (then: client version) increase
as well.

-- 
Pieter

Re: [Bitcoin-development] Missing fRelayTxes in version

[Mike Hearn]

[-- Attachment #1: Type: text/plain, Size: 2583 bytes --]

There's no problem, but there's no benefit either. It also locks us in to a
potentially problematic guarantee - what if in future we want to have, say,
two optional new pieces of data in two different messages. We don't want to
require that if version > X then you have to implement all features up to
and including that point.

Essentially the number of fields in a message is like a little version
number, just for that message. It adds flexibility to keep it that way, and
there's no downside, seeing as that bridge was already crossed and people
with parsers that can't handle it need to fix their code anyway.

So I have a slight preference for keeping things the way they are, it keeps
things flexible for future and costs nothing.



On Thu, Jun 20, 2013 at 11:06 AM, Pieter Wuille <pieter.wuille@gmail.com>wrote:

> On Thu, Jun 20, 2013 at 09:36:40AM +0200, Mike Hearn wrote:
> > Sure but why not do that when there's an actual new field to add? Does
> > anyone have a proposal for a feature that needs a new version field at
> the
> > moment? There's no point changing the protocol now unless there's
> actually
> > a new field to add.
> >
> > Anyway I still don't see why anyone cares about this issue. The Bitcoin
> > protocol does not and never has required that all messages have a fixed
> > number of fields per version. Any parser written on the assumption it did
> > was just buggy. Look at how tx messages are relayed for the most obvious
> > example of that pattern in action - it's actually the raw byte stream
> > that's stored and relayed to ensure that fields added in new versions
> > aren't dropped during round-tripping. Old versions are supposed to
> preserve
> > fields from the future.
>
> Actually, that is not the same issue. What is being argued for here is that
> the version in the version message itself should indicate which fields are
> present, so a parser doesn't need to look at the length of the message.
> That
> seems like a minor but very reasonable request to me, and it's trivial to
> do.
> That doesn't mean that you may receive versions higher than what you know
> of,
> and thus messages with fields you don't know about. That doesn't matter,
> you
> can just ignore them.
>
> I see no problem with raising the protocol version number to indicate
> "all fields up to fRelayTxes are required, if the announced nVersion is
> above N".
> In fact, I believe (though haven't checked) all previous additions to the
> version
> message were accompanied with a protocol version (then: client version)
> increase
> as well.
>
> --
> Pieter
>
>

[-- Attachment #2: Type: text/html, Size: 3236 bytes --]

Re: [Bitcoin-development] Missing fRelayTxes in version

[Turkey Breast]

[-- Attachment #1: Type: text/plain, Size: 5469 bytes --]

I don't get why this is such a contentious change?

Before I was able to use asserts to check the expected length of length of messages per protocol version, I could pass in dumb iterators that just parse the byte stream and I could serialize and deserialize a message to check the parser is correct (in debug mode).

This 'simple' change causes all that behaviour to be lost. You can no longer just use iterators but must know the remaining length (or if you use std::distance, you can only use specific std containers - not just anything with an iterator and an operator++). You cannot check the deserialization process by serializing the deserialized message and comparing it to the original data (because the bool is always present in the serializer).

It's a bit stupid you call it buggy code when this behaviour has never been present in Bitcoin. The BIP doesn't introduce any unwanted side-effects and is a trivial reasonable change.

If you want optional fields then the proper way to do it, is to either set a flag in the Services field of the "version" message to indicate different formats for messages (i.e use this template structure for a message, not that one), introduce a new message (if the changes are big), to approve/improve Stefan's BIP 32 for custom services or to have a value in the byte stream indicating which fields are present (maybe a bitfield or so).

Using a quirk of an implementation is just bad form and sloppy coding. Optional fields should have their own mechanism that allows them to remain as optional fields between protocol version upgrades.


