Is replier_id always needed in Reply?

[Mallets]

Describe the release item

Reply includes a mandatory replier_id of type ZenohId.
However, the Sample accepted in the reply() contains a SourceInfo containing an optional ZenohId.

In the light of this, shouldn't replier_id be of type Option<ZenohId> instead of ZenohId? If ZenohId is always present in the Reply, then who is setting it all the time? Is this behaviour compatible with zenoh-pico?

[Mallets]

It seems the ZenohId is set in https://github.com/eclipse-zenoh/zenoh/blob/5601669ae4f1409e2080682acdb54a8f1d700ee8/zenoh/src/session.rs#L2400C58-L2400C65.
I believe the current behaviour is not the expected one and the replier_id should be an Option<ZenohId>.

[Mallets]

Linking C structure in zenoh-pico: z_reply_data_t

[p-avital]

IIRC, the distinction is there so that storages that reply can be identified, while the sample would retain its original SourceInfo. @OlivierHecart?

[milyin]

@Mallets @OlivierHecart
@DariusIMP noticed, that there is a problem with this issue. Before the fix the reply_id was set to random, which was definitely wrong: https://github.com/eclipse-zenoh/zenoh/pull/1138/files#diff-f0aac9d83f0d53d0be9a6ee02269c26605c3920dbd6782f744e59c21aec235d5L2311

But it seems that this field is unrelated to SourceId field of the sample. In case of error this field is set to zid of the replying session. I see no reason to not to do the same for normal replies. So maybe we should remove this option and just always set this field with current session zid ?

[OlivierHecart]

@Mallets @OlivierHecart @DariusIMP noticed, that there is a problem with this issue. Before the fix the reply_id was set to random, which was definitely wrong: https://github.com/eclipse-zenoh/zenoh/pull/1138/files#diff-f0aac9d83f0d53d0be9a6ee02269c26605c3920dbd6782f744e59c21aec235d5L2311

But it seems that this field is unrelated to SourceId field of the sample. In case of error this field is set to zid of the replying session. I see no reason to not to do the same for normal replies. So maybe we should remove this option and just always set this field with current session zid ?

replier_id has indeed nothing to do with SourceId. The reason for making this an Option is that sometimes we might not have the information on the real replier or there may not be a clearly identifiable replier (infrastructure error for example). In such cases it's better to set None rather to mislead users to believe that the local session replied.

[milyin]

@OlivierHecart also mentioned, that it's probably better to use EntityGlobalId here instead of ZenohId

[milyin]

[stepkun]

Imho the fields/structs source_id & replier_id should either be reliable or removed.
In case they are not reliable, users who need that information have to implement and transport the information themselves, so they are obsolete and only bloating the transported data amount.
And in case they are reliable, access methods need not return an Option anymore - will make programmers life much easier.

[milyin]

@stepkun thank you for your thoughts
After discussion with @OlivierHecart and @Mallets it was decided to put this option under "unstable" until final solution. Anyway there is no coherent way to set it on sending side, so before it's done let's keep it unstable.