商品簡介
It is 2026, and a language model can generate a lexer, a type checker, and a working compiler in seconds. So why learn to build them by hand? Because a generated artifact is an assertion - and an assertion you cannot evaluate is not knowledge, it is a hope with good syntax. The value was never in the typing. This book carries one language - Lark, the Lambda Affine Resource Kernel - from a real instruction set to a machine-checked proof that its meaning is sound, treating the hardware, the theory, and the proof as a single argument. You begin at the silicon and build upward through lexer, parser, type system, interpreter, and optimizer, to a code generator that runs Lark as native code on an affordable Raspberry Pi Pico 2/2W. Theory is earned, not assumed: Hindley-Milner inference, affine ownership, and traits, each introduced when the implementation needs it. And the book teaches the discipline that code-generating tools demand - an interpreter that defines what a program means, a differential-testing harness that says which back end is wrong, a type-safety theorem that bounds what any transformation may do. And then it turns that machinery on itself: the finished compiler is fuzzed, the tests are checked with planted bugs, and even the proof kernel is attacked until it gives up its own flaws - because a guarantee is worth exactly as much as the adversary it has survived. For the technically literate programmer who has always meant to understand what is under their language and never found the entry point. It assumes familiarity with Python or C and basic data structures - but not a line of compiler theory.
