April 30, 2026
Search party turns into flame war
You can beat the binary search
A tiny search trick sparked a huge fight over whether the old classic was ever that great
TLDR: A programmer built a faster search method for certain small sorted lists by checking many values at once, but commenters instantly argued over whether it’s a real breakthrough or just a dressed-up old idea. The biggest reaction was less “wow” and more “show me bigger tests.”
A programmer showed off a new way to search a sorted list faster than the famous binary search, and the internet immediately did what it does best: turned a niche speed test into a full-blown comment-section cage match. The basic claim is simple enough for normal humans: instead of checking one spot at a time, this method chops the list into chunks and uses special chip features to check a bunch of numbers at once. On paper, it’s a clever shortcut for the small, tightly packed lists used in Roaring Bitmap, a data format built for speed.
But the comments? Absolute chaos. One camp was impressed by the hardware-savvy thinking. Another basically said, “Cool trick, but this only matters because you picked a very specific kind of data.” One commenter argued that if you know more about your data, you can beat binary search in other ways too. Another was openly skeptical, asking whether “quaternary search” is just binary search wearing a fake mustache and calling itself innovative.
Then came the jokes. One person roasted the benchmarks for topping out around 4,000 items, calling that “basically non-data” and comparing the whole contest to a 1cm car race. Another dropped a delightfully nerdy flex: the log base 2 of all atoms in the universe is only about 300, so maybe let’s all calm down about logarithms. In other words, the article offered an optimization, but the community delivered the real entertainment: pedantry, skepticism, bragging rights, and top-tier nerd comedy.
Key Points
- •The article contrasts linear search and binary search for locating values in sorted arrays and notes their C++ equivalents as `std::find` and `std::binary_search`.
- •It uses Roaring Bitmap arrays of 16-bit integers, with sizes from 1 to 4096, as the practical setting for optimizing membership tests.
- •The article argues that modern 64-bit ARM and x64 processors can compare multiple 16-bit integers in parallel using SIMD instructions.
- •It proposes a search method called SIMD Quad that combines quaternary interpolation search across 16-element blocks with SIMD equality checks inside the selected block.
- •For arrays smaller than 16 elements, the method falls back to linear search; for larger arrays, it uses block boundaries and SIMD instructions such as NEON or SSE2.