The bitcoind software can probably be improved too, by checking that the length of the version message is consistent for the protocol version given by the connected node. Right now it makes no assumptions based on that which is a mistake (new clients can broadcast older version messages that don't have all the fields required). Probably the software should penalise hosts which do that.

What's the big deal to update the protocol version number from 70001 to 70002? It's not like we'll run out of integers. The field has now gone from optional to required now anyway - that's a behaviour change. It'd be good to enforce that. I see this as a bug.



________________________________
 From: Mike Hearn <mike@plan99.net>
To: Pieter Wuille <pieter.wuille@gmail.com> 
Cc: Bitcoin Dev <bitcoin-development@lists.sourceforge.net>; Tamas Blummer <tamas@bitsofproof.com> 
Sent: Thursday, June 20, 2013 11:17 AM
Subject: Re: [Bitcoin-development] Missing fRelayTxes in version
 


There's no problem, but there's no benefit either. It also locks us in to a potentially problematic guarantee - what if in future we want to have, say, two optional new pieces of data in two different messages. We don't want to require that if version > X then you have to implement all features up to and including that point.

Essentially the number of fields in a message is like a little version number, just for that message. It adds flexibility to keep it that way, and there's no downside, seeing as that bridge was already crossed and people with parsers that can't handle it need to fix their code anyway.

So I have a slight preference for keeping things the way they are, it keeps things flexible for future and costs nothing.




On Thu, Jun 20, 2013 at 11:06 AM, Pieter Wuille <pieter.wuille@gmail.com> wrote:

On Thu, Jun 20, 2013 at 09:36:40AM +0200, Mike Hearn wrote:
>> Sure but why not do that when there's an actual new field to add? Does
>> anyone have a proposal for a feature that needs a new version field at the
>> moment? There's no point changing the protocol now unless there's actually
>> a new field to add.
>>
>> Anyway I still don't see why anyone cares about this issue. The Bitcoin
>> protocol does not and never has required that all messages have a fixed
>> number of fields per version. Any parser written on the assumption it did
>> was just buggy. Look at how tx messages are relayed for the most obvious
>> example of that pattern in action - it's actually the raw byte stream
>> that's stored and relayed to ensure that fields added in new versions
>> aren't dropped during round-tripping. Old versions are supposed to preserve
>> fields from the future.
>
>Actually, that is not the same issue. What is being argued for here is that
>the version in the version message itself should indicate which fields are
>present, so a parser doesn't need to look at the length of the message. That
>seems like a minor but very reasonable request to me, and it's trivial to do.
>That doesn't mean that you may receive versions higher than what you know of,
>and thus messages with fields you don't know about. That doesn't matter, you
>can just ignore them.
>
>I see no problem with raising the protocol version number to indicate
>"all fields up to fRelayTxes are required, if the announced nVersion is above N".
>In fact, I believe (though haven't checked) all previous additions to the version
>message were accompanied with a protocol version (then: client version) increase
>as well.
>
>--
>Pieter
>
>

------------------------------------------------------------------------------
This SF.net email is sponsored by Windows:

Build for Windows Store.

http://p.sf.net/sfu/windows-dev2dev
_______________________________________________
Bitcoin-development mailing list
Bitcoin-development@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/bitcoin-development

[-- Attachment #2: Type: text/html, Size: 9790 bytes --]

Re: [Bitcoin-development] Missing fRelayTxes in version

[Mike Hearn]

[-- Attachment #1: Type: text/plain, Size: 7206 bytes --]

Sure, the issue isn't running out of integers, it's that you have to handle
the case of truncated messages whether you like it or not so it doesn't add
any simplicity. Even if Bitcoin-Qt starts only sending the new field with a
new version number, there are tens of thousands of bitcoinj based wallets
out there now that send the current version number and the fRelayTx field
as well, so you cannot assume anything about whether the field will exist
or not based on the version number regardless of what is changed on the C++
side. Assuming you care about your code being able to serve Bloom-filtering
clients of course.

