Fix data loss in TCP streams.

[jmattsson]

• This PR is for the dev branch rather than for master.
• This PR is compliant with the other contributing guidelines as well (if not, please describe why).
• I have thoroughly tested my contribution.
• [n/a] The code changes are reflected in the documentation at docs/en/*.

So it turns out that very occasionally^ the lwip stack hands over chained pbufs to the tcp_recv() callback. As only the first pbuf in the chain was passed into Lua, this meant that data went missing from the middle of the TCP stream (which one would certainly not expect). The fix here is to simply do multiple callbacks into Lua until the end of the chain is reached (though I've never seen more than two pbufs in the chain so far).

With this patch I'm now always successfully getting the full TCP stream, whereas before I never did.

^) During each 1meg transfer, I saw it happen 1-4 times.

[devsaurus]

Looks good 👍

[djphoenix]

Can be moved outside a loop

[jmattsson]

Yes, though it becomes (slightly) harder to follow, and you double up on the if statement as you'll need to push the port/ip each time through the loop regardless. Given that we'd normally be doing the ets_sprintf for each pbuf anyway, I felt the tradeoff wasn't worth the readability cost. If you really reckon it is, I can change it.

[djphoenix]

Agree for "harder to follow"...
Will accept 'FIXME' comment there as a solution for now.

[joysfera]

wondering if this patch could fix also the mysterious MQTT issues (where bits of TCP data were missing randomly) #1811

[jmattsson]

Afaiu mqtt uses espconn, so this shouldn't be applicable there. Someone is welcome to prove me wrong though :)