Passionate Algorithmist, Compiler Engineer @ Microsoft
Every point cloud above was computed in plain C++ and reached Python through a binding I didn’t write. No .def() , no trampoline classes, no keyword-arg boilerplate. Just bind_class<PointCloud>(m, "PointCloud") . Reflection fills in the rest at compile time. pybind11 mirror_bridge Hand-written binding lines (Open3D geom) 3,610 71 (51× less, auto-generated) PointCloud(list_of_1M_points) ctor 585 ms…
“I Love to Specialize Hash Functions for My C++ Classes” (No One, Ever)
We know Mirror Bridge generated bindings beat the standard Python interpreter. But how do they fare against the big guns?
You have a Python codebase. Thousands of lines. It works. Your team knows it. Your tests cover it. Your CI/CD deploys it.
JSON (JavaScript Object Notation) is a popular format for storing and transmitting data. It uses human-readable text to represent structured data in the form of attribute–value pairs and arrays. E.g., {"age":5, "name":"Daniel", toys:["wooden dog", "little car"]}. Ingesting and producing JSON documents can be a performance bottleneck.
TLDR Implemented a complex expected linear time algorithm for finding Minimum Spanning Tree. Found a bug in an implementation in a 13 years old paper. Code available at github