Daniel Shahaf <d.s_at_daniel.shahaf.name> writes:
> Philip Martin wrote on Wed, Mar 19, 2014 at 13:17:08 +0000:
>> We have:
>> - the documented protocol
>> (? want-iprops:boolean )
>> - the released server implementation of the protocol
>> ? want-iprops:boolean
>> - the released behaviour
>> properties always sent
>> - the trunk behaviour
>> properties only sent when want-iprops is true
>> A third party client could already be using want-iprops with a 1.8
>> server to get inherited props, even with the broken behaviour.
> Not our problem. We have no obligation to support third parties who
> reimplement our protocol. We also have no obligation to make the server
> accept constructions that no released client ever generated.
So the ra_svn protocol is a private API. Is this a position we have
taken in the past? I'm not opposed to it but I wasn't aware that it was
> If I understand correctly, you're saying that no *released* client
> ever sent a want-iprops boolean on get-file and get-dir;
> so we have zero
> obligation to support a want-iprops parameter there, period. (To clarify,
> we have no obligation to support either "?B" or "(?B)" or "?(?B)".)
> We may choose to support the form that the 1.8.0-1.8.9 server code
> accepts (that is, "?B"), or we may choose to declare that a bug in
Does anyone have any opinions about which one we should choose?
> In the former case we may apply the patch.
If we decide to retain "?B", to allow further optional arguments in
future, then applying the patch has two benefits, it improves clients
using the buggy servers and it add the placeholder that will be needed
when an optional argument is added.
> In the latter case, we shouldn't apply the patch, and moreover we will
> have to "seal" get-file and get-dir --- that is, document that no
> further optional arguments may be appended to them. (That's not a
> problem --- it just means that if we want to add an argument at the end,
> we create a new protocol command "get-file2" and "get-dir2", or
> condition the new argument on a new server capability. That's to ensure
> we never send the extra arguments to a 1.8.0-1.8.9 server (which will
> misinterpret them).)
> Also in the latter case, we can either remove the "?B" from the server's
> code, or keep it --- released clients and third-parties following the
> documented protocol should behave the same with or without it, since
> they never send a bare "B" there.
> In either case, we update libsvn_ra_svn/protocol to document the
> divergence, and may choose to issue an API errata (in spite of the
> protocol spec nominally being a private API).
Philip Martin | Subversion Committer
WANdisco // *Non-Stop Data*
Received on 2014-04-08 21:13:08 CEST