dev-2023-03

New Language Features

• BREAKING CHANGE: Brand New Default context.temp_allocator Implementation
  • New version is a growing arena based approach
    • Old version used an unsafe ring buffer
  • IMPORTANT free_all(context.temp_allocator) must be called to clear the contents of the internal arena (unlike the previous ring buffer)
    • It is recommended that this is done per frame/cycle/etc
• expand_values()
  • Built-in procedure which expands a struct or fixed-length array into multiple values

Foo :: struct {x: f32, y: i32}
f := Foo{1, 2}
a, b := expand_values(f)
assert(a == 1 && b == 2)

• Allow comparisons between empty struct{} and union{}
• Allow for assigning to fields through using in a compound literal

Foo :: struct {
	using x: struct { // named `using` fields
		y: int,
	},
	z: f32
	using _: struct { // anonymous `using` fields
		w: bool,
	},
}

f := Foo{
	y = 123, // equivalent to `x = {y = 123},`
	z = 456,
	w = true,
}

• Verbose error messages by default with optional ANSI colouring too:
  • Previous behaviour is available with -terse-errors
  • Automatic detection of ANSI colours is currently only supported for Windows
  • set ODIN_TERMINAL=ansi to force ANSI colouring
  • Line printing with indication of the error location below
    

Compiler Improvements

• @(fini) to complement its opposite @(init)
• Fix to byval parameters on Darwin SysV ABI which in turn fixes #by_ptr
• Rename to runtime.Type_Info_Parameters for procedures inputs and outputs
• Fix issue that conflicts with constant parapoly procedure literals and @(deferred_*)
• Numerous improvements to error messages

New Packages

Package Improvements

• Introduction of time.tsc_frequency()
• Improvements to core:mem/virtual's Arena
  • On free_all free all memory blocks except for the first one
  • virtual.arena_destroy will free all memory blocks including the first one (assuming Growing or Static)
• Add #optional_allocator_error to make_map
• Numerous improvements to the vendor:directx packages

dev-2023-02

New Language Features

Compiler Improvements

Compile Time Speed Optimizations

• Multithreaded frontend by default
  • Using the new and improved work stealing thread pool system
  • Fix numerous race conditions
  • Replace many instances of MPMC queues with MPSC queues
  • Replace many instances of BlockingMutex with RwMutex
  • Replace many instances of RecursiveMutex with BlockingMutex
• Multithread backend when the experimental -use-separate-modules is used
  • -use-separate-modules turns each package into a separate translation unit (object file) and then links them all together. This allows LLVM be multithreaded at the translation unit stage (LLVM Module).
  • -use-separate-modules will not produce as good code compared to without it because of the multiple translation units.
• Improvements to the hash table related structures (StringMap, StringSet, PtrMap, PtrSet)
• Big improvement to compile times due to the above improvements

General fixes

• Improved type inference for ternary if expressions
• Remove auto_cast from procedure parameters
• Fix bug with built-in matrix transpose caused by misalignment of the value
• Fix valgrind assembly generation
• Allow when statements within foreign blocks at the file scope

New Packages

Package Improvements

• core:mem/virtual Arena_Temp improvements
  • arena_temp_ignore allowing for ignoring of a temporary section if necessary
• Numerous bug fixes

dev-2023-01

New Language Features

Compiler Improvements

• Improvements to multiple return values ABI for Odin calling conventions
  • #2208
• Improved internal thread pool implementation (ready for next month's release)
  • Futex usage too
• Numerous bug and typo fixes

New Packages

• core:text/match

Package Improvements

• Make sync calls contextless where possible
• Add different variants for sync.once_do
• Make os.get_last_error contextless
• Numerous fixes to core:fmt

dev-2022-12

New Language Features

• New map: High performance, cache-friendly, open-addressed Robin Hood hashing hash map data structure with various optimizations for Odin
  • PR #2181
  • Smaller internal data structure than previously (4 pointers down from 7 pointers)
  • SOA based memory layout
  • Up to 4x-6x faster insertion and lookup on average
  • Entries are stored in non-contiguous cell-layout which means no element straddles across a cache line
  • Only one allocation per map rather than two (previously hashes and entries)
    • Only requires alloc_non_zeroed and free internally
  • Robin Hood hashing
  • Smaller header information (runtime.Map_Info) for dynamic calls.
  • Allows for calling delete_key whilst iterating across the map

Compiler Improvements

• Add debug symbols for global constants for integers, booleans, enums, runes, and pointers
  • Variables are namedspaced with pkg::name, or name if built-in (or the initial package for convenience)
• Improvements to debug information generation
• -default-to-nil-allocator also enabled -no-dynamic-literals
• Improvements to intrinsics.ptr_sub code generation
• Numerous bug fixes