With regards to relying on quirks, etc, this is the old "is the protocol
defined by Satoshi's code" debate again ... as I said, version messages
have always had a variable number of fields. You didn't notice before
because it was a long time since any fields were added. Perhaps it's indeed
not ideal, perhaps if Bitcoin was designed in 2013 it'd be using protobufs
or some other pre-packaged serialization system. But it is what it is.



On Thu, Jun 20, 2013 at 12:37 PM, Turkey Breast <turkeybreast@yahoo.com>wrote:

> I don't get why this is such a contentious change?
>
> Before I was able to use asserts to check the expected length of length of
> messages per protocol version, I could pass in dumb iterators that just
> parse the byte stream and I could serialize and deserialize a message to
> check the parser is correct (in debug mode).
>
> This 'simple' change causes all that behaviour to be lost. You can no
> longer just use iterators but must know the remaining length (or if you use
> std::distance, you can only use specific std containers - not just anything
> with an iterator and an operator++). You cannot check the deserialization
> process by serializing the deserialized message and comparing it to the
> original data (because the bool is always present in the serializer).
>
> It's a bit stupid you call it buggy code when this behaviour has never
> been present in Bitcoin. The BIP doesn't introduce any unwanted
> side-effects and is a trivial reasonable change.
>
> If you want optional fields then the proper way to do it, is to either set
> a flag in the Services field of the "version" message to indicate different
> formats for messages (i.e use this template structure for a message, not
> that one), introduce a new message (if the changes are big), to
> approve/improve Stefan's BIP 32 for custom services or to have a value in
> the byte stream indicating which fields are present (maybe a bitfield or
> so).
>
> Using a quirk of an implementation is just bad form and sloppy coding.
> Optional fields should have their own mechanism that allows them to remain
> as optional fields between protocol version upgrades.
>
> The bitcoind software can probably be improved too, by checking that the
> length of the version message is consistent for the protocol version given
> by the connected node. Right now it makes no assumptions based on that
> which is a mistake (new clients can broadcast older version messages that
> don't have all the fields required). Probably the software should penalise
> hosts which do that.
>
> What's the big deal to update the protocol version number from 70001 to
> 70002? It's not like we'll run out of integers. The field has now gone from
> optional to required now anyway - that's a behaviour change. It'd be good
> to enforce that. I see this as a bug.
>
>   ------------------------------
>  *From:* Mike Hearn <mike@plan99.net>
> *To:* Pieter Wuille <pieter.wuille@gmail.com>
> *Cc:* Bitcoin Dev <bitcoin-development@lists.sourceforge.net>; Tamas
> Blummer <tamas@bitsofproof.com>
> *Sent:* Thursday, June 20, 2013 11:17 AM
> *Subject:* Re: [Bitcoin-development] Missing fRelayTxes in version
>
> There's no problem, but there's no benefit either. It also locks us in to
> a potentially problematic guarantee - what if in future we want to have,
> say, two optional new pieces of data in two different messages. We don't
> want to require that if version > X then you have to implement all features
> up to and including that point.
>
> Essentially the number of fields in a message is like a little version
> number, just for that message. It adds flexibility to keep it that way, and
> there's no downside, seeing as that bridge was already crossed and people
> with parsers that can't handle it need to fix their code anyway.
>
> So I have a slight preference for keeping things the way they are, it
> keeps things flexible for future and costs nothing.
>
>
>
> On Thu, Jun 20, 2013 at 11:06 AM, Pieter Wuille <pieter.wuille@gmail.com>wrote:
>
> On Thu, Jun 20, 2013 at 09:36:40AM +0200, Mike Hearn wrote:
> > Sure but why not do that when there's an actual new field to add? Does
> > anyone have a proposal for a feature that needs a new version field at
> the
> > moment? There's no point changing the protocol now unless there's
> actually
> > a new field to add.
> >
> > Anyway I still don't see why anyone cares about this issue. The Bitcoin
> > protocol does not and never has required that all messages have a fixed
> > number of fields per version. Any parser written on the assumption it did
> > was just buggy. Look at how tx messages are relayed for the most obvious
> > example of that pattern in action - it's actually the raw byte stream
> > that's stored and relayed to ensure that fields added in new versions
> > aren't dropped during round-tripping. Old versions are supposed to
> preserve
> > fields from the future.
>
> Actually, that is not the same issue. What is being argued for here is that
> the version in the version message itself should indicate which fields are
> present, so a parser doesn't need to look at the length of the message.
> That
> seems like a minor but very reasonable request to me, and it's trivial to
> do.
> That doesn't mean that you may receive versions higher than what you know
> of,
> and thus messages with fields you don't know about. That doesn't matter,
> you
> can just ignore them.
>
> I see no problem with raising the protocol version number to indicate
> "all fields up to fRelayTxes are required, if the announced nVersion is
> above N".
> In fact, I believe (though haven't checked) all previous additions to the
> version
> message were accompanied with a protocol version (then: client version)
> increase
> as well.
>
> --
> Pieter
>
>
>
>
> ------------------------------------------------------------------------------
> This SF.net email is sponsored by Windows:
>
> Build for Windows Store.
>
> http://p.sf.net/sfu/windows-dev2dev
> _______________________________________________
> Bitcoin-development mailing list
> Bitcoin-development@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/bitcoin-development
>
>
>
>
> ------------------------------------------------------------------------------
> This SF.net email is sponsored by Windows:
>
> Build for Windows Store.
>
> http://p.sf.net/sfu/windows-dev2dev
> _______________________________________________
> Bitcoin-development mailing list
> Bitcoin-development@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/bitcoin-development
>
>

