slang

A stack-based bytecode Vm written in C. It's a dynamically typed (currently), garbage-collected and object-oriented programming language with a syntax that is inspired by C#, JavaScript and Python - just less verbose.

Note

Slang is still in development. The language is not stable and the syntax is subject to change.

Why?

The main goal of this language is to be able to solve problems with succinct code. It's aimed at developers who like competitive programming or just want to write code that is concise. It's also a great language to learn how programming languages work under the hood, because it's relatively simple.

It's a less-verbose language that relies on the JavaScript syntax, but also incorporates some of the features of C# and Python. It also tries to approach some things differently than JavaScript - The major "thing" being the fact that you can use reference types to index into objects and the presence of a tuple type, which features a reproducible hash. This is particularly nice if you're into dynamic-programming. You can, for example, easily cache stuff:

Tip

Currently, there is no Wiki or "Getting Started" or the like. The best way to see what it can do is to checkout sample.sl in the repository root, or the tests in the test directory.

References

• Based on the book Crafting Interpreters. In contrast to the book, this Vm is not focused on space-efficiency, but more on ease of implementation and speed.
• https://github.com/kuroko-lang/kuroko
• https://github.com/Janko-dev/yabil/
• https://luajit.org/luajit.html

---

Roadmap for Version 1.0

Compiler rebuild

• Fix "unused var" warnings for late-bound globals
• Resolve natives and imports of cached modules in the compiler maybe? We could easily just look the stuff up and emit a value for it (e.g. constant, or some new OP like OP_PRECOMPILED) instead of looking it up in the VM.
• Test if assignment to patterns works, because declarations do.
• After testing: Refactor module imports without Module name (imports using "from").
• After testing: Reorder the operands for OP_GET_PROPERTY(and set too) in the VM from [recv][value] (top) to [value][recv] (top). This would eliminate the need for the "prelude" functions for assignment.
• After testing: Reorder the operands for OP_GET_SUBSCRIPT(and set too) in the VM from [recv][idx][value] (top) to [value][idx][recv] (top). This would eliminate the need for the "prelude" functions for assignment.
• Add resolver warn for vars that could be constant.
• Make parser marking possible and remove disabling the GC during parsing.
• Turn globals / natives into an array. Because we can resolve it now at compile time. This would also allow for constant time global variable lookup
• Remove run-old completely
• Make REPL use the new compiler
• Move path resolution from the compiler to the resolver.
• Remove in_global_scope and current_scope from the compiler - that should be handled via the resolver. Currently needed for destructuring, but I think that should be possible without it.

Features

• 🐛 Fix tuple hashing for tuples containing negative values (Encountered this in AOC '24 day 22 when hasing tuples containing negative Ints). UPDATE: Found it. It's the hashing of integers thats the issue.
• Add not for is and in: e.g. x not in y and x is not Int
• Allow Tuple.inside = fn(this) -> (this[0]>=0 and this[0]<ROWS) and (this[1]>=0 and this[1]<COLS)
• Implement for ... in ...; loops (Implement Iterators)
• Add nillish coalescing operator ?? e.g. let x = [1] <newline> let v = x[1] ?? 0
• Implement Seq.mapat(Int, Fn) -> Seq. (Map only the element at the given index but return the whole sequence)
• Implement Seq.cull(Value|Fn) -> Seq. (Remove all elements that are equal to the argument or satisfy the predicate)
• Implement Seq.zip(Seq, Seq) -> Seq. (Zip two sequences into one sequence of tuples)
• Add a variant of log (Maybe tap/info/dump/peek?) which accepts a single argument and also, return it. That'd be awesome: const x = a + b + c + tap(d) + e
• Add more error classes to std. Add a native Error base class, from which managed-code errors inherit. Check vm_inherits(error.type, vm.error_class) in handle_runtime_error and - if true - use error.type->name as the prefix instead of Uncaught error.
• Implement Str.ints() -> Seq. (Split a string into a sequence of integers (also negative ones))
• Implement Str.rep(Int) -> Str. (Repeat a string n times)
• Implement Str.chars() -> Seq. (Split a string into a sequence of characters, shorthand for Str.split(""))
• Implement Seq.sum() -> Num, Tuple.sum() -> Num. (Sum all elements. Requires some sort of __add)
• Implement Seq.sort(sort_fn) -> Seq. (Sort a sequence in place. Requires some sort of __lt)
• Implement Seq.min(type: Type) -> Value. (Get the minimum value of a sequence, should use SP_METHOD_LT of the type)
• Implement Set class (Part of the std module - not a native type)
  • Implement Set.add(Obj) -> Nil.
  • Implement Set.del(Obj) -> Nil.
  • Implement Seq(Set) constructor Set.to_seq() -> Seq
• Implement native Math module.
  • Implement Math.abs(Num) -> Num.
  • Implement Math.ceil(Num) -> Int.
  • Implement Math.floor(Num) -> Int.
• Implement Gc module
  • Implement Gc collect() -> Nil.
  • Implement Gc stats() -> Obj.
• 🐛 Fix Str.split(Str) for strings which have multiple submatches per match, e.g. " 0 w e r".split(" ") segfaults. (Encountered in AOC '24 day 25)
• map and some other array functions should also accept arity=0 functions, not only arity=1 and arity=2.
• Implement Test class / module with Assert.that(expected, Is.equal_to(actual))
• Add destructuring to module imports.
• Add const (See Challenge 22.3)

