5a6de314ab
* ok wgpu runs hello world * toy hasher * separate wgsl file * kinda poseidon2 * test vectors * closer * a bit of debug * closer * sort of almost * fixed gf_