I'm curious why the author chose to use gcc 5.4.0 for the code size benchmark. Do older gcc versions generate smaller code?
We use many different compilers and compiler versions for correctness testing - various versions of GCC, LLVM, and MSVC with various optimization settings. That's important because different compilers generate different code and we want to ensure that they all generate correct code. (SQLite has found bugs in historical versions of all of those compilers!) But for benchmarking, as long as the same compiler and version is used consistently, does it really matter which compiler is used?
Additional compiler comparison data (from 2017): https://www.sqlite.org/footprint.html