New Packages

• vendor:cgltf

Package Improvements

• Add Allocator_Mode.Alloc_Non_Zeroed
• DirectX packages to use bit_set for flags where possible instead of just an enum
• Implement numerous core:math procedures in native Odin
• Add math.pow10
• Add strings.write_(f16|f32|f64)
• Add user_data: rawptr to filepath.Walk_Proc

dev-2022-11

New Language Features

-

Compiler Improvements

• Make raw_data an intrinsic rather than a @(builtin) runtime procedure
• Allow transmute to be constant for integers of the same internal endianness
• Throw type checker error when scalar cast to non-square matrix
• Fix #defined(I)
• Build script: Detect which and complain if not found
• Add early LLVM > 14 detection, as LLVM 15 and above are not (yet) supported
• Remove previously deprecated -opt flag
• Improve SysV ABI LLVM IR generation for development purposes
• Use direct parameter value in lb_find_ident when possible
• Optimize #caller_location and #location to use read only data section where possible
• Ad-hoc pass source code location directly by pointer without stack copy
• Clarify odin help for -define
• Add -minimum-os-version for Darwin targets, e.g. -minimum-os-version:12.0.0
• Add -target-features flag to force extra LLVM options
• Add safety check for #2161

New Packages

-

Package Improvements

• Update Darwin release map for core:sys/info
• Unify Static_Arena and Growing_Arena into Arena
• Many improvements to core:mem/virtual in general
• Add and correct various Windows, Darwin, glfw, objC and other bindings
• Allow for N = -1 in wstring_to_utf8
• Add core:math/rand.choice
• Heavily improve time handling on Windows for time.now() and os.File_Info

dev-2022-10

New Language Features

• cap(Enum), equivalent to max(Enum)-min(Enum)+1
• ODIN_BUILD_PROJECT_NAME constant
• //+build-project-name build directive
• intrinsics.type_convert_variants_to_pointers
• Add helgrind and valgrind support

Compiler Improvements

• Greatly improve error messages
• Preparations for Compiler Explorer support
• Use uint instead of int to improve bounds checking code generation
• Replace #optional_second with #optional_allocator_error
• Remove extra pointer indirection
• Make intrinsics.count_{ones,zeros,trailing_zeros,leading_zeros} work at compile time
• Improve map
• Improve parapoly support for ^T to [^]$V and vice versa
• Various bugfixes
• Simplify win32 resource file linking
• Add Windows 32-bit build system error
• Add help command, so you can now use odin help build in addition to odin build -help
• Improve macOS version detection for odin report

New Packages

• Add vendor:zlib
• Begin work on core:debug/pe to parse PE files
• Add core:sys/info to query CPU, GPU, RAM, etc.

Package Improvements

• Add system:legacy_stdio_definitions.lib to core:c
• Allow reading/writing files larger than max(i32) on Darwin
• Allow skipping unused fields in core:encoding/json
• Add parsing of +/-Inf and NaN to core:strconv
• Add additional win32 imports and constants
• Fix typo in map_insert
• Fix libc.aligned_alloc
• Add slice.enumerated_array
• Add serial comms support to core:sys/windows
• Add complex support to core:libc
• Correct json unmarshal of maps
• Fix Darwin libs for vendor:stb/image
• Added support for ID3D11InfoQueue to vendor:directx
• Improve core:mem/virtual handling of out of memory on Windows
• Simplify core:io by removing different unnecessary types and calls
• Add the builtin procedures abs, clamp, min, max to core:math as aliases

dev-2022-09

New Language Features

• #soa pointer type to ad with refactory to #soa data types
  • e.g. #soa ^#soa[]T

New Compiler Features

• Make unreachable() a built-in compiler-level procedure
• Allow for foo() or_else unreachable() or other diverging procedures

Compiler Improvements

• #load can now optionally import the loaded file data as a different type than []u8
  • #load(path)
  • #load(path, []T) (where T is a type that can be trivially loaded from memory (i.e. no pointers))
  • #load(path, string) (shorthand for string(#load(path)))
• Deprecate #load_or(path, default) in favour of #load(path) or_else default
• Allow chaining of #load(path1) or_else #load(path2)
• Pass size information to allocator to delete, when possible, to improve tracking information
• Improved detection of rc.exe
• Improvements to debug symbols for range intervals
• Fix Objective-C related x->y() calls in Odin
• Prepare for future LLVM versions
• Various bugfixes

