Milestone 365: xtask in modules, taken while the board is thin

[calef]

Lane: milestone/365-xtask-in-modules, written by an agent; calef's account is the author GitHub shows.

Milestone 365. xtask/src/main.rs was 11,124 lines with no module structure, and one of the three merge hotspots every lane wires its test into. It is now 19 modules plus a 294-line dispatcher, split along the seams the file already had, which are the commands. Base 638b6bfee83b2f17e5cef97178332fa92537ef39.

The layout, and what lost

The largest file is now shell_check.rs at 1,866 lines; the smallest is host.rs at 130. The four places where something other than "one module per command" won:

• Shared machinery got a module when it is a thing, not when it is leftovers. scanout.rs, inbound.rs and measure.rs are instruments several commands use, each named for what it does. A common.rs would have been a second hotspot wearing a new name.
• host.rs is the residue, deliberately small: cargo, run, capture, llvm_tool, workspace_root, bin_elf, flag_value. Named for what it holds rather than for the fact of being shared, which is what keeps it from growing into a common.
• The soak and the job mix left the board. They share a recogniser with board-console and run under QEMU, never touching a board, so they are soak.rs.
• The tests went with the code they test, not into a tests.rs, which would have been a fourth hotspot.

initrd_aarch64 is the only thing that moved out of file order, to sit with the two packers it is a sibling of.

The evidence that behaviour did not change

• The move is provably a move. Every non-blank line of the original, ignoring use lines and the pub(crate) prefixes the split forced, appears in the new files: zero missing, and the only additions are the module doc comments and the mod tests wrappers.
• script/fastpath-footprint is byte-identical on all three ISAs, before and after, down to the per-symbol totals.
• cargo xtask <unknown> prints the same usage text byte for byte, which is the one output naming every subcommand and flag.
• The 21 host tests are the same 21 by name.

Milestone 130's refused kernel_main split is why that list exists rather than a claim it was unneeded. xtask is a host binary compiled in one configuration with no cfg-gated early park, so the compiler really does verify this split completely.

Three things it broke, all build errors

script/lint's host-pass check read xtask/src/main.rs by name for the bare-metal exclusion list (it reads every xtask/src/*.rs now); three intra-doc links stopped resolving across files; and a banner lifted from // to //! became documentation that clippy's doc_markdown then had an opinion about.

Eighteen citations in notes/ and script/ that direct a reader at a symbol were repointed at its module, one at a time against the symbol rather than by a sweep. Seven that describe the past were left alone.

Gates, all from this worktree

gate	exit
script/lint	0
script/names (refusals unchanged)	0
cargo test -p xtask (21 passed)	0
script/test (aarch64, riscv64, x86_64/OVMF)	0
script/shell-check (all three architectures)	0
script/shell-check --graphical	0
script/shell-check --graphical-serial	0
script/boot-check	0
script/bench --check	0
script/icount	0
script/fastpath-footprint (byte-identical)	0

This lane took the machine-global nife-dev link, as every gating lane does.

Commits

The split is its own commit and does not build on its own, by construction: a reviewer can read it as "these lines went there" with no visibility change or import line in the way. The next commit is the fixups the compiler then demanded.

🤖 Generated with Claude Code

https://claude.ai/code/session_01STu3VeDYnEHem3iKCbzBF2

[calef]

merge-drain: STALLED. #1001 has conflicts a person must resolve (Milestone 365: xtask in modules, taken while the board is thin)