You should not use gperf anymore, very probably. It "may" be still usefull in
use cases with _very_ intensive, time/memory critical, and  massive, _really_
massive, keyword lists. And even there, you should seriously consider
alternatives which are saner and semantically optimized to your use case. And
nowadays we know that c++ is never a good idea: the elephant in the room is the
obvious c++ compiler implementation cost, which is a significant detriment to
foster "working" alternative compilers, not to mention it is very prone to the
rube goldberg machine syndrome.

Don't be a masochist trying to reverse engineer the heuristics from the code
first hand: Read gperfp.pdf before anything else.

This is a "port to _simple_ C89 with benign bits of c99/c11". It is a properly
namespace-ized (then "reuse-able" in any other project), "one compilation unit"
project: just compile all.c and link it to your libc and libm the way you want.
Of course, bugs were introduced while porting.