config: the resolution pipeline's dispatcher leaves the record

[ch-wan]

The series

Stacked, each PR based on the one above it. Read them in order; every boundary is
self-sufficient and green on its own.

1. #36896 — the resolution pipeline's dispatcher leaves the record ← you are here
2. #36972 — the callbacks into the record go to zero
3. #36973 — six more runtime readers ask the bags
4. #36974 — the dead record parameters go
5. #36975 — the lazy imports that buy nothing become eager

• #36925 — CI vehicle — runs the whole series against main, not for merge

Motivation

_run_resolution_pipeline was the last piece of resolution living on ServerArgs: 331 lines of
ordered dispatch, plus the four handler methods it still reached. Every step it calls already
lives in arg_groups/, and none of them is a member of the record — a step decides about the
record, it is not part of it.

Keeping the dispatcher on the class also kept the awkward direction of travel: a hook in
arg_groups/ was called by a method of the very object it was extracted from.

Modifications

The dispatcher becomes arg_groups.pipeline.run_resolution_pipeline(server_args).

Three of the four remaining handlers were one-line forwards to run_post_process_pass, the same
shape as the ninety-three slots #36792 cut, and are now that call. The fourth is the MLA opt-in
check; it joins the platform family as handle_hardware_runtime_validation. resolve_once keeps
the gate — the resolved/failed flags — and calls the function.

ServerArgs goes from 5643 to 5277 lines and hosts no resolution handler at all.

That changes what one guard can say. test_no_handler_reads_a_field_off_self said a handler on
the record must not read a raw field off self; there is no such handler left, so it becomes the
stronger test_the_record_hosts_no_resolution_handler, and the package-side
test_no_hook_reads_a_field_off_the_record covers what the steps read.

Five guards follow the dispatcher to its new file: the two that count pipeline entries patch the
module instead of the class; the "one caller" scan matches a bare-name call rather than a self.
attribute call; the laziness and ModelConfig-ordering checks parse arg_groups/pipeline.py,
which is the file whose imports map a bare-name call to the family defining it; and the platform
write-capture plugin is installed in the namespace the capture now runs in.

That last one is the seam worth noticing. The test installs an out-of-tree platform by patching
current_platform in the dispatcher's module. Moving the dispatcher moved that namespace, and
without following it the plugin's defaults silently stopped reaching the projection — a real
behaviour the test exists to protect, failing loudly rather than quietly.

The MLA check runs before the dummy short circuit, so platform_hook is imported at its call site
like every other family rather than at module scope.

Accuracy Tests

The 62-shape resolution probe (resolve_once() + check_server_args(), exception type and full
message) is byte-identical to the base commit. The registered tests that mention the record report
the same failure set as main, file for file.

A dummy resolution loads two more modules (pipeline, platform_hook) and is unchanged at ~9 ms.

Speed Tests and Profiling

None. The dispatch is the same sequence of calls, one stack frame shallower.

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.
• Follow the SGLang code style guidance.

Review and Merge Process

Mostly mechanical: arg_groups/pipeline.py is _run_resolution_pipeline verbatim with self
renamed and four call sites rewritten. The parts worth reading are the guard changes in test/,
and in particular the platform-plugin patch target — the one place where moving the dispatcher
moves a namespace a test depends on.

Checked on this tip: the 77 handler and pass calls stay in the same order, the MLX validation still
runs immediately before the dummy return, _run_resolution_pipeline has no leftover caller, and
the plugin test patches pipeline.current_platform.

---

CI States

Latest PR Test (Base): ❌ Run #33242578573
Latest PR Test (Extra): ❌ Run #33242578477
Latest PR Test (AMD ROCm 7.2): ❌ Run #33242578540

[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 2026-08-29T06:37:34.325612Z	4c96ae3	New commits

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.