[-- Attachment #2: Type: text/html, Size: 11621 bytes --]

Re: [Bitcoin-development] Missing fRelayTxes in version

[Pieter Wuille]

[-- Attachment #1: Type: text/plain, Size: 6744 bytes --]

Let's just increase the version number and be done with this discussion.
It's a small benefit, but it simplifies things and it's trivial to do.

I don't understand how a policy of requiring version increases could limit
future extensions: after the version/verack exchange, the protocol version
is negotiated between peers, and there is no need for anything optional
anymore.

Note thay this is just about parsing, not about relaying - you should still
relay parts of a message you haven't parsed. But that doesn't apply to the
version message anyway, which is the only place where this matters.

-- 
Pieter
 On 20 Jun 2013 12:38, "Turkey Breast" <turkeybreast@yahoo.com> wrote:

> I don't get why this is such a contentious change?
>
> Before I was able to use asserts to check the expected length of length of
> messages per protocol version, I could pass in dumb iterators that just
> parse the byte stream and I could serialize and deserialize a message to
> check the parser is correct (in debug mode).
>
> This 'simple' change causes all that behaviour to be lost. You can no
> longer just use iterators but must know the remaining length (or if you use
> std::distance, you can only use specific std containers - not just anything
> with an iterator and an operator++). You cannot check the deserialization
> process by serializing the deserialized message and comparing it to the
> original data (because the bool is always present in the serializer).
>
> It's a bit stupid you call it buggy code when this behaviour has never
> been present in Bitcoin. The BIP doesn't introduce any unwanted
> side-effects and is a trivial reasonable change.
>
> If you want optional fields then the proper way to do it, is to either set
> a flag in the Services field of the "version" message to indicate different
> formats for messages (i.e use this template structure for a message, not
> that one), introduce a new message (if the changes are big), to
> approve/improve Stefan's BIP 32 for custom services or to have a value in
> the byte stream indicating which fields are present (maybe a bitfield or
> so).
>
> Using a quirk of an implementation is just bad form and sloppy coding.
> Optional fields should have their own mechanism that allows them to remain
> as optional fields between protocol version upgrades.
>
> The bitcoind software can probably be improved too, by checking that the
> length of the version message is consistent for the protocol version given
> by the connected node. Right now it makes no assumptions based on that
> which is a mistake (new clients can broadcast older version messages that
> don't have all the fields required). Probably the software should penalise
> hosts which do that.
>
> What's the big deal to update the protocol version number from 70001 to
> 70002? It's not like we'll run out of integers. The field has now gone from
> optional to required now anyway - that's a behaviour change. It'd be good
> to enforce that. I see this as a bug.
>
>   ------------------------------
>  *From:* Mike Hearn <mike@plan99.net>
> *To:* Pieter Wuille <pieter.wuille@gmail.com>
> *Cc:* Bitcoin Dev <bitcoin-development@lists.sourceforge.net>; Tamas
> Blummer <tamas@bitsofproof.com>
> *Sent:* Thursday, June 20, 2013 11:17 AM
> *Subject:* Re: [Bitcoin-development] Missing fRelayTxes in version
>
> There's no problem, but there's no benefit either. It also locks us in to
> a potentially problematic guarantee - what if in future we want to have,
> say, two optional new pieces of data in two different messages. We don't
