We have approximately 5 formatter_items per buffer_line. Let's assume
we're on a 64-bit machine. Then there were (5 * 2) + 3 useless pointers
(104 bytes) as well as 5 * (4 + 4) = 40 bytes of wasted space because
of needless padding. That's 144 bytes already. Compared to that, this
change adds 16 bytes of overhead for an array sentinel, i.e. 128B less.
With a limit of 1000 lines per buffer, we've saved ~128kB per buffer
on completely useless data, and code complexity stays roughly the same.
All in all, memory usage for buffers should be about 50% lower.
Involves some rewrites to fit the new APIs.
SNI has been implemented Mostly just because we can, I don't think it's
widely in use and kike doesn't support this feature of the protocol either.
