Odin Files

New Language Features

• [#7664]: Allow bit_set to be backed by arrays of integers

New Compiler Improvements

Speedups

• [#7720]: LLVM Backend Performance Improvements to Reduce Compile Times Part 1
• [#7729]: LLVM Backend Performance Improvements to Reduce Compile Times Part 2
• [#7747]: Compiler Performance Improvements to Reduce Compile Times Part 3
• [#7685]: Complete Parapoly Rewrite + Race condition fixes
• [#7694]: Make global checking codegen order-independent
• [#7686]: Parse integer literals without allocating per digit and reject non digits in their exponent
• [#7739]: Make radlink default linker on Windows
• [#7701]: Multithread The Entire Semantic Checker
• [#7728]: map performance improvements

General fixes and improvements

• [#7654]: Experimental support for cross-linking Windows executables
• [#7788]: Add the new -windows-sdk-root to the odin help build command
• [#7497]: Fix [#7488]: -vet error in core_builtin_soa.odin when appending multiple empty structs to #soa[dynamic]
• [#7500]: Fix [#5851]: Detect and error on chained offset_of expressions
• [#7505]: Apply mojibake magic to test runner, too
• [#7509]: Fix [#7499]: Add ctrl-c handler to run_subprocess
• [#7070]: Fix vendor:ENet.Buffer data/length ordering issue
• [#7524]: Fix alias nullptr ICE
• [#7528]: intrinsics.type_elem_type() added case for Type_FixedCapacityDynamicArray
• [#7531]: Improve error message for constant indexing
• [#7530]: simd_abs codegen improvement + fix for unsigned integers
• [#7545]: Fix [#6917]: LLVM IR va_arg instruction uses wrong stride for C varargs on Windows
• [#7548]: Fixes simd_approx_recip and simd_approx_recip_sqrt
• [#7549]: Fixes simd_select and simd_runtime_swizzle operand construction
• [#7574]: Fix [#7573]: Check Enumerations iota fits explicit base type
• [#7570]: Fix [#7544]: Constant bit_field field access returning zero value
• [#7568]: Fix side_effects vs volatile typo in error message
• [#7561]: Fixes [#6888]: SIMD gather/scatter and masked load/store randomly crashing the compiler
• [#7550]: Fix syntax error type with -json-errors
• [#7577]: Fix #location in global variable initializers
• [#7582]: Fix [#7581]: Compiler crash when checking code with swizzled SOA fields
• [#7605]: Fix [#7604]: Allow building at /
• [#7607]: Fix [#7598]: Keep entity iteration valid across a scope map rehash
• [#7623]: Add -file suggestion when main is duplicated or package names differ
• [#7628]: Fix quadratic type checking case: don't cache polymorphic record specializations
• [#7631]: Fix record instantiated poly proc on selector when used as a proc value
• [#7629]: Fix [#7596] and [#7511]
• [#7621]: Fixes the attributes of the context parameter in dummy procedures
• [#7641]: Fix field sorting for bit_field literals
• [#7656]: Fix [#7609]: Emit union constants for non-constantable unions
• [#7658]: Fixes #must_tail on WASM targets
• [#7652]: Fix [#6951] and [#5214]: $ procedure arguments matching the wrong instantiation
• [#7639]: Fix [#6302]: Broadcasting arrays of matrices and matrices
• [#7661]: Fix [#7587]: Full length swizzles through a pointer keeping the pointer type
• [#7594]: Fix [#7583]: -strict-style requires comma before newline after proc parameters which have a default value
• [#7588]: Fix [#7427] and [#7193]: Error on OOB tuple element in compound literals
• [#7668]: Fix [#5318]: Procedures passed to $ parameters share an instantiation's name
• [#7666]: Add ExactValue_Rational to improve precision on compiler float computation
• [#7602]: Adds lifetime markers for named locals
• [#7674]: Fix [#5105] and [#5569]: Deferred call in short circuits causes LLVM codegen failure
• [#7682]: Fixes the nocapture attribute on LLVM >= 21
• [#7679]: Fix [#7477] and [#7506]: Casts between nested arrays of the same depth
• [#7697]: Fix compound literals of array backed bit_fields
• [#7692]: Disallows the "odin" calling convention with #must_tail
• [#7706]: Fix [#7430]: Fixed capacity dynamic array literals lose their length
• [#7709]: Fix [#7708]: Compare a polymorphic record's constant parameters by value in subst_unify
• [#7711]: Fix [#7700]: Constant #soa field access yields the per-element field array; compare and canonicalize typeid constants
• [#7716]: Fix intrinsics.procedure_of with parapoly specializations
• [#7714]: Fixes [#3995], [#4337], [#5950], [#7197], and [#7640]: Handle globals initialized with the address of a non constant literal
• [#7718]: Fix [#7490]: Preserve zero-initialized arrays in intrinsics.concatenate
• [#7722]: Use a response file for MSVC-style linkers (link.exe and radlink.exe) when the command line exceeds the CreateProcess limit of 32KiB
• [#7723]: Fix mark a generated polymorphic procedure before publishing it to other threads
• [#7520]: Fix #soa pointer field access through using
• [#7743]: Allow using on #soa pointers, and make the fields a using #soa pointer parameter promotes assignable
• [#7746]: Fix -linker:radlink being rejected on Windows
• [#7740]: Fix [#7689]: Incorrect debug source location at the end of if/else chains

