Hi, Actually i thought about the tombstone problem myself a few days ago, and i also don't see another solution than to retract all revisions, otherwise we would need to implement some corrections aware server, which is probably overkill when the client can simply send out 3 moderate messages instead of one. I guess in practice moderating/retracting often corrected messages will probably not be the case anyway.
This of course will pose a problem for many clients, as nothing in 308 even hints at it being necessary to keep all correction ids around, which would now be a necessity to implement message moderation and retraction. I vaguely remember that multiple clients simply replace the content of the original message in the database and probably not storing correction message ids. Also for reactions (XEP-0444) this feels all crazy complicated, if you have to gather reactions over multiple messages, deduplicating, then picking the right message id if you want to change your reaction, because you reacted on the 2nd of 3 corrections .. Maybe im the only one, but all this feels quite hard to implement and get right, especially if you don't know all this beforehand and try to gradually implement a new xmpp client. If you make the wrong architecture decision when implementing corrections, you can go back to zero. Also libraries cannot help with that at all, because its all in the right database design from start. All in all it feels a bit frustrating to implement these new reference XEPs. I expect that many bugs will plague the eco system because of how complicated they can get to implement. Regards Philipp On Mon, Apr 21, 2025, at 22:40, Marvin W wrote: > Hi, > > XEP-0308 is currently not a requirement of XEP-0424/0425/0461 and is largely > an unrelated feature. It's of course possible to design around it being a > requirement, but that also would be weird, given that XEP-0308 has a number > of known limitations (e.g. correcting older than the latest messages is > explicitly out of scope) and thus is likely to be replaced eventually. > > As long as XEP-0308 is not a requirement, you must live under the assumption > of some clients supporting it and some not supporting it. There's also a > number of practically relevant cases where XEP-0308 can't be used safely > (e.g. semi-anonymous MUCs without occupant-id, if the messages are received > from archive or without all participants online between the original message > and its correction) and clients are also typically partial knowledge, meaning > they might not be able to fully comply with XEP-0308 (e.g. if they don't know > the original message because it is out of the sync window). As such, it is > just a sane default to assume that even if clients are aware of XEP-0308, > they might not be able to use it. And to extend on this, it's easier to just > assume the client does not support it (because every client can just fallback > to the behavior of not supporting it). > > XEP-0424/XEP-0425 apply to almost all clients - even if they only support > very basic XMPP features - simply due to it potentially tombstone'ing the > message before it arrives at a client. The behavior of tombstone'ing becomes > very vague if you include XEP-0308 and only the reference the original > message, as it means only that is tombstone'd, not any of the other > revisions, which thus remain stored and available. If later a client syncs > the correction, but not the tombstone, it wouldn't even know that the message > was meant to be retracted, simply because it's out of the sync window. > > For the reasons laid out before, a rule to require to always only reference > the initial message in a multi-revision scenario is conceptually flawed in > that the sending client might just be unaware what the initial message was. > In XEP-0308 that was acceptable, given the restrictions it puts on the > sending clients (by only allowing to edit ones own last message). > > As a side note: XEP-0308 is currently stable, but lacks multiple of the > formal requirements for being stable, as it has neither a schema, nor a > normative description of the <replace/> element, but only an example. > > Marvin > > On Mon, 2025-04-21 at 21:04 +0200, Philipp Hörist wrote: >> Hi, >> >> Not sure why you spec out stuff for XEP-0308 non-aware clients. >> >> XEP-308 is the oldest, its in all the compliance XEPs as a core feature that >> every IM Client should support, i think we all agree its a core part of any >> messenger application. >> >> Why would it be a problem for these XEPs to depend on XEP-308? You solve a >> problem nobody has. >> >> You state >> >> > And I agree that we should generally apply the same logic to other XEPs. >> >> then go on and make an over engineered spec to do exactly the opposite. >> >> We changed XEP-0308 to make it extra so that it is not a chain, that all >> corrections reference always the original, now you want to introduce right >> now 4 different XEPs that in the future can reference any of these in >> between correction messages? >> >> The sane thing to do is, reference the original message, always. An easy >> rule, no need to define what it means if the 2nd of 3 corrections is >> retracted or moderated or replied to. >> >> Regards >> Philipp >> >> On Mon, Apr 21, 2025, at 19:04, Marvin W wrote: >>> Hi, >>> >>> I understand the nature of XEP-0308 in such that it changes the body of >>> another message and is not to be understood as a message itself. This is >>> why if the same message is corrected multiple times, they all refer to the >>> initial message id. And I agree that we should generally apply the same >>> logic to other XEPs. >>> >>> That said, other XEPs don't depend on XEP-0308 and generally should not >>> assume XEP-0308 to be supported