The Magic of Partial Evaluation
aray07
105 points
28 comments
November 18, 2026
Related Discussions
Found 5 related stories in 137.6ms across 13,334 title embeddings via pgvector HNSW
- Abstracting Effects with Continuations crowdhailer · 55 pts · July 16, 2026 · 47% similar
- Zig's Incremental Compilation Internals garyhtou · 218 pts · July 28, 2026 · 46% similar
- Evals Are the Next Bottleneck ssatia · 14 pts · August 20, 2026 · 46% similar
- I accidentally turned LLM memory into program analysis matt_d · 89 pts · August 28, 2026 · 44% similar
- Jev, Prolog, Pi, and the dream of probabilistic logic programming schmuhblaster · 16 pts · September 21, 2026 · 44% similar
Discussion Highlights (10 comments)
measurablefunc
Combining this w/ a JIT gets you truffle/graal.
owlstuffing
This type of optimization is also known as compile-time constant expression folding. The problem with expanding it to include function calls is that it compromises binary compatibility. A code change on a function that results in a different range or has different side effects, will not be realized unless the caller is recompiled.
apignotti
We implemented a similar concept in our C++ -> JavaScript/WebAssembly compiler (Cheerp) several years ago. It was a nice challenge. https://labs.leaningtech.com/blog/reducing-webassembly-size-...
infogulch
Doesn't Zig have a pretty nice implementation of compile-time evaluation built in?
artrax
We build this as a c++ library, do to query-compilation for our stream processing engine. https://github.com/nebulastream/nautilus It allows us to implement operators or functions without having to write them in terms of LLVM IR DSL (or similar). Rather, you write your functions over a wrapper type (and returning one), for which there is an overload of anything which you can overload in c++. You can instantiate a wrapper from some c++ value, in which case the wrapper will just forward everything to this value, which was nice for debugging in the preagentic era. But, you can also instantiate such a wrapper as a tracer, and the result of the functions evaluation will be a trace. For controlflow, we record whenever a wrapper type is evaluated as a bool, first return true, and when we reach the end of one branch, throw an exception to get back to the beginning, and start all over again, but evaluate it to false. The resulting trace we can turn into LLVM IR and compile, or run for example in wasm
johntortugo
This seminar explains how GraalVM Truffle uses partial evaluation: https://youtu.be/pksRrON5XfU
airtnp
I keep thinking the partial evaluator and Futamura projection application on LLM context. Skill => Generic Program LLM skill optimizer => Partial evaluator Context => Env/Input Agent => Interpreter First projection: mix(interpreter, program) => specialized skill AOT compiling => offline skill rewriting JIT compiling => runtime specializing, defer loading Second projection: mix(mix, interpreter) => Agent/Model specific skill optimizer Third projection: mix(mix, mix) => self-improvement skill optimizer Whatever maybe I'm just thinking LLM as magic partial evaluator for any tasks
iFreilicht
Very interesting concept! It reminds me of Hazel ( https://hazel.org/ ) which takes the idea further and into the editor; any program written in it can be partially evaluated, even if variables or even part of the syntax are missing. Missing parts are called "holes", and those are preserved during evaluation so that something like abs( + 2 * 3 - 10) becomes abs( - 4) which is great fun. My first attempts using this showed me that if you're not used to this, it actually gets pretty confusing quite quickly, but I guess it can be learned.
TuringTest
> For me, the most beautiful pieces of software are those that reveal something about the nature of programming languages. There are three moments that make the functional programmer evolve their understanding of how a programming language really works as a symbolic representation of human thought, beyond the metaphor taught in school of a computer program as a grinder that crunches input data and produces results. The first one is when you really understand recursion and how a function can be defined in terms of itself (with varying parameters for each call). The second is when you discover Prolog variable unification or Haskell lazy evaluation and how you can represent infinite data structures without having the program hang in an infinite loop. Third one is when you learn that the behavior of any imperative language (down to the working of the microprocessor data paths) can be fully represented in a pure functional rewrite language (without destructive assignment / cell state changes). Those insights make one realize that all programming paradigms have the nature of symbols processing in common and there's no "true description" of a program's runtime execution - the imperative, functional and logic views of the running program are all equivalent representations.
p1necone
All the examples in here look pretty clean, but doesn't partial application like this in practice on "average" codebases result in massive binary size explosion?