> want to require that if version > X then you have to implement all features
> up to and including that point.
>
> Essentially the number of fields in a message is like a little version
> number, just for that message. It adds flexibility to keep it that way, and
> there's no downside, seeing as that bridge was already crossed and people
> with parsers that can't handle it need to fix their code anyway.
>
> So I have a slight preference for keeping things the way they are, it
> keeps things flexible for future and costs nothing.
>
>
>
> On Thu, Jun 20, 2013 at 11:06 AM, Pieter Wuille <pieter.wuille@gmail.com>wrote:
>
> On Thu, Jun 20, 2013 at 09:36:40AM +0200, Mike Hearn wrote:
> > Sure but why not do that when there's an actual new field to add? Does
> > anyone have a proposal for a feature that needs a new version field at
> the
> > moment? There's no point changing the protocol now unless there's
> actually
> > a new field to add.
> >
> > Anyway I still don't see why anyone cares about this issue. The Bitcoin
> > protocol does not and never has required that all messages have a fixed
> > number of fields per version. Any parser written on the assumption it did
> > was just buggy. Look at how tx messages are relayed for the most obvious
> > example of that pattern in action - it's actually the raw byte stream
> > that's stored and relayed to ensure that fields added in new versions
> > aren't dropped during round-tripping. Old versions are supposed to
> preserve
> > fields from the future.
>
> Actually, that is not the same issue. What is being argued for here is that
> the version in the version message itself should indicate which fields are
> present, so a parser doesn't need to look at the length of the message.
> That
> seems like a minor but very reasonable request to me, and it's trivial to
> do.
> That doesn't mean that you may receive versions higher than what you know
> of,
> and thus messages with fields you don't know about. That doesn't matter,
> you
> can just ignore them.
>
> I see no problem with raising the protocol version number to indicate
> "all fields up to fRelayTxes are required, if the announced nVersion is
> above N".
> In fact, I believe (though haven't checked) all previous additions to the
> version
> message were accompanied with a protocol version (then: client version)
> increase
> as well.
>
> --
> Pieter
>
>
>
>
> ------------------------------------------------------------------------------
> This SF.net email is sponsored by Windows:
>
> Build for Windows Store.
>
> http://p.sf.net/sfu/windows-dev2dev
> _______________________________________________
> Bitcoin-development mailing list
> Bitcoin-development@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/bitcoin-development
>
>
>
>
> ------------------------------------------------------------------------------
> This SF.net email is sponsored by Windows:
>
> Build for Windows Store.
>
> http://p.sf.net/sfu/windows-dev2dev
> _______________________________________________
> Bitcoin-development mailing list
> Bitcoin-development@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/bitcoin-development
>
>

[-- Attachment #2: Type: text/html, Size: 11057 bytes --]

Re: [Bitcoin-development] Missing fRelayTxes in version

[Mike Hearn]

[-- Attachment #1: Type: text/plain, Size: 8638 bytes --]

As I said, there's no benefit. Even if we do that on the C++ side, you
still have to handle connections from bitcoinj clients which will send the
field with the old version number. You can't assume they'll all be updated
simultaneously, even though both the Android app and MultiBit do have
update notifications these days and eventually old versions will presumably
disappear.

