Want to write a compiler? Just read these two papers

It emphasizes that when it comes to compilers, it's far better to first look at a small working implementation rather than starting with a grand theoretical book.

Traditional compiler books cover such a wide scope that it's hard to know where to start, and spending a long time on topics like regex-to-state-machine conversion or grammar theory still leaves you far from actually building a working compiler.

Instead, it recommends Jack Crenshaw's Let's Build a Compiler! series.

• It covers a single-pass compiler at the level of Turbo Pascal.
• It uses a simple structure where parsing and code generation are intertwined.
• Only the most basic level of optimization is applied.
• Besides the original Pascal, there's also a C version and a Forth translation, making it easy to experiment with.

However, this approach has one major limitation. Since there's no internal representation like an AST (abstract syntax tree), it lacks the flexibility to handle the program as a tree structure.

So as a next step, it suggests reading A Nanopass Framework for Compiler Education by Sarkar, Waddell, and Dybvig.

• It views a compiler as a series of transformations over an internal representation.
• Rather than bundling transformations into large chunks at once, it breaks them into very small passes.
• There can be dozens to hundreds of passes, and each pass should be as simple as possible.
• The framework lets you specify the input and output of each pass.
• The implementation language is Scheme, and since it's dynamically typed, verification happens at runtime.

The conclusion is clear. First, build a small compiler yourself, and then, if needed, pick up the Dragon Book or other standard references. You might not even need that.