That's awkward, because the gperf dependency is only needed every 2 years or so, when a new keyword is added. The genererated C code needs to be added to git, and only needs a regen if changed. I do that all the time via gperf or via some perl-generated perfect hash tables, eg for static keywords, Config tables, unicode tables, ... The memory win is always dramatic, like 20x better typically. I also use some standard gperf-fixup scripts for easier inclusion into my code.