Re: flexibility. Let's say version V+1 adds a complicated new set of data
to some messages. Not every client wants or needs the feature enabled by
them.

Now version V+2 adds a simple extension to a basic message that everyone
wants/needs.

To get the latter feature, all clients now have to support the first
feature as well because the version number is monotonic.

OK, we can use a service bit to handle these cases, if we anticipate that
not all clients will want the first feature. But then again, we can also
use the presence of the additional data as the ground truth instead of
duplicating that fact. I don't really mind either way. It just seems that
parsing always requires you to be able to handle truncated messages anyway
(without asserting or crashing), because a bogus client can always send you
partial data. So I don't see what effort is saved.



On Thu, Jun 20, 2013 at 12:52 PM, Pieter Wuille <pieter.wuille@gmail.com>wrote:

> Let's just increase the version number and be done with this discussion.
> It's a small benefit, but it simplifies things and it's trivial to do.
>
> I don't understand how a policy of requiring version increases could limit
> future extensions: after the version/verack exchange, the protocol version
> is negotiated between peers, and there is no need for anything optional
> anymore.
>
> Note thay this is just about parsing, not about relaying - you should
> still relay parts of a message you haven't parsed. But that doesn't apply
> to the version message anyway, which is the only place where this matters.
>
> --
> Pieter
>  On 20 Jun 2013 12:38, "Turkey Breast" <turkeybreast@yahoo.com> wrote:
>
>> I don't get why this is such a contentious change?
>>
>> Before I was able to use asserts to check the expected length of length
>> of messages per protocol version, I could pass in dumb iterators that just
>> parse the byte stream and I could serialize and deserialize a message to
>> check the parser is correct (in debug mode).
>>
>> This 'simple' change causes all that behaviour to be lost. You can no
>> longer just use iterators but must know the remaining length (or if you use
>> std::distance, you can only use specific std containers - not just anything
>> with an iterator and an operator++). You cannot check the deserialization
>> process by serializing the deserialized message and comparing it to the
>> original data (because the bool is always present in the serializer).
>>
>> It's a bit stupid you call it buggy code when this behaviour has never
>> been present in Bitcoin. The BIP doesn't introduce any unwanted
>> side-effects and is a trivial reasonable change.
>>
>> If you want optional fields then the proper way to do it, is to either
>> set a flag in the Services field of the "version" message to indicate
>> different formats for messages (i.e use this template structure for a
>> message, not that one), introduce a new message (if the changes are big),
>> to approve/improve Stefan's BIP 32 for custom services or to have a value
>> in the byte stream indicating which fields are present (maybe a bitfield or
>> so).
>>
>> Using a quirk of an implementation is just bad form and sloppy coding.
>> Optional fields should have their own mechanism that allows them to remain
>> as optional fields between protocol version upgrades.
>>
>> The bitcoind software can probably be improved too, by checking that the
>> length of the version message is consistent for the protocol version given
>> by the connected node. Right now it makes no assumptions based on that
>> which is a mistake (new clients can broadcast older version messages that
>> don't have all the fields required). Probably the software should penalise
>> hosts which do that.
>>
>> What's the big deal to update the protocol version number from 70001 to
>> 70002? It's not like we'll run out of integers. The field has now gone from
>> optional to required now anyway - that's a behaviour change. It'd be good
>> to enforce that. I see this as a bug.
>>
>>   ------------------------------
>>  *From:* Mike Hearn <mike@plan99.net>
>> *To:* Pieter Wuille <pieter.wuille@gmail.com>
