Thanks for the PR!

If I understand correctly, your are supporting latching only for the TCP transport. In that case, I do not understand the need for the complex "latch helper" setup. I would suggest to re-send latched messages whenever TCPSender::connect() is successful, indicating a new receiver has started.

Oh man, could it be that simple? I admit I haven't understood the whole relay code, but it looked impossible to me to tell when the client has connected. But as you pointed me to the right direction, it now seems clear to me.

Few questions arise:

1. Isn't there a possible infinite loop if I resend the latched messages from connect, which is itself called from the send method? I mean in case the connection is very unreliable or so...
2. Now, when it's almost clear the whole cumbersome thing was not needed, is there a chance to make use of it on the UDP transport?

Isn't there a possible infinite loop if I resend the latched messages from connect, which is itself called from the send method? I mean in case the connection is very unreliable or so...

I guess so. At the very least we should prevent the inner send() routine (resending a latched message) from calling connect() again. Maybe an additional parameter reconnect with default value true?

Now, when it's almost clear the whole cumbersome thing was not needed, is there a chance to make use of it on the UDP transport?

I have never seen a use case for latched topics over UDP. Typically, UDP data is streaming data, where latching makes no sense. Or do you have an actual use case for that?

Ah, and there is one more issue: If there are no messages to transmit, send will never get called, and will never notice the dropped connection. So we should introduce a timer which checks whether the connection is still open, and if not, reconnects (and re-sends latched messages).

Ok, I'm trying to rewrite and will send a new PR in a while. Your anti-looping suggestion sounds good to me.

And regarding latched UDP topics - it probably really doesn't make sense to latch UDP topics... Should I add a warning to the UDP transport when it finds a topic which requires latching?

Well, now I think if this wasn't my original issue... I kind of get lost thinking about all the possible cases. Wouldn't periodical connection check be a problem? Either performance-wise, or just for the case that the receiver is stopped and is not mentioned to be re-run? (e.g. unsubscribing from all topics on the receiving side).

Wait, is the dropped connection really a problem?

The receiver also sets the "latched" flag at the publishers it creates. So even if the connection drops, the subscribers on the receiving side will get properly handled. Or do I miss some other important point?