Improvements

• Verify that value-array sorting should use SP_METHOD_LT (currently used), and not SP_METHOD_LTEQ instead?
• Refactor comparators to FLOAT_COMPARATOR and INT_COMPARATOR respectively - just like in native_type_str.c (see STR_COMPARATOR).
• Rename Str.reps to Str.rep for consistency.
• String refactoring:
  • Move to UTF-16 / unicode codepoints.
  • Refactor/remove Str.ascii()
  • Refactor/remove Str.ascii_at()
  • Refactor/remove static Str.from_ascii(Int) -> Str
  • Refactor str comparison functions
• Improve destructuring assignment:
  • If you destructure a Seq into a Tuple, the rest of the elements should be of the type of the lhs. E.g. let (a, ...b) = [1, 2, 3] where a is an Int and b is a Tuple. Currently, b is a Seq.
• Make managed-code callables accept less arguments than it has parameters. There is an inconsistency, since native-callables allow this. Should be easy, just pass nil to the missing arguments.
• Add a guard in compiler.c -> number() to check for overflow.
• Remove OP_PRINT completely in favor of native log function
• Add a mimalloc segfault handler.
• Implement Float.nan and Float.inf constants (Would require static fields).
• Add test for Math.pow(Int, Int) -> Int (Or maybe move to Int?)
• Add test for Math.xor(Int, Int) -> Int (Or maybe move to Int?)
• Add test for Math.shl(Int, Int) -> Int (Or maybe move to Int?)
• Add test for Math.shr(Int, Int) -> Int (Or maybe move to Int?)
• Add test for Math.bor(Int, Int) -> Int (Or maybe move to Int?)
• Add test for Math.band(Int, Int) -> Int (Or maybe move to Int?)
• Add test for Str.ascii() -> Seq which includes special characters.
• Add test for Str.ascii_at(Int) -> Int which includes special characters.
• Add test for static Str.from_ascii(Int) -> Str which includes special characters.
• Add test for Str.SP_METHOD_LT(Str) -> Bool, also for sorting.
• Add test for Str.SP_METHOD_LTEQ(Str) -> Bool
• Add test for Str.SP_METHOD_GT(Str) -> Bool
• Add test for Str.SP_METHOD_GTEQ(Str) -> Bool
• Maybe check for NULL functions in vm_exec_callable instead of before calling it - would add some overhead though.
• Collect compile-time errors as strings and print them either directly when they occur (when compiling an entry point), or, as part of a failed import error message (runtime error). Currently, the compiler pipeline directly prints to stderr, which is a little confusing, as e.g. parser errors will be printed before the runtime error message for a failed import. see module-import-wiht-compile-error.spec.sl for an example.
• Currently, i++ behaves more like ++i (Which we don't support). Fix it.
• Add Tuple.order test.
• Remove "" + value.to_str() throughout the codebase, "" + value should now work.
• Align DO_OP_IN with MAKE_OP(), there's some unnecessary push/pop-int in there.
• Call to_str implicitly when adding a string to a non-string. Only if the left side is a string.
• Fix VM finishing a program in error state not exiting with EXIT_FAILURE. (Probably, the flags are reset in vm_free or reset_stack or something).
• Make compiler errors recoverable. Just let it sync, then continue - instead of aborting.
• Use obj_get_prop in obj_has instead of just checking the hashtable for a value. Otherwise, they behave differently - which sucks.
• Make sure managed code classes do not override internal classes. Not necessary, since we don't update the Vms internalized natives.
• Add error and other contextual keywords to a list of reserved words (Maybe including all natives). Check them when declaring anything. Not necessary, since other ctx keywords have their own TOKEN type. As for errors, we'll just allow redeclaration, even inside a catch-block. The reason being that the error-var gets injected into a scope surrounding the whole try/catch statement, not just in the catch-block.
• Remove string_to_double and use number from the compiler instead.
• Generalized calls. This is optional, but could enhance the language.
  • Move hash_value to types (uint64_t ObjClass.__hash(value: Value))
  • Move values_equal to types (bool ObjClass.__equals(a: Value, b: Value)) DUPE, see "Optimizations".
  • Move to_str to types (Obj ObjClass.__to_str(value: Value))
• Remove the "Class" prefix in VALUE_STR_CLASS to just return the class name. This is a bit more consistent with the rest of the code.
• Restructure test: compiler, vm (types, modules)
• (When Gc is implemented) Add some test cases where we call Gc.collect() to ensure that relevant objects are not collected.
• Add tests for Fn.bind(Obj)
• Add Tests with tabs in source code. Especially to test uncaught runtime error reporting.
• Add tests for OP_MODULO
• Module Caching refactor: When trying to load a module from cache, we need to check for the absolute file path of the module instead of only the name. When importing a module with destructuring, we use the absolute path as the module name - when you load the same module with a relative path and a module name, it will currently be cached with the key "module name" and therefore not be found in the cache.

