Next steps: what's left, and moving to Claude Code containers

[zmaril]

Stacked on #1. Adds docs/next-steps.md: where headstart stands, how to run everything in a Claude Code container, and what's left, in priority order, with how to do each item.

Moving to containers

• What an environment needs:
  • Linux with 16 cores.
  • About 100 GB of disk: the rustc build is 12 GB, cargo's target 13 GB, and the registry 8 GB. A single zed or lemmy build needs tens of GB while it runs.
  • Network access to GitHub, crates.io and CI LLVM.
  • Rust 1.98.0.
  • System packages for the real projects.
• Setup: scripts/setup.sh, the rustc-perf submodule, and scripts/real-projects.sh.
• Run times on 16 cores:
  • sweeps: 7–18 min each;
  • UI and cargo test suites: about 3 min each;
  • timing: 25 min to 4 h per pass.
• Lessons from the VM:
  • Timing needs a quiet machine.
  • Detach long runs.
  • Background jobs ignore SIGINT, which hangs cargo's ctrl-C test.
  • Set RUSTC_WRAPPER explicitly.

Open work, in priority order

1. Scheduling when the machine is saturated (atuin under -Zthreads).
2. Fewer cores (-j 8, -j 4).
3. Peak memory under cargo build, and a cap on paused compilers.
4. Timing release builds and incremental edits.
5. Tests that could live upstream: check-swap as a run-make test, the UI suite with the flag forced on, and cargo testsuite cases.
6. Platforms: Windows retraction, macOS timing.
7. Re-verifying Clippy and rustdoc on the current version.
8. A rebase onto a newer rustc.
9. The path upstream: four separable pieces, and the questions reviewers will ask.
10. Measurements to redo: error delay and memory.
11. Tooling fixes found in review.
12. Smaller items.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Tet2pfGBx9bM6ge28w8rgM