Channel Runtime Conformance

This document defines the stabilization gate for std.channel behavior across runtime backend selections and host OS behavior lanes.

Scope

The gate focuses on behavior that must remain stable while channel/runtime APIs are consolidated:

• Return-code compatibility for core send/recv/select paths
• Select fairness under sustained ready-state load
• Timeout behavior under empty-channel waits
• Cross-backend consistency when selected via environment override
• Host OS behavior coverage for scheduler/timer/runtime differences

Backends are selected through FUN_RUNTIME_BACKEND and validated with both values:

• posix
• windows

Host lanes are also validated in CI:

• Linux (ubuntu-latest)
• macOS (macos-latest)
• Windows (windows-latest)

CI Coverage Matrix

Lane	runs-on	FUN_RUNTIME_BACKEND	Purpose
linux-posix	ubuntu-latest	posix	Primary POSIX host behavior + regression gate
macos-posix	macos-latest	posix	Native macOS scheduler/timer behavior
windows-native	windows-latest	windows	Native Windows scheduler/timer behavior

Compatibility Matrix

The following operations are treated as compatibility anchors:

Operation	Expected result
try_recv on open+empty channel	channel_rc_empty() (2)
send_timeout when space available	channel_rc_ok() (0)
try_send on full channel	channel_rc_full() (2)
recv_timeout_into with queued value	channel_rc_ok() (0) and value consumed
recv_timeout_into on open+empty channel	channel_rc_timeout() (2)
send_timeout_with_cancel with raised cancel flag	channel_rc_cancelled() (3)
send_timeout after close()	channel_rc_closed() (1)
recv_timeout_into after close+drain	channel_rc_closed() (1)
select_recv_default_with on two open+empty channels	channel_rc_default() (3), index channel_select_index_default() (-1)
select_recv_timeout_with_tuning_cancel on two open+empty channels	channel_rc_timeout() (2)
select_recv_timeout_with_tuning_cancel with raised cancel flag	channel_rc_cancelled() (3)

Alias constraints are also part of compatibility:

• channel_rc_timeout() == channel_rc_full() == channel_rc_empty() == 2
• channel_rc_default() == channel_rc_cancelled() == 3

Stress Benchmark Thresholds

A micro-benchmark validates select round-robin fairness and timeout stability:

• Workload:
  • select_recv_timeout3_rr_with_tuning_cancel over three prefilled channels (360 receives total)
  • 20 timeout rounds over empty 3-way select with 15ms timeout each
• Fairness threshold: fairness_skew <= 1
• Completion threshold: count_total == 360
• Timeout threshold: timeout_failures == 0 across all timeout rounds
• Process elapsed budget per host lane run: 150ms <= elapsed_ms <= 5000ms
• Cross-backend elapsed drift budget within a host lane: abs(posix_elapsed_ms - windows_elapsed_ms) <= 800ms

CI Gate

The tests in tests/codegen_test.zig are the executable gate:

• std.channel runtime conformance matrix is stable across backend selectors
• std.channel fairness and timeout benchmark stays within backend thresholds

These gates run in every CI lane from the matrix above.

When these tests fail, treat it as a behavior regression and stabilize before adding new channel/runtime API surface.