More than my first attempt at writing a programming language, but the first to be generate code, though it cannot (yet) parse anything.
This was largely written to write down a/my mental model for compilers. Though also a
majority of the time spent so fars was encoding x86_64 instructions from an embedded
DSL. The x86_64 encoder is incomplete, only handling ADD-like instructions.
There is a good amount of hardcoded tests that compare encoded instructions
to a slice of bytes. However this can a be a little error prone, so there is
also some "integration testing" that compares the instruction encoding here
with that of objdump.
Test cases also have an associated string name which is actually the intel
syntax version of the instruction. The name and encodings are written to files
test_xxx.asm and test_xxx.bin respectively. Then the output of
objdump ... test_xxx.bin can be diffed with that of test_xxx.asm
-
- parser
- - basic types
- - tuples
- - abstract data types
- - abstraction
- - application
- - let binding
- - patterns (missing ADTs)
- - types (missing ADTs)
- - multiple lets
- - pattern lets
-
- string interning
-
- interpreter
- - values
- - beta reduction (function calling)
- - pattern matching
- - let bindings
- - multiple lets
- - pattern lets
-
- type check
- - basic types
- - patterns
- - sizing
- - match arm consitency
- - match exhaustiveness
- - polymorphism
- - type inference
-
- pointers?
- - requires some level of polymorphism
-
- optimizer
- - transform local shadowing into mutability
- - automatically demote args to references
-
- x86_64 instruction encoding
- - add-like
- - ret
- - call
- - jmp
- - syscall
-
- codegen
- - everything on the stack
- - use regs
- - basic types prelude
- - non-builtin string
- - debug info
-
- more arches
- - Z80/Gameboy
- - RISC-V
- - arm