LLVM Toolchain

Kira ships pinned LLVM bundles so contributors do not need to build LLVM manually as the normal path.

Install The Bundle

kira fetch-llvm

or:

zig build fetch-llvm

That flow reads llvm-metadata.toml and installs the matching host bundle under:

~/.kira/toolchains/llvm/<llvm-version>/<host-key>/

What The Repo Pins

The repository source of truth is llvm-metadata.toml.

It records:

• the pinned LLVM version
• the release tag that owns the bundle assets
• supported host keys
• exact asset filenames

The current repo pin is LLVM 22.1.2.

Supported Hosts

The first published bundle set is intentionally small and tied to real CI-built hosts:

• x86_64-windows-msvc
• x86_64-linux-gnu
• aarch64-macos

Discovery Order

The LLVM backend checks, in order:

1. KIRA_LLVM_HOME
2. ~/.kira/toolchains/llvm/<llvm-version>/<host-key>/
3. older repo-local fallback paths if they already exist

If llvm-config exists in the selected tree, Kira uses it to refine the bin/lib directories. Otherwise it uses the standard install layout directly.

Native And Hybrid Both Use It

The same LLVM discovery path is used by:

• the pure LLVM/native backend
• the native half of hybrid mode

So if LLVM is not available, both native and hybrid work will fail until you install or override the toolchain.

No Checksum Verification

The current LLVM fetch flow intentionally does not use checksum verification.

The source of truth is:

• the pinned version in llvm-metadata.toml
• the configured release tag
• the host key
• the exact asset filename

The docs are spelling that out because it is a real implementation choice in this repo, not an omitted detail.