l: A new runtime for k and q
A new proprietary runtime called L for the k and q array languages—best known as the basis of kdb+, a high‑performance time‑series database widely used on Wall Street—is drawing attention for promising major speedups, including SIMD‑friendly execution and computation directly on compressed columns. Commenters are split between enthusiasm for opening up a historically expensive, niche technology to more developers and skepticism about the closed-source model, vague marketing claims, and an AI‑generated “vibecoded” website that makes the system hard to evaluate from the outside. The thread also touches on legal risks given Kx Systems’ history of aggressive IP enforcement and on the broader appeal of array languages as a way to rethink how programmers work with large datasets.
What L Is and Why It Matters
- L is described as a new runtime/interpreter for k and q/qSQL, the array languages behind kdb+, widely used for tick/time-series data in finance.
- It aims for full language and database compatibility with existing k/q systems so users can run similar workloads locally, for free.
- Several comments frame this as a big deal for quants who currently pay high prices for kdb+, especially as its vendor is seen as expensive and aggressive on IP.
Array Languages and Use Cases
- k and q are characterized as terse, array-oriented languages in the APL/J family, where vectors/arrays are the primary mental and execution unit.
- Commenters note that learning an array language often changes how one thinks about programming, beyond the niche financial use.
- Some say the main practical motivation to learn q is access to highly paid jobs in finance and trading.
Performance, Design, and Benchmarks
- L positions itself as a high-performance engine with:
- Fusion of chained operations to avoid intermediates and enable in-place updates.
- “Compute on compressed” vectors: using frame-of-reference–style compression to reduce bit-width (e.g., i64→i16/i8) and then operating directly on compressed data.
- This is claimed to give large speedups for primitives like sum/avg by becoming more memory- and SIMD-efficient, addressing the memory wall.
- Public benchmark suites are linked (master-benchmark, db-benchmark, TSBS), and readers are encouraged to run them themselves.
- Some want direct comparisons against other k runtimes (including closed ones like Shakti). Current results are seen as promising but incomplete.
Website, “Vibecoding,” and AI Involvement
- The landing page is widely described as “vibecoded”: cryptic copy, marketing phrases like “vector as unit of thought,” and minimal explanation for newcomers.
- There is debate whether this is acceptable or a red flag. Some see it as a modern norm; others view it as a signal of uncertain quality.
- The site explicitly states it was designed with Claude; the runtime is claimed to be hand-written.
- The author reports using AI to analyze and optimize generated assembly, but not to write the core runtime.
Licensing and Openness
- L is closed source, which many array-language enthusiasts accept as common in this ecosystem but others reject as a “non-starter.”
- Some prefer existing open-source implementations as safer or more inspectable, while still finding L technically intriguing.