Changed

• runtime type checking errors are on stderr instead of stdout
• runtime exceptions are on stderr instead of stdout

Changed

• VM error outputs are on stderr instead of stdout

Added

• new TAIL_CALL_SELF instruction to take care of tail calls in functions: jumps to address 0 in the current page, and reset the scope

Changed

• error outputs are on stderr instead of stdout

Breaking change

• in macros, len, empty?, head, tail, @ have been renamed to $len, $empty?, $head, $tail and $at. Those versions only work inside macros too, inside of having a weird dichotomy where they sometimes got applied and sometimes not

Added

• apply function: (apply func [args...]), to call a function with a set of arguments stored in a list. Works with functions, closures and builtins
• +, -, *, / and many other operators can now be passed around, like builtins. This now works: (list:reduce [1 2 3] +), where before we would get a compile time error about a "freestanding operator '+'"
• builtin__slice builtin, for strings and lists: (builtin__slice data start end [step=1]) ; this is an experimentation and may be removed in future versions
• arguments of builtin macros are properly type-checked and will now raise runtime errors if the type is incorrect
• -fno-cache cli option to disable the creation of the bytecode cache folder __arkscript__
• in the CLI, file can be - to read code from stdin

Changed

• when using the cli flag -fdump-ir, the IR is dumped in the cache folder

Breaking changes

• assert is no longer an instruction but a builtin
• when comparing values of different types using <, >, <=, >= and =, the result will always be false (it used to rely on the type index)

Added

• added BREAKPOINT instruction
• breakpoints can be placed using (breakpoint) and (breakpoint condition)
• added a debugger that can be triggered on error or on breakpoint by passing -fdebugger to the CLI (see the docs for the debugger)
• diagnostics can now be generated when using State.doString, using Diagnostics::generateWithCode (as the original code must be passed to the diagnostics generator)
• empty? can now take a dict, and returns true if it has no key/value pairs
• len can now work on dictionaries, counting the number of keys

Changed

• changed the runpath of arkscript to look for libArkReactor under its (arkscript's) directory, {arkscript}/bin, {arkscript}/lib, and {arkscript}/../lib
• and and or require valid expressions, so (or 1 (mut x 3)) is no longer valid code, as (mut x 3) doesn't return a value
• (not (dict "a" 2)) now returns false, as not checks if the dict is empty

Added

• the repl prints the output of the last expression it ran
• new super instructions: MUL_BY, MUL_BY_INDEX, MUL_SET_VAL that can do multiplications (and optional storing in vars) in place
• new super instruction: FUSED_MATH, which can fuse 2 to 3 math operations in one go (ADD, SUB, MUL, DIV)
• new LOAD_SYMBOL instruction that avoids creating a reference

Fixed

• the REPL doesn't color import in two colors (red for imp__t and blue for ___or_), it keeps the first color that matched (red for import here)
• page numbers are correctly counted when using the bytecode reader with '--only-names', instead of displaying 0 every time

Changed

• quotes are added around strings in type errors
• disassemble can show a file bytecode
• empty? now accepts nil and returns true for this value
• the REPL adds (repl:history) and (repl:save filename) as builtins
• the REPL attempts to load a file from ARKSCRIPT_REPL_STARTUP environment variable, to preload code
• rename LOAD_SYMBOL and LOAD_SYMBOL_BY_INDEX to LOAD_FAST and LOAD_FAST_BY_INDEX to emphasize they load refs

Fixed

• the formatter was breaking functions' arguments list containing argument attributes on multiple lines for no reason
• the formatter was formatting begin nodes inside conditions badly, putting the { on the same line as the condition, making it hard to know if the condition had then and else nodes or a single multi nodes then node

Changed

• long function calls are split on multiple lines

Breaking changes

• Function arguments are now immutable by default and an argument attribute mut must be added: (fun (a b c) (set b 5)) -> (fun (a (mut b) c) (set b 5))

Deprecated

• dict:contains, use dict:contains?
• math:even, use math:even?
• math:odd, use math:odd?

Added

• new builtin disassemble to print the bytecode of a function
• new builtin io:readFileLines to read lines from a file as a list of strings

Changed

• the formatter properly formats dictionaries (key-value pairs on their own line, always)
• renamed dict:contains to dict:contains? so that all functions returning booleans have ? suffix ; added temporary alias dict:contains
• renamed math:even to math:even?, and math:odd to math:odd?
• string:removeAt can work with negative indexes

Changes for you, as a user

• macros are defined using (macro name value) and (macro name (foo bar ...args) body) instead of !{name value} and !{name (foo bar ...args) body}
• quote keyword got removed. You can emulate it with a macro: (macro quote (value) (fun () value)) (it wasn't really a Lisp quote in the first place...)
• while loops have their own scope
• imports changed a lot: from (import "path/to/file.ark") to java like imports (the root is deduced from the main script emplacement): (import path.to.file :foo :bar)
  • for more details, see the documentation
• you can format your code using the cli: arkscript -f myfile.ark
• new dict datatype: https://arkscript-lang.dev/docs/std/dict/

For more information, check the language documentation!

Technical changelog

Added

• added program name under builtin__sys:programName

Changed

• execution contexts can be reused for async calls if they are not active, to avoid constantly requesting memory and creating (heavy) contexts
  • if there is more than 5 contexts, the 6th one will be destroyed once it completes
• execution contexts are now marked as free to be reused (or deleted) once a value has been computed, without waiting for a call to await
• captures are not renamed anymore by the NameResolutionPass (which used to fully qualify captured names when possible, which isn't desirable: when you capture &foo, you expect to be able to use .foo not .module:foo)
• when loading a module, its mappings are loaded in the current scope instead of the global scope
• argument order in the CLI changed: the file to run (and its optional script arguments) are now last, to be more consistent with all the other existing tooling (Python, Docker...)