Testing and CI

Relevant source files

• .github/workflows/test.yaml 
• tests/ 
• tests/aarch64_jit.rs 
• tests/aarch64/managed.rs 
• tests/fixtures/axiomos/README.md 
• clippy.toml 
• jit.md 

Test suites

tests/common.rs holds shared fixtures (TCP_SACK_ASM, TCP_SACK_BIN, TCP_SACK_MATCH, TCP_SACK_NOMATCH). Doc tests include the README, which lib.rs pulls in with #![doc = include_str!("../README.md")].

AArch64 native tests

The native module is an executable-memory owner used only in tests:

It never creates RWX mappings, and a protection failure fails the test. Native tests cover:

• every_alu_opcode_immediate_and_register_at_both_widths, checked against an independent reference model,
• all_conditional_branches_taken_and_not_taken_at_both_widths,
• alu_widths_wide_immediates_and_forward_control_flow, compact_immediates_clear_old_register_bits,
• arithmetic_failure_stops_execution (zero divisor sets fault_pc),
• callbacks_return_values_effects_and_failures, helper_failure_stops_before_store_or_second_call,
• helper_callback_keeps_argument_order_and_callee_saved_registers,
• memory_callback_preserves_live_r0_and_all_widths, where the callback deliberately clobbers X9,
• generated_entry_preserves_host_registers, where a global_asm! shim seeds X19–X29 and counts how many changed,
• eight_kibibyte_stack_places_r10_at_compiled_top,
• copied_stack_pointer_has_no_r10_flag_and_still_needs_bounds_check.

AxiomOS controller fixtures

tests/fixtures/axiomos/ contains three controllers copied unchanged from AxiomOS feat/v0.5-runtime at 9be704c8: conservative_obstacle, slow_approach and clear_streak (11, 13 and 32 instructions). The checked-in .bin files are their .text, built with Clang 22.1.8 (-target bpfel -mcpu=v3 -O2), so Cargo tests need neither Clang nor an AxiomOS checkout. Regenerate and compare with:

tests/fixtures/axiomos/build.sh target/managed-bpf
for name in conservative_obstacle slow_approach clear_streak; do
    cmp "tests/fixtures/axiomos/$name.bin" "target/managed-bpf/$name.bin"
done

The harness models the R1 context wrapper, a 56-byte payload, an 8 KiB stack, an instance-private map, and helpers 5, 6 and 1009. Its tests:

• controller_trajectories_and_invalid_sensor checks expected drive outputs and map values across sensor sequences,
• relocated_code_reuses_private_instance_maps runs two images of the same program with separate runtimes,
• controller_helper_failures_stop_at_the_call checks that a failing helper yields STATUS_CALLBACK_ERROR at the exact slot with no later effects.

Running on an x86 host via QEMU

CARGO_TARGET_AARCH64_UNKNOWN_LINUX_GNU_LINKER=aarch64-linux-gnu-gcc \
CARGO_TARGET_AARCH64_UNKNOWN_LINUX_GNU_RUNNER='qemu-aarch64 -L /usr/aarch64-linux-gnu' \
cargo test --target aarch64-unknown-linux-gnu --features aarch64-jit

QEMU executes the emitted instructions, callbacks included. It does not validate Pi 5 cache maintenance, kernel mappings, or timing.

CI (.github/workflows/test.yaml)

The workflow runs on pull requests, on pushes to main, and by manual dispatch. Every job uses Swatinem/rust-cache, which saves the cache only on main. In-progress runs for the same ref are cancelled.

Lint configuration

clippy.toml contains a single setting:

doc-valid-idents = ["eBPF", "uBPF"]

jit.md notes that a repository-wide cargo clippy --all-targets -- -D warnings hits an older manual_is_multiple_of warning in examples/date.rs. That warning is unrelated to the AArch64 backend. CI runs plain cargo clippy, without -D warnings.