>> *Cc:* Bitcoin Dev <bitcoin-development@lists.sourceforge.net>; Tamas
>> Blummer <tamas@bitsofproof.com>
>> *Sent:* Thursday, June 20, 2013 11:17 AM
>> *Subject:* Re: [Bitcoin-development] Missing fRelayTxes in version
>>
>> There's no problem, but there's no benefit either. It also locks us in to
>> a potentially problematic guarantee - what if in future we want to have,
>> say, two optional new pieces of data in two different messages. We don't
>> want to require that if version > X then you have to implement all features
>> up to and including that point.
>>
>> Essentially the number of fields in a message is like a little version
>> number, just for that message. It adds flexibility to keep it that way, and
>> there's no downside, seeing as that bridge was already crossed and people
>> with parsers that can't handle it need to fix their code anyway.
>>
>> So I have a slight preference for keeping things the way they are, it
>> keeps things flexible for future and costs nothing.
>>
>>
>>
>> On Thu, Jun 20, 2013 at 11:06 AM, Pieter Wuille <pieter.wuille@gmail.com>wrote:
>>
>> On Thu, Jun 20, 2013 at 09:36:40AM +0200, Mike Hearn wrote:
>> > Sure but why not do that when there's an actual new field to add? Does
>> > anyone have a proposal for a feature that needs a new version field at
>> the
>> > moment? There's no point changing the protocol now unless there's
>> actually
>> > a new field to add.
>> >
>> > Anyway I still don't see why anyone cares about this issue. The Bitcoin
>> > protocol does not and never has required that all messages have a fixed
>> > number of fields per version. Any parser written on the assumption it
>> did
>> > was just buggy. Look at how tx messages are relayed for the most obvious
>> > example of that pattern in action - it's actually the raw byte stream
>> > that's stored and relayed to ensure that fields added in new versions
>> > aren't dropped during round-tripping. Old versions are supposed to
>> preserve
>> > fields from the future.
>>
>> Actually, that is not the same issue. What is being argued for here is
>> that
>> the version in the version message itself should indicate which fields are
>> present, so a parser doesn't need to look at the length of the message.
>> That
>> seems like a minor but very reasonable request to me, and it's trivial to
>> do.
>> That doesn't mean that you may receive versions higher than what you know
>> of,
>> and thus messages with fields you don't know about. That doesn't matter,
>> you
>> can just ignore them.
>>
>> I see no problem with raising the protocol version number to indicate
>> "all fields up to fRelayTxes are required, if the announced nVersion is
>> above N".
>> In fact, I believe (though haven't checked) all previous additions to the
>> version
>> message were accompanied with a protocol version (then: client version)
>> increase
>> as well.
>>
>> --
>> Pieter
>>
>>
>>
>>
>> ------------------------------------------------------------------------------
>> This SF.net email is sponsored by Windows:
>>
>> Build for Windows Store.
>>
>> http://p.sf.net/sfu/windows-dev2dev
>> _______________________________________________
>> Bitcoin-development mailing list
>> Bitcoin-development@lists.sourceforge.net
>> https://lists.sourceforge.net/lists/listinfo/bitcoin-development
>>
>>
>>
>>
>> ------------------------------------------------------------------------------
>> This SF.net email is sponsored by Windows:
>>
>> Build for Windows Store.
>>
>> http://p.sf.net/sfu/windows-dev2dev
>> _______________________________________________
>> Bitcoin-development mailing list
>> Bitcoin-development@lists.sourceforge.net
>> https://lists.sourceforge.net/lists/listinfo/bitcoin-development
>>
>>
>
> ------------------------------------------------------------------------------
> This SF.net email is sponsored by Windows:
>
> Build for Windows Store.
>
> http://p.sf.net/sfu/windows-dev2dev
> _______________________________________________
> Bitcoin-development mailing list
> Bitcoin-development@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/bitcoin-development
>
>

[-- Attachment #2: Type: text/html, Size: 13662 bytes --]