Summary

cel-expr-python currently publishes wheels for:

• manylinux_2_27_x86_64 / manylinux_2_28_x86_64 (Linux x86_64)
• macosx_11_0_arm64 (macOS Apple Silicon)
• win_amd64 (Windows x86_64)

There are no manylinux_*_aarch64 wheels, so pip install cel-expr-python fails on Linux ARM64 when --only-binary=:all: is set (common in Docker/CI), and otherwise requires a full Rust toolchain to build from source.

Motivation

I'm a maintainer of omnigent, where we use cel-expr-python for inline policy evaluation (CEL expressions for agent sandboxing). We've had to add a platform_machine != "aarch64" environment marker and conditional imports to work around the missing wheel (see omnigent-ai/omnigent#300), which means our ARM64 users lose access to CEL policies entirely.

Linux aarch64 is increasingly common across cloud infrastructure:

• AWS Graviton (EC2, ECS, Lambda)
• Azure Cobalt (ARM-based VMs)
• OCI Ampere (ARM instances)
• GCP Tau T2A (ARM VMs)
• Self-hosted (Raspberry Pi, NVIDIA Jetson, etc.)

Suggestion

Add manylinux_*_aarch64 targets to the CI wheel build matrix. Since the project already uses PyO3/maturin, this should be achievable by adding the aarch64 target to the existing GitHub Actions workflow (e.g., via maturin-action with target: aarch64-unknown-linux-gnu and QEMU emulation or cross-compilation).

Thank you for the great library!