refactor(pipeline): split pipeline.zig into focused modules (#434)

[apotema]

This is a refactor with no behaviour change. src/cli/pipeline.zig was 3,985 lines, which is over the toolkit's 1,000-line limit. It follows the serve.zig (#436) and provider_github.zig (#437) splits. The root re-exports the same public API (run, plus the three specs cli.zig re-exports), and every module is referenced from a root test {} block. The helpers shared by the tests live in pipeline/testing.zig.

One exception to "moves only": run

run was a single function of about 1,310 lines, so moving it whole was not an option. It is now split into stage functions. Each one is a verbatim block of the old body and runs the same statements in the same order. The mechanical changes:

• Early exits. A stage that used to return code / return 1 now returns ?u8 (null means go on), or a small union { proceed / ready, exit: u8 }. run returns that code at the same point, so the exit statuses are unchanged.
• Defers. Every resource with a defer stays declared in run: the arena, target_dir, the reporter and its errdefer failActiveStage, asm_bin, provider_sources, output_dir, the zig args/env and hook_site. Stages borrow them through pipeline/context.zig (Context). The watched serve's defer watch_replan.deinit(...) still fires after finishServe is evaluated and before run's own defers.
• Hook plans. The anonymous hook_plans = .{ ... } literal is now the named HookPlans struct. The fields and the order they are evaluated in are unchanged.
• Locals. Inside the stages, locals are bound from cx under their old names. &hook_site became hook_site, and &parsed_args became parsed_args (a pointer now).
• No reordering. Nothing moved in or out of the order. For example, generate_out / build_out are still computed right before their stage runs.
• Test fixture. One edit in test code: the anonymous fixture installer in the re-gate test (now in generate.zig) had to become pub fn install, because gateThenInstall lives in another file now.

File map

File	Lines	Contents
src/cli/pipeline.zig	406	Root: module doc and map, re-exports, ok, run (stale gate, parse, upgrade, feed/reporter setup, asm resolve, install call, hook site, ios/android handoff, zig env/args, stage calls), test {}
pipeline/args_resolve.zig	251	resolve (the NAME half of target resolution, platform/backend derivation), TargetRefusal, TargetVerdict, confirmTarget, EarlyVerdict, earlyTargetCheck + 2 tests
pipeline/install.zig	291	preInstall (compatibility, .prebuild, Python/SDL2 wiring), gateThenInstall, AssemblerInstaller, discoverAndPlan (discovery, ownership, plans, #378 gate, bundle check, plugin-core compat, lock write) + 1 test
pipeline/generate.zig	314	run (before/core/after generate, ASTC and --bake prepasses, fingerprints, emsdk activation), beforeGenerate + 1 test
pipeline/build.zig	210	run (before/core/after build, docker/progress/captured zig build, SDL2 DLL staging, labelle build finalisation), bundleStep
pipeline/run.zig	582	launch (run hooks, wasm export/serve + --watch wiring, iOS/Android deploys, docker/host launches), serveNoBuild, noteRunSharesStdout, crossTargetLaunchSkipped + 1 test, runOutcome, runOptionEnv
pipeline/context.zig	39	Context, HookPlans
pipeline/watch.zig	747	WasmRebuildCtx + its 3 tests, collectPrebuildIgnorePaths, CollectPrebuildIgnorePathsSpec
pipeline/watch_replan.zig	286	WatchReplan, Generation, Installer
pipeline/watch_replan_tests.zig	812	The 8 WatchReplan tests and the TwoStage fixture (they left the struct because WatchReplan plus its tests was about 1,060 lines)
pipeline/screenshot.zig	235	screenshot_suffixes, FileStamp, ScreenshotProbe, ScreenshotProbeSpec
pipeline/export_output.zig	209	resolveExportOutput and its path helpers, ResolveExportOutputSpec
pipeline/testing.zig	20	testSite (was WasmRebuildCtx.testSite)

The largest file is 812 lines.

Agnosticism guard allowlist delta

• Added (11): every new file except context.zig: args_resolve, build, export_output, generate, install, run, screenshot, testing, watch, watch_replan and watch_replan_tests. Each one carries moved platform words (the legacy target/backend sites, wasm serve, bgfx#57, fixture "wasm"/"bgfx" tags).
• Kept: cli/pipeline.zig. The root is still dirty (its module doc, .wasm, the ios/android handoff), and the stale-entry test that names it is unchanged.
• Clean, so not listed: context.zig.
• The allowlist goes from 58 to 69 entries, and the doc comment is updated to match.

Tests: same named-test set

	Before	After
Build summary (zig build test --summary all)	1072/1077 passed (5 skipped)	1073/1078 passed (5 skipped)
Main CLI test binary	1040 total	1041 total
Named test "…" decls from pipeline	35	35 (same names)
Guard	17 pass	17 pass

The test names were pulled from both test binaries (the builtin.test_functions strings) and diffed after stripping the namespace path, which the move changes. The two sets match: 1,084 entries each. The only differences are five compiler-numbered __struct_N ids on anonymous test fixtures. The +1 in the total is the new unnamed test {} aggregator in pipeline.zig, so none of the moved tests were dropped by lazy analysis.

Verification

Each command's exit code was checked on its own:

• zig fmt --check (new and changed files): 0
• zig build: 0
• zig build -Dtarget=x86_64-windows: 0
• zig build test (includes the guard): 0
• zig build test-provider-dispatch: 0
• test/provider_dispatch_e2e.py (25 invocations), provider_github_e2e.py (70), provider_config_e2e.py (21), provider_hooks_e2e.py (43), provider_targets_e2e.py (40), plugin_compat_e2e.py: all 0

Closes #434

https://claude.ai/code/session_01R2K5sJDL5Msn1uRTkybiDP

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-26T22:07:18.070372Z	8ec59b0	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Team

Run ID: 3c4190ae-fe74-42c4-b476-6c80d6982f15

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

_{Comment @coderabbitai help to get the list of available commands.}