New Packages

• vendor:commonmark
  • Markdown Variant: https://commonmark.org/

Package Improvements

• core:image/tga can now load as well
• Add math.divmod and math.floor_divmod
• Pretty marshal output for core:encoding/json
• Add ID3D11Debug to vendor:directx/d3d11 package
• Add wgl, raw input, and additional Windows imports
• Add more core:container/queue helpers and peek_front and peek_back
• Add reflect.get_union_variant
• Add slice.sort_with_indices family of procedures
• Improvements to sync.Futex support on windows
• Add core:sys/valgrind package for valgrind, memcheck, and callgrind
• Various bugfixes

dev-2022-08

New Language Features

None

New Compiler Features

• #by_ptr procedure attribute to enforce a parameter to be passed by pointer internally

Compiler Improvements

• Improved build scripts for *nix systems
• Improved error message when doing ^x instead of &x for addressable variables
• Improved debug symbol generation
• Fix and improve x->y() behaviour to minimize duplicate evaluation
• Disallow proc do stmt and only allow proc() { stmt }
  • This removed an ambiguity with parsing
• Fix minor issue with a lack of a trailing comma in parsing structs without -vet
• Numerous bug fixes

New Packages

None

Package Improvements

• Improvements to strings.Buffer related procedures
• Update vendor:raylib to support macos-arm64
• Minor improvements to mathematical procedures
• Minor clean up of the WASI API
• Simplify logic of append of zero sized elements

dev-2022-07

New Language Features

• Add builtin shrink for dynamic maps and arrays

New Compiler Features

Compiler Improvements

• Fix odin test runner.
• Use memmove if possible when emitting store over a certain size

New Packages

• core:slice/heap - A max heap implementation

Package Improvements

• Tighter allocation of Arena allocator
• Remove simd_rem; Disallow simd_dev for integers
• Remove strings dependency from core:sys/windows
• Additional Win32 bindings
• Add new verbs and qualifiers to core:fmt
• Support utf-16 printing with []u16 and [^]u16
• Add strings.prefix_length and slice.prefix_length
• Add runtime.dll_forward_reason for Windows

dev-2022-06

New Language Features

• Generic #simd type and intrinsics

  • Supports array programming semantics (operators)
  • New generic intrinsics
  • Platform specific intrinsics for the x86 family:
    • sse, sse2, sse41, sse42, ssse3
    • cmpxchg16b, fxsr, lzcnt, pclmulqdq, popcnt, sha
  • PR #1807

• Merge functionality of #maybe with standard union functionality

  • Maybe :: union($T: typeid) {T} No need for #maybe any more

• Deprecation of a..b range syntax to prefer a..=b

New Compiler Features

• New intrinsics:

  • non_temporal_load
  • non_temporal_store
  • fused_mul_add

• Generic #simd intrinsics:

  • simd_add
  • simd_sub
  • simd_mul
  • simd_div
  • simd_shl
  • simd_shr
  • simd_shl_masked
  • simd_shr_masked
  • simd_add_sat
  • simd_sub_sat
  • simd_and
  • simd_or
  • simd_xor
  • simd_and_not
  • simd_neg
  • simd_abs
  • simd_min
  • simd_max
  • simd_clamp
  • simd_lanes_eq
  • simd_lanes_ne
  • simd_lanes_lt
  • simd_lanes_le
  • simd_lanes_gt
  • simd_lanes_ge
  • simd_extract
  • simd_replace
  • simd_reduce_add_ordered
  • simd_reduce_mul_ordered
  • simd_reduce_min
  • simd_reduce_max
  • simd_reduce_and
  • simd_reduce_or
  • simd_reduce_xor
  • simd_shuffle
  • simd_select
  • simd_ceil
  • simd_floor
  • simd_trunc
  • simd_nearest
  • simd_to_bits
  • simd_lanes_reverse
  • simd_lanes_rotate_left
  • simd_lanes_rotate_right

• Platform specific intrinsics:

  • x86_cpuid
  • x86_xgetbv

• @(priority_index=<int>) for foreign import

  • Force certain things to be linked before others
  • Required for the deterministic link order of foreign imports

Compiler Improvements

• Allow transmute on constant expressions
• Heavily improved support for js_wasm32 target
  • Basic DOM UI procedures
  • Event system including listeners
  • Page Allocator
• Compiler flag error message improvements
• Improved ternary-if type inference
• Improved -vet shadowing with ternary-if expressions
• Allow import _ "foo" to allow for @(init) procedures without producing an import name
• Correct @(require_results) on parapoly procedures

New Packages

• core:simd
• core:simd/x86
• core:encoding/endian
• vendor:ggpo
• vendor:openexr

Package Improvements

• Unify raw_data in core:mem with core:runtime
  • Make raw_data return [^]T types
• core:image generic interface improvements
  • image.which
  • Generic loader
• Add #optional_ok to dynlib.symbol_address
• Add mem.DEFAULT_PAGE_SIZE
• Add rand.init_as_system to allow for system-level based random number generation
• Numerous minor fixes to core: packages
• Interface cleans up to numerous packages keeping platform specific code private