parse_number() as a SQLite extension: C vs Rust

Gives measured numbers: optimized C ran 16 to 19 ns per row, about twice as fast as Rust, with a 14 KB binary versus roughly 300 KB. It also shows an agent-driven benchmark workflow backed by fuzz-testing.

• Willison's parse_number() strips commas, $, £ and % from strings like -$1,234.56 or '12.5%', keeps a leading minus, and returns NULL for anything it cannot parse, so formatted text can be summed and sorted as numbers inside SQLite.
• The first C build lost to Rust because it called strtod(): on decimal-only data it cost 106 ns per row versus 47 to 55 ns for the two Rust builds. Willison reports that Rust's equivalent of strtod is about twice as fast.
• A second prompt asking Claude to write its own C decimal parser cut decimal-only parsing from 106 ns to 18 ns per row over 2 million rows, flipping the result so C beat both Rust builds on every data mix tested.
• The two Rust builds used different bindings, Rusqlite (321 KB) and sqlite-loadable-rs (298 KB), and the sqlite-loadable build was slightly faster on every mix in Willison's table, for example 41 ns versus 46 ns per row on mixed data.

• AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
• benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.

• articlesparqlite: SPARQL 1.1 inside SQLite, written in Rustsimonwillison.net
• postAgents rebuilt SQLite in Rust from the manual aloneCursor
• articleMarkdown is 40× more token-efficient with SQLiterichardcrng

Checking sign-in…

Loading comments…