Fun facts

A few things noticed while surveying the repo.

The largest hand-written file is a backend

src/codegen/x86_64/CodeGen.zig is approximately 10.4 MB. That is one Zig file. It is the self-hosted x86_64 backend, with per-instruction lowering for AIR. The runner-up, src/Sema.zig, is "only" ~1.6 MB.

For comparison, src/main.zig (the entire CLI, including all 30+ sub-commands) is about 350 KB.

A self-deprecating linter inside the compiler

src/main.zig declares:

/// Shaming all the locations that inappropriately use an O(N) search algorithm.
/// Please delete this and fix the compilation errors!
pub const @"bad O(N)" = void;

It is referenced from places that need to do linear scans the author considers wrong, so that "fixing every O(N) hot-spot" is a tracked refactor task in the type system itself.

zig ships its own zen

Running zig zen prints a small zen-of-Python-style poem (the message lives in src/main.zig). It is a real subcommand.

A Zen of bundled toolchains

A single zig binary contains:

• libc sources for glibc, musl, mingw, wasi, darwin, freebsd, netbsd.
• libc++, libc++abi, libunwind, libtsan.
• Clang resource headers (lib/include/).
• The Aro C frontend (lib/compiler/aro/).
• The resinator Windows resource compiler (lib/compiler/resinator/).
• A pure-Zig compiler-rt (lib/compiler_rt/).

That is what makes zig cc -target aarch64-linux-musl "just work" on a fresh Linux box without an external cross toolchain.

ZIR is older than the file size suggests

lib/std/zig/AstGen.zig is ~576 KB and lib/std/zig/Zir.zig is ~210 KB. Despite their size, the names have been stable across all stage2 history — the IRs grew opcodes, not concepts.

Two ELF linkers, side by side

src/link/Elf.zig (159 KB) and src/link/Elf2.zig (158 KB) are both checked in. Elf2 is an in-progress rewrite, and the side-by-side state makes the migration visible to anyone reading src/link/.

The encodings table is bigger than most files

src/codegen/x86_64/encodings.zon — the x86_64 instruction encoding table — is ~161 KB. It is generated/maintained by hand and consumed by Encoding.zig and encoder.zig next to it. The corresponding LLVM disassembler comparison helper is src/codegen/x86_64/Disassembler.zig.

The compiler ships its own GDB pretty-printers

tools/stage1_gdb_pretty_printers.py, tools/stage2_gdb_pretty_printers.py, tools/zig_gdb_pretty_printers.py, and tools/std_gdb_pretty_printers.py (plus tools/lldb_pretty_printers.py, the largest at ~58 KB) make debugging the compiler with GDB or LLDB readable.

Self-hosted HTML rendering

tools/docgen.zig and tools/doctest.zig together compile, run, and embed Zig snippets directly into doc/langref.html.in. The resulting langref.html is part of every install (build.zig's langref step), and -Denable-superhtml even validates the output HTML.