[Defilan]

What this changes

zig build compile-dsv41-server fails on main with zig/src/server/clef_systemone.zig:5:25: error: no module named 'lanes' available within module 'root', because serverOn in zig/build/dsv41_draft.zig doesn't give the server module lanes. dsv41-serve goes through the same serverOn. This adds that one line, with no source changes.

#2379 already has this same line, in the same place. I test-merged the two and the result has one copy, so they won't conflict. If #2379 lands first, close this one.

• Issue: n/a. @ashhart asked for this as a separate PR on deepseek_v41: plan and emit a world of 3, host tests only #2325.
• Stacked on: main.
• atc overlap --json: run 2026-11-07 02:00Z with main at c92956e7c. On its own it returns complete: false, so I read the file lists of all 223 open pull requests. Five touch this file: TensorFold Web: a small local web app, and tensorfold web to open it #1781, DeepSeek-V4.1 TP=2: every-rank receipts and opt-in per-layer hash files for repeat divergence (#1395) #2030, Unlimited-OCR-2 on native Metal: the publisher's BF16 release, page images to structured text #2074, Unlimited-OCR-2 on CUDA: the family's steps on NVIDIA GPUs, the router's long-context speedup (stacked on #2074) #2077 and DeepSeek-V4.1 on CUDA: refuse a malformed TF_DSV41_* setting by name before loading, and register the ones that change outputs #2379. atc overlap --pulls reports all five clean. The full run also lists conflicts on LightOnOCR-3 (0.8B, 1B, 4B) on native HIP, with the model card's oracle #1965 and gfx12 WMMA dense tile: 64-deep swizzled K step, LDS-only barrier (same bytes) #2280, and neither touches this file.

Checks on this head

• Head commit: 82b390d96, base e2b2dc118. It merges cleanly with c92956e7c.

• Lean checker, python3 tools/zig/lean_check.py, on the base and on this head: 146 and 146 by tools/zig/lean_new.py, 0 new. The 146 are inherited.

• Tests run, with the command and pass, fail and skip counts:

  • zig build compile-dsv41-server: exit 0, 8 of 8 steps. It failed without this line when I last checked, at b3933ecd3.
  • zig build compile-dsv41-cuda: exit 0, 6 of 6 steps.
  • zig fmt --check build.zig zig and git diff --check: clean.
  • zig build test test-golden -Dcpu=apple_m1, in a checkout at a 49-character path: 11043 passed, 89 skipped, 1 failed, explained below.
  • zig build test-cuda-host -Dcpu=apple_m1, in a checkout at a 40-character path: 5110 passed, 30 skipped, 0 failed.

• That one failure isn't from this change. exl3_load: a scan that fails an allocation frees every group prefix it held fails with NondeterministicMemoryUsage depending on the checkout's path length, with or without this line. Running the test binary directly on an Apple M5 Max:

  Checkout path	Base e2b2dc118	This head
  49 characters	0 of 5 pass	0 of 10 pass
  40 characters	10 of 10 pass	5 of 5 pass

  Those are the only two lengths I tried, and I don't know the mechanism. I can file it as its own issue if you'd like.

• CI: waiting for approval on this head. I'll classify any failure here.

• Not run, and why: the full test test-golden at the shorter path, since the failing binary was the only one affected and I reran it there. dev-tools, since no API changed. dsv41-serve on this exact head: the same line on fe8148d8e built it on Linux with nvcc and CUDA 13.0.3, and those binaries loaded weights on two GB10s (numbers on deepseek_v41: plan and emit a world of 3, host tests only #2325), but they didn't reach serving.

Receipt

n/a. This is build wiring only, with no source changes, and it can't change what an existing binary does: the DeepSeek server doesn't build on main without it.

Checklist

• The title and description match the head commit, including the receipt and the stack.
• New tests fail before the change, pass after it, and skip cleanly without their dependency. n/a: no new test. compile-dsv41-server is the check.
• Comments and docstrings are one line. No measurements or history in the source.
• No personal data, machine names, internal hosts or local paths in the diff or its commits. No AI attribution lines.

[ashhart]

Thanks @Defilan, and for checking it against #2379. #2379 carries the same line and is in our landing queue tonight, so this one closes once #2379 is on main. If #2379 stops for anything, we'll take this instead.

[ashhart]

A change of plan, @Defilan: #2379 is held until after tonight's release cut, so this one lands on its own. It's in the landing queue now, with a CHANGELOG line we've added. Thanks for splitting it out.

[ashhart]

Thank you @Defilan. This is merged into main with your commits under your own name, so you're listed as a TensorFold contributor. It ships in the next release, credited to you in the changelog and the release notes.

[ashhart]

Thanks @Defilan! This is in TensorFold 1.2.4, out now: https://github.com/ashhart/TensorFold/releases/tag/v1.2.4