[core] Run the accept hook before opening caller's socket

[xhaakon]

In listener mode, open the socket created for a connecting caller only after its accept hook has been run. This gives the hook an opportunity to set some 'pre' options (like SRTO_RCVBUF) on the socket before it gets passed to srt_accept().

[maxsharabayko]

Hi @xhaakon
In your suggestion the "accept" hook gets a chance to set socket options with "PRE" binding, like SRTO_RCVBUF.
Before the change, setting e.g. SRTO_RCVBUF inside the hook is impossible, and will return the MJ_NOTSUP, MN_ISBOUND error.

Could you please elaborate a bit on a possible use case? I wonder why would a listener need to change the settings it had configured. I see, for example, a possible need to set receiver buffer size per connection when e.g. the streaming bitrate is known from the connection, maybe read from StreamID... 🤔

TODO

• Check if setting a different SRTO_ENFORCEDENCRYPTION from inside the accept callback is handled properly.

[xhaakon]

Hi @maxsharabayko

I see, for example, a possible need to set receiver buffer size per connection

Yes, what you're hinting at is basically our desired use case.

Our application would like to tune the receive buffer based on the formulas you laid down in #703. The sender (caller) would measure the link bandwidth and RTT, do the calculations and remember the resulting buffer sizes. Next time it connects to the receiver (listener), it would request the desired buffer size by putting it into StreamID.

Setting SRTO_RCVBUF beforehand on the listening socket is not useful here, because each connecting caller has a different link quality and so the buffer size we want to use for each individual client will also differ.