New Packages

• [#7523]: asm: Generalize memory operand syntax to a flat term list

Package Improvements

Speedups

• [#7630]: core:strconv: Ported Eisel-Lemire's fast float algorithm to Odin
• [#7672]: core:strconv: Second part of the Fast Float port (SWAR based scanning)
• [#7725]: core:strconv: Adds a #config to opt out of fast float (and its table)
• [#7726]: core:hash/xxhash: Add a few more #no_bounds_check to XXH3
• [#7728]: base:runtime: map performance improvements

General fixes and improvements

• [#7496]: core:sys/windows: Add bindings for advanced mouse/pointer/pen manipulation
• [#7508]: vendor:windows/wasapi: Fixes
• [#7516]: core:mem: Revert [#6476], which misunderstood the mem.Scratch allocator's semantics
• [#7517]: core:mem: Implement tests for the mem.Scratch allocator
• [#7489]: core:simd/arm: Implement SIMD Arm Neon shift left
• [#6787]: core:fmt: Improved fmt_os documentation
• [#7527]: core:simd/x86: Fixed typo in avx.odin
• [#7539]: vendor:curl (Linux): Remove import declarations for crypto libraries that differ by distro
• [#7532]: core:sys/windows: Add line-drawing functions in gdi32.odin
• [#7563]: core:simd/arm: Implement SIMD Arm Neon shift right
• [#7576]: vendor:sdl3/ttf: Use [^]byte rather than cstring and add ExpandString wrapper
• [#7572]: vendor:libc-shim: Replace mem_zero with null termination in strncpy
• [#7565]: core:sys/windows: Add performance counter PSAPI bindings
• [#7553]: vendor:box2d: Add missing GetVersion binding
• [#7546]: vendor/box3d: Support BOX3D_SHARED on Windows
• [#7543]: vendor:egl: Bind more procedures and constants
• [#7599]: core:os: Fix the 0th env var being duplicated on Linux
• [#7600]: vendor:OpenEXRCore Fixes
• [#7536]: core:odin: Fix [#7526]: Handle triple quote/tick string
• [#7603]: core:simd/arm: Implement SIMD Arm Neon zip unzip extract
• [#7615]: base:runtime: Fix [#7562]: Windows AMD64 no-CRT stack probe support
• [#7620]: core:dynlib: Improvements and explanations in example
• [#7622]: core:strconv: Fixes three bugs in hex float parsing
• [#7624]: core:math/big: Fix squaring path writing the denominator into the numerator
• [#7633]: core:encoding/json: Size the unquote buffer for replaced invalid UTF-8
• [#6722]: core:sys/linux: Fix Time_Val precision comment
• [#7634]: core:crypto/_fiat: Fix P-256 fe_sqrt addition chain, unblocking compressed SEC1 point decoding
• [#7627]: core:rexcode: Add 512bit vmovntdq(a)
• [#7659]: core:encoding/base64: Add strict decoding options
• [#7663]: core:simd/x86: Implement simd x86 avx2 avx512 shift logical
• [#7632]: core:simd/arm: Implement simd arm neon transpose reverse
• [#7579]: core:c: Fixes [#7540]: fastX type widths
• [#7584]: core:sys/windows: Extend Win32 common dialog bindings
• [#7610]: core:sys/windows: Fix [#7547]: Double-null-terminate path in SHFileOperationW
• [#7665]: core:container/bit_array: Add create_from_enum helper. Improve documentation, fix examples.
• [#7671]: core:container/xar: Fix chunk slot count capping Array capacity
• [#7484]: core:sort: Document all public procedures and Interface types
• [#7676]: core:crypto/aegis: Minor documentation cleanups
• [#7673]: core:encoding/json: Removes @(private) from unmarshal_value for use in user unmarshalers
• [#7590]: core:os: Allocate the long-path buffer in the caller's allocator
• [#7320]: vendor:sdl3: WASM foreign import library configurable
• [#7684]: core:rexcode: Changed append to append_elem for a sizable improvement
• [#7683]: core:simd/x86: Implement simd x86 avx512 evex512
• [#7690]: core:encoding/base64: Add unpadded base64 encoding
• [#7695]: core:simd/x86: Implement simd x86 avx512 typo fix
• [#7703]: core:encoding/csv: Preserve final CSV records in multiline mode
• [#7704]: core:encoding/cbor: Fix bit_set byte order
• [#7643]: vendor:sdl2: Add RenderSetVSync
• [#7721]: core:sync/chan: Accept .Recv channels in chan.recv and chan.try_recv
• [#7733]: core:odin: Fix parser assert with invalid for-in stmt
• [#7731]: core:time/timezone: Preserve fractional timestamps in timezone conversions
• [#7738]: core:math/linalg/glsl: Fix mix_vec4 and mix_dvec4 implementations
• [#7753]: core:fmt: Fix incorrect number of bytes written returned when formatting floats with h/H
• [#7751]: core:simd/x86: Implement simd x86 sse2 _mm_insert_epi16 fix
• [#7755]: core:sys/windows: Add GetCursor and GetDpiForSystem
• [#7756]: vendor:box3d: Fixing typo from userMateriald to userMaterialId

Miscellaneous

• [#7495]: macOS Intel removed from CI tests and releases, due to general flakiness of the CI. Odin can still be built from source on Intel Macs.
• [#7569]: Fix [#7525]: Add test coverage for inline-asm label definitions on AMD64
• [#7670]: Update NetBSD CI from 2026Q1 -> 2026Q3
• [#7707]: Adjust doc comments that misrender as CommonMark
• [#7754]: Upgrade Ubuntu CI to LLVM 22
• [#7744]: Update test_core_runtime.odin

Full Changelog: https://github.com/odin-lang/Odin/compare/dev-2026-09...dev-2026-10