by sending or receiving clients. Thus these >>> XEPs don't only need to specify "which ID should be used when a message was >>> XEP-0308 corrected", but also might specify other rules and need to >>> consider how this will behave with other clients supporting or not >>> supporting XEP-0308. >>> >>> For XEP-0461, I think the best handling would be, that it is allowed to >>> reference not only the initial message, but also any of the corrections. If >>> entities receive a reply to a message that was corrected later (or was a >>> correction itself) it may decide the display the body of this specific >>> iteration of the message (instead of the latest iteration) if it displays >>> the body of the message replied to next to the reply. >>> >>> For the following log of messages: >>> - from=Alice id=a body=First >>> - from=Alice id=b correction-of=a body="Second" >>> - from=Alice id=c correction-of=a body="Third" >>> - from=Bob reply-to=b body="> Second\nReply" fallback=reply:0:9 >>> >>> The XEP-0308-aware client would then display something like >>> Alice: Third (edited) >>> Bob: > Alice wrote: *(link to above message "Third")* >>> > Second (edited) >>> Reply >>> >>> The XEP-0308-unaware client would instead display something like >>> Alice: First >>> Alice: Second >>> Alice: Third >>> >>> Bob: > Alice wrote *(link to above message "Second")* >>> > Second >>> Reply >>> >>> I think this is fair. >>> >>> For XEP-0424 my personal suggestion would be, that a retraction for a >>> corrected or correction message, should only be understood as a retraction >>> of that revision of the message. If that revision is not the latest (the >>> message was corrected further later) retracting an older revision has no >>> effect - this also applies to the original message. This makes the behavior >>> across XEP-0308-aware and -unaware clients the most consistent: >>> >>> For the following log of messages: >>> - from=Alice id=a body=First >>> - from=Alice id=b correction-of=a body="Second" >>> - from=Alice id=c correction-of=a body="Third" >>> - from=Alice retraction-of=b >>> >>> The XEP-0308-aware client would then display something like: >>> Alice: Third (edited) >>> It the client supports showing a correction history, it would display that >>> the second revision of the message was retracted. If instead the latest >>> revision is retracted, the previous (unretracted) revision is displayed. >>> >>> The XEP-0308-unaware client would instead display something like: >>> Alice: First >>> Alice: (retracted) >>> Alice: Third >>> >>> Only once all revisions of a message have been retracted, such message will >>> completely be retracted (for both XEP-0308-aware and- unaware clients). As >>> such, if a client wants to remove a message with corrections entirely, it >>> should send retractions for each revision (it's probably worth explicitly >>> allowing this to happen in a single message). >>> >>> If we agree on above for XEP-0424, XEP-0425 wouldn't strictly need any >>> additions. And because XEP-0308 is based on the messages @id even in MUCs >>> and we don't want to ask MUC servers to keep track of those, there's not a >>> lot a MUC server can do here either. >>> >>> Marvin >>> >>> >>> On Mon, 2025-04-21 at 17:01 +0200, Philipp Hörist wrote: >>>> XEP-0444 specifies it in the business rules, although its not very strict, >>>> basically says you should use the original id but everyone should also >>>> support everything else. >>>> >>>> XEP-0308 does not overwrite the original message id, the client just >>>> replaces the payload of the message. >>>> >>>> From XEP-308 >>>> > If the same message is to be corrected several times, the id of the >>>> > original message is used in each case (e.g. If a message of id 1 is >>>> > corrected by a message of id 2, a subsequent correction should correct >>>> > 1, not 2). >>>> >>>> So all corrections point to the original message. >>>> >>>> I would argue the case if the corrections XEP itself always references the >>>> original message, all other XEPs should do aswell. >>>> >>>> Regards >>>> Philipp >>>> >>>> >>>> On Mon, Apr 21, 2025, at 16:41, Tedd Sterr wrote: >>>>> Additionally, XEP-0444 (Message Reactions), and pretty much anywhere an >>>>> ID will ever be used to reference another message. >>>>> >>>>> The issue stems from XEP-0308 replacing/overwriting the message ID, and >>>>> consequently all references already using the previous message ID are >>>>> left pointing to a non-existent message. >>>>> >>>>> Amongst other things, this provides a way to avoid moderation — send an >>>>> offensive message, it gets moderated, just LMC the message and it's no >>>>> longer moderated. >>>>> >>>>> So, while it feels 'wrong' from a purist view, NOT replacing the message >>>>> ID would be the better option - old and new references will then all >>>>> point to the same message. Otherwise, presumably, clients (and servers?) >>>>> will be required to track all past message IDs for each corrected message. >>>>> >>>> >>>> _______________________________________________ >>>> Standards mailing list -- [email protected] >>>> To unsubscribe send an email to [email protected] >>> >>> >>> _______________________________________________ >>> Standards mailing list -- [email protected] >>> To unsubscribe send an email to [email protected] >>> >> >> _______________________________________________ >> Standards mailing list -- [email protected] >> To unsubscribe send an email to [email protected] > > > _______________________________________________ > Standards mailing list -- [email protected] > To unsubscribe send an email to [email protected] >
_______________________________________________ Standards mailing list -- [email protected] To unsubscribe send an email to [email protected]
