[partialzero]

Indeed this is very confusing! The program implements a pretty standard state machine (ok), but there is nothing apparently async here. The auth alludes to combining the state machine with async IO in this paper (https://github.com/angea/pocorgtfo/blob/master/contents/arti...), but this implementation is just using fread to (synchronously) read a chunk of bytes.

Furthermore, given disk caching and memory mapping, I'm not convinced async IO would really be that astonishingly different, as individual reads are going to be amortized over pretty much the same bulk reads that the sample program is doing.

As the author says themselves, it seems the main win is hand implementing the incremental utf8 parsing instead of calling a library/os function.