A Python-based compiler for A7, a statically-typed systems programming language. A7 combines the simplicity of C-style syntax with modern features like generics, type inference, and compiler-managed reference operations.

The compiler features a complete pipeline: tokenizer, parser, semantic analysis, internal safety proof planning, AST preprocessing, and Zig code generation.

A7 draws inspiration from practical systems programming languages that prioritize clarity and programmer productivity:

• JAI by Jonathan Blow - Design philosophy and compile-time features
• Odin by Ginger Bill - Simplicity and explicit memory management

Requirements: Python 3.13+ and uv (recommended package manager). Install Zig 0.16.0 to build and run generated Zig output; CI pins the same version for repeatable artifact checks.

# Install uv (if needed)
curl -LsSf https://astral.sh/uv/install.sh | sh  # Linux/macOS
# or: pip install uv

# Clone and setup
git clone <repository-url>
cd a7-py
uv sync

Run from the repository checkout:

Compile an A7 program to Zig (default backend):

uv run python main.py examples/001_hello.a7
# Output: examples/001_hello.zig

Learn the current language from one commented file:

uv run python main.py examples/037_language_tour.a7

Modes and output formats:

uv run python main.py --mode tokens examples/006_if.a7                     # Tokens only
uv run python main.py --mode ast examples/004_func.a7                      # Tokens + AST
uv run python main.py --mode semantic examples/009_struct.a7               # Through semantic passes
uv run python main.py --mode pipeline examples/014_generics.a7             # Full pipeline, no file write
uv run python main.py --format json examples/014_generics.a7               # Machine-readable JSON
uv run python main.py examples/001_hello.a7 --mode compile --doc-out auto  # Compile + auto docs
uv run python main.py examples/001_hello.a7 --mode compile --doc-out out.md  # Compile + custom docs
uv run python main.py --mode doc examples/001_hello.a7                     # Doc-only run
uv run python main.py --verbose examples/009_struct.a7                     # Full pipeline details

Use the installed console script after uv sync:

uv run a7 examples/001_hello.a7

Exit codes for automation:

0 success, 2 usage, 3 io, 4 tokenize, 5 parse, 6 semantic, 7 codegen, 8 internal

Run tests:

PYTHONPATH=. uv run pytest                         # All tests
PYTHONPATH=. uv run pytest test/test_tokenizer.py  # Specific test file
PYTHONPATH=. uv run pytest -k "generic" -v         # Targeted tests
uv run python scripts/verify_examples_e2e.py       # Compile/build/run + output checks for all examples
uv run python scripts/verify_error_stages.py       # Error-stage audit across modes and formats
uv run python scripts/build_examples.py --profile debug --backend zig --clean
uv run python scripts/build_examples.py --profile release --backend zig --clean
./run_all_tests.sh                                 # Full release-oriented local gate

The compiler emits Zig source. scripts/build_examples.py is the native artifact builder used for release smoke checks:

uv run python scripts/build_examples.py --profile debug --backend zig --clean
uv run python scripts/build_examples.py --profile release --backend zig --clean

Artifacts are written under build/debug/ and build/release/:

build/<profile>/zig/src/*.zig
build/<profile>/zig/bin/*

Each built binary is executed and compared with test/fixtures/golden_outputs/*.out.

Build the Python package:

rm -rf dist
uv build

Cleaning dist/ first prevents stale older package artifacts from being mixed with the current release build.

Generate a checksum manifest for release artifacts:

uv run python scripts/generate_release_manifest.py dist --output dist/SHA256SUMS
uv run python scripts/verify_release_manifest.py dist/SHA256SUMS
uv run python scripts/verify_archive_contents.py dist/a7-docs-site.tar.gz --require dist/llms.txt --require dist/llms-full.txt

The installed CLI entrypoint is a7:

uv run a7 --help

Release tags build distributions and attach them to the draft GitHub release. There is no package-registry publishing job in the current workflow.

