April 18, 2026
Parse wars: grab the popcorn
Binary Encodings for JSON and Variant
Faster than JSON? Devs cheer while skeptics warn of hidden costs
TLDR: A new post argues binary encodings and “VARIANT”-style storage can make JSON lookups dramatically faster, spotlighting parsing as the real slowdown. The comments split hard: speed fans love the boost, while others warn that pre-indexing, extra data, and memory trade-offs make this no free lunch.
JSON just got dragged. A new post claims the slow part isn’t JSON itself—it’s the parsing—and shows that a simple binary format and a “VARIANT” style design can make repeated lookups scream, with a headline-grabbing “2,346x faster” microbenchmark. The punchline: text JSON has no random access, so you waste time reading stuff you don’t need. Cue the comment fireworks.
Fans of speed are high‑fiving, pointing to microbench numbers and saying “please, end my coffee‑break parses.” The “coffee” line from the post became a meme, with commenters joking they can brew a pour-over while a million JSON rows get parsed. But the top counter‑take, from kstenerud, is a cold shower: “Fast internal scanning isn’t free”—you pay with pre‑indexing, extra data, and losing easy, incremental writes. Translation: you can go faster, but you’ll carry more baggage.
On the other flank, C++ developer boricj waves the memory flag, praising streaming: formats like JSON, YAML, and binary CBOR can stream one character at a time so you don’t blow out RAM. Meanwhile, the post’s jab that “BSON isn’t enough” riled the MongoDB crowd (clutch those pearls), while others rallied around “VARIANT will replace JSON” like it’s the new house band. For the curious, here’s simdjson and a primer on Parquet. The vibe: speed demons vs trade‑off realists, with plenty of popcorn.
Key Points
- •The article measures JSON parsing and retrieval costs using simdjson, focusing on parse-plus-query time.
- •On a 2019 Intel MacBook Pro, a realistic twitter.json query took ~136,784 ns per lookup.
- •Synthetic benchmarks show linear growth in lookup time for later keys/indices, revealing JSON’s lack of random access.
- •The author states that minimal binary encodings can dramatically accelerate lookups, citing a 2,346x improvement.
- •The piece frames why BSON may not suffice for repeated-query acceleration and introduces Parquet’s VARIANT as a potential alternative.