Freeciv

Has this had an actual noticeable impact, or are we prematurely optimizing? As much as it pains me to admit it, there does come a point where performance optimization might not be worth making the code more complex. Most of our network traffic is compressed, so I'd hope if there's enough extra null bytes to be a significant concern, compression would mitigate that again.

If we're gonna do this, I'd prefer not to complicate the packets.def format; and more importantly, not to specify the index size limit in two places, so whenever we change a constant used as an array's declared/maximum size, we also have to check whether we now have to change a bunch of index sizes.
We could determine the required index type statically by comparing the declared max size to MAX_UINT8; or we could even do it dynamically (if there were any diff arrays with non-constant size), since at the time we send or receive a deep-diff array, we already know the length of that array, and thus the max index. Though doing it dynamically would make the generated code a lot more complex.
(If and when we're gonna be deep-diffing vectors, we'll only have the dynamic option, naturally.)

Voila. I did reformat and move around some parts of it to make that part of the code more readable, I think it's okay now.

Marko Lindqvist wrote in #note-4:

Should we have this also in S3_2, as the original change to 16bit array lengths was applied there?

Probably, yeah.

Fixed a typo (MAX_UNIT8 instead of MAX_UINT8).

Fixed an off-by-one error in JSON mode (receiving side was always ignoring the first diff element).