Source (.a7) → Tokenizer → Parser → Semantic Analysis → Safety Proof Planning → AST Preprocessing → Zig Codegen → Output (.zig)

1. Tokenizer. Lexes source into tokens. Handles single-token generics ($T), nested comments, and number formats.
2. Parser. Uses recursive descent with precedence climbing. Parses all A7 constructs.
3. Semantic Analysis. Runs name resolution, base type checking with inference, control flow validation, and recursion rejection.
4. Safety Proof Planning. Proves or rejects risky operations such as casts, division/modulo, indexing/slicing, reference dereferences, and direct use after del before backend lowering.
5. AST Preprocessing. Runs stdlib resolution, struct init normalization, mutation and usage analysis, type inference, shadowing resolution, function hoisting, and constant folding.
6. Backend Code Generation. Translates approved AST operations to valid Zig source code.

Semantic analysis, AST preprocessing, formatter/reporting AST walks, and backend binary-expression emission use explicit stacks. The parser is recursive descent, and backend statement/non-binary expression generation still uses visitor-style recursive emission in some paths. Current low-recursion coverage validates the supported pipeline at Python recursion limit 100 for representative programs. A7 source recursion, including local function-pointer aliases and higher-order callback trampolines, is rejected during semantic validation; use loops, explicit stacks, or index-based worklists instead.

Use usize for sizes, lengths, capacities, allocation byte counts, and array/slice/string indices. Index and slice-bound variables are required to be usize; non-negative integer literals are accepted for simple indexing. usize maps directly to Zig usize.

Use isize only when the value is a signed pointer-sized offset or a difference between positions. It exists for pointer-adjacent signed math, not as the default signed integer type.

Use fixed-width integers such as i32, i64, u32, or u64 when the data itself has that width or range. Small arithmetic examples can use i32; counters and indexes should usually use usize.

• Types: Primitives, arrays, slices, pointers, generics, raw and aliased function types, inline struct return values
• Declarations: Functions, structs, enums, unions, variables, constants, type aliases
• Control Flow: if/else, while, for loops, for-in, labeled loops with break/continue, match statements, defer
• Function Rules: Direct, mutual, alias-mediated, and callback-trampoline recursion are semantic errors
• Expressions: All operators with proper precedence, fixed-array + for same-shape numeric arrays, casts, if-expressions, struct/array literals, untagged union field literals/access
• Memory: ref parameters use ordinary lvalue arguments, ref struct fields are accessed directly after nil-proofing, and scalar/struct new plus del support defer cleanup. Heap fixed arrays (new [N]T) are rejected until the language model is defined.
• Safety proofing: casts, division/modulo, bounds-sensitive indexing/slicing, reference dereferences, operation-specific backend approvals, and direct use after del are checked by internal facts before Zig emission.
• Imports: Virtual std/io and std/math modules with aliases; simple file-backed alias imports can lower into the same generated Zig file
• Generics: Type parameters ($T), constraints, type sets, generic structs, generic struct literals, and simple top-level generic function calls
• Code Generation: A7 → Zig, with generated std/io print helpers that preserve stdout/stderr on current Zig toolchains
• Standard Library: Registry with io and math modules, backend-specific mappings
• Error Messages: Rich formatting with source context and structured error types

• Test status depends on current branch state. Check with PYTHONPATH=. uv run pytest --tb=no -q.
• Example end-to-end verification is available through uv run python scripts/verify_examples_e2e.py.
• Debug/release artifact verification is available through scripts/build_examples.py.
• Tag releases attach package artifacts to a draft GitHub release.
• Parser covers the implemented language surface, but spec/implementation gaps remain tracked in docs/STATUS.md.
• Zig backend handles the current example suite and most AST node types; unsupported source constructs should continue moving to compiler-side diagnostics.
• This compiler is not a sandbox. Do not compile and execute untrusted A7 source.

• Documentation website: https://code5717.github.io/a7-py/
• Agent/curl.md docs entry point: https://code5717.github.io/a7-py/llms.txt
• Agent docs index: https://code5717.github.io/a7-py/docs/index.md
• Full agent context: https://code5717.github.io/a7-py/llms-full.txt
• docs/SPEC.md - Language specification
• docs/SAFETY_CONTRACT.md - Compiler safety contract and proof/backend-plan invariants
• docs/STATUS.md - Current gaps, priorities, and roadmap
• docs/RELEASE.md - Release/debug build checklist
• docs/SECURITY.md - Security policy and trust boundary
• examples/ - 43 sample programs
• docs/CHANGELOG.md - Change history
• docs/ERROR_ANALYSIS.md - Historical error-analysis snapshot, not current status

cd site
bun install
bun run dev

Work in progress. Contributions welcome!