Winking allocation scheme is employed to dynamically allocate structures that are emitted during parsing process. All dynamic memory requirements of the parser are fulfilled using pre-allocated array that is co-located with on-wire message buffer and deallocated in one shot.
Original message is copied at the very beginning of the allocate buffer, followed by the empty "message heap" area. As message parsing progressing, this area is filled with data structures, for the most part referencing parts of the message area.
Internal API is provided to perform speculative reservation into the heap area and return space back to heap in case of parsing errors.
It is conjectured that such design results in an extremely efficient use of L1 cache memories, which are in the order of few tenths of kilobytes per core.
[On-wire HF Name] -=7 bit=-> [De-capitalization / Bit-compaction] -=5 bit=->
[Pre-computed Pearson Hash using 64-entries array] -=Canonical Header Id=->
[Verification using strcasencmp()].