Codebase Refactors

• Align doc-comment style to use /** everywhere.
• Align error messages. Some use ' around names, or type names, some don't.
• Make a NATIVE_METHOD_RUNTIME_ERROR(class_name, method_name) macro, which throws a runtime error with a nice prefix and always returns NIL_VAL. Use this in all NATIVE_METHOD_IMPL functions.
• Move utility functions from vm.c to respective files.

Optimizations

• Exclude the natives from the gc
• Add OP_CALL_SP and use a lookup table similar to finalize_new_class to call the method. Would make sense now that we have so many SP methods. We could also remove all arithmetic and comparison operations.
• The features I want to implement for this language kinda require a more complex compilation process. Specifically, ditching the single-pass approach for a two-pass compiler and an actual AST.
• Maybe delete NATIVE_CHECK_RECEIVER as most of the time the method is called from the receiver.
• Use memcpy for concat and such (See Seq(Tuple) ctor for an example). Check for for-loops in the native methods. Need to test if this copies values by reference or by value. Needs a decision on on how we want to handle that.
• Implement a string builder and use it everywhere. This is a must-have.
• Inline vm_push(), peek() and vm_pop() in the Vm.
• Maybe add a fast hashtable-set function (key must be ObjString).
• Move ip into a register. This is a must-have. (See Challenge 24.1)
• Store strings as flexible array members (See Challenge 19.1)
• Split globals into a Hastable global_names and a ValueArray global_values. This would allow for constant time global variable lookup. (See Challenge 21.2) https://github.com/munificent/craftinginterpreters/blob/master/note/answers/chapter21_global.md
• Only necessary closures. Evaluate this, maybe it's not worth it. (See Challenge 25.1)
• Closing over the loop variable. (See Challenge 25.2)
• Single-op unaries. Not fully-fledged constant folding, but a good start. (See Challenge 15.4)
• Move values_equal to types (bool ObjClass.__equals(a: Value, b: Value)) - this would make values_equal obsolete.

---

Roadmap for Version 2.0

Features

• String interpolation. C#-style $"Hello {name}" (See Challenge 16.1)
• Implement match Statement. (See Challenge 23.1, on how to impl switch, that's a start.)
• Implement @memoize decorator. Would put args into a Tuple and use that as a key in a Obj.
• Implement native Json module.
  • Implement Json.parse(Str) -> Obj.
  • Implement Json.stringify(Value) -> Str.
  • Implement Json.stringify(Value, Int) -> Str. (Indentation)

---

Roadmap for Version 3.0

Features

Type-checking

• Implement syntax for type annotations. E.g. let x: Int = 1. We'll check as much as possible at compile-time. Locals are already cared for, because they live on the stack and are not referenced by a name. All locals are resolved during compile time. The only exception being locals that are not initialized with a value. That should be allowed, but the type must be declared. E.g. let x: Int. This is a bit more flexible than C# and a bit less flexible than TypeScript. We'll see how it goes.
• Implement a solution for functions. We should introduce a signature field in fn objects. The sig struct should probably consist of: args: ObjClass** to store the types of the arguments, ret: ObjClass* to store the return type of the function, argc: int to store the number of arguments.
• Introduce type-modifiers. ? for nullable, and maybe something for exact type match. This should be done as a flag on the Value struct. Maybe make a Type struct that holds the ObjClass* and the flags. This could then also be used for function signatures.
• Actually add typechecking and remove all NATIVE_CHECK_ARG things.

LSP Server

• Implement an LSP server. Rough idea: Compile the source and generate the bytecode. Maybe with some heuristics. Like, don't recompile imported modules. We should be able to map a line + column position to a instruction, since we store the relevant information in a Token for each instruction. We should also be able to retrieve a list of possible strings to write next in the sourcecode - e.g. globals, methods etc.

---

Further Ideas

• Rebuild to a register-based Vm https://www.lua.org/doc/jucs05.pdf
• Constant folding directly in the compiler
• Implement a JIT compiler