%u and %llu use sizes that we want them to use, that's why we have our own printf implementation — to have the same output independently from local libc. We can postulate that %u is 32-bit unsigned int and that %llu is 64-bit unsigned long long int. Cannot we?

%u and %llu are 32-bit and 64-bit in modern desktops and servers, yes, but not necessarily because the standards don’t enforce so. If say, someone wants to embed a client in an Arduino, their compiler would complain that our code should use %lu.

And we can’t deviate our my_snprintf from the standards either, lest our implementation be inconsistent with GCC -Wformat (MDEV-21978), not to mention that C sprintf also uses %u-%u-%llu.

you're right, and this is technically true. but practically I don't think MariaDB will compile anywhere where longlong isn't 64 bit and int isn't 32 bit.

long is non-portable and can be either 32-bit or 64-bit, both will work.

rpl_gtid::seq_no currently uses uint64 which is defined as ulonglong, so it doesn’t matter currently,
However, if MDEV-35460 replaces uint64 with uint64_t, we’ll get -Wformat warnings on Linux because their uint64_t is unsigned long.