Update Orchestrion to v1.12.0 and preserve partial uploads - #214

Summary

• update the maintained Orchestrion integration from v1.9.0 to v1.12.0 and the default dd-trace-go/v2 pin from v2.9.0 to v2.9.1
• preserve Orchestrion's upstream test-variant resolver while retaining the Bazel execroot/tempdir and resolver-context compatibility patches
• make the resolver cycle-guard source patch work with both LF and CRLF checkouts
• preserve exact configured dd-trace-go module versions in the offline proxy, including prerelease pins that differ from the module graph's selected stable version
• regenerate the public profiles for rules_go v0.60.0, v0.61.1, and v0.62.0, and update fixtures, onboarding helpers, docs, and skills
• make declared rules_go stdlib cache outputs deterministic by building through a private scratch GOCACHE and publishing manifested woven archives under content-addressed paths
• isolate non-Go runtime data tools from Orchestrion and avoid reinstrumenting an already synthetic Go test main during Test Optimization linking
• batch CODEOWNERS enrichment once per target set instead of repeating it for every payload
• keep processing every available fresh valid payload when another target or validation phase fails, while preserving the earliest failure as the command result
• apply the same partial-result behavior to Bash, PowerShell, generated Go validation scripts, documentation, and onboarding skills

Orchestrion v1.12.0 requires Go 1.25.0 or newer.

Why

The v1.12 resolver tracks test variants for generated test mains and recursively resolves the correct dependency archives. The previous v1.9 integration disabled that recursive toolexec path for Bazel compatibility; this update keeps the upstream behavior and narrows the local patch to Bazel filesystem and invocation-context differences.

The offline proxy downloads both the resolved module graph and the exact configured dd-trace-go pins. This keeps normal Go module resolution intact while ensuring synthetic Orchestrion dependencies can resolve a consumer's prerelease pin without network access.

A failed test can still produce a valid Test Optimization payload containing the failure details. Validation failures and missing sibling outputs are therefore reported without discarding valid fresh payloads from the same invocation. The final status still reports the earliest test, doctor, dry-run, or upload failure.

Repository transitions reset Orchestrion for non-Go runtime data tools. The Go linker also recognizes a Test Optimization synthetic test main and does not route that already generated source through Orchestrion again. This keeps the recursive resolver available where required without applying it to unrelated tool binaries.

The resolver source downloaded by the repository rule can use the checkout.s native line endings. Building the replacement text with the detected line ending keeps the same patch semantics on Linux, macOS, and Windows.

The stdlib builder never exposes its declared cache TreeArtifact as a writable GOCACHE. Plain builds leave it empty, while Orchestrion builds publish only sorted, manifested *-d archives from an action-private scratch cache. This removes nondeterministic Go cache metadata and scratch entries from Bazel outputs. Content-addressing the published archives also prevents environment-sensitive Go action IDs from changing declared output paths for byte-identical archives.

Validation

• python3 tools/dev/generate_rules_go_fork_maps.py --check
• python3 tools/dev/materialize_rules_go_fork.py check --all
• python3 tools/dev/check_release_archive_contents.py
• python3 tools/dev/verify_rules_go_profiles.py --public-denylist tools/dev/private_leak_public_denylist.txt --bazel <macOS Bazel wrapper>
• focused bootstrap, onboarding pin, workspace helper, CODEOWNERS batching, runtime-transition, and profile-tool tests
• focused resolver-cycle patch tests for LF and CRLF across rules_go v0.60.0, v0.61.1, and v0.62.0
• WORKSPACE and Bzlmod Go integration scripts for rules_go v0.60.0, v0.61.1, and v0.62.0, including exact v2.9.1-rc.3 offline-proxy coverage
• ./bazelw test //... --noexperimental_split_xml_generation (482 tests passed)
• deterministic stdlib cache verification across two isolated output roots for rules_go v0.60.0, v0.61.1, and v0.62.0 on macOS, plus Linux/amd64 verification for v0.60.0, with disk and remote caches disabled
• local-archive consumer validation across all three rules_go versions, covering disabled mode, fresh Test Optimization payloads, unchanged action-cache hits, doctor, and mock uploads
• ./bazelw test //tools/tests/python:python_tools_test --noexperimental_split_xml_generation
• ./bazelw test //modules/go/tools/dd_topt_go_bootstrap:bootstrap_test --noexperimental_split_xml_generation
• local-archive WORKSPACE and Bzlmod consumer fixtures, including real uploads of failed-test payloads to their mock backend
• dd-source load run with 83 optimized Go targets across 14 contexts: 80 passed, 3 preserved test failures, doctor and dry-run passed, and all 506 fresh payloads (261 test plus 245 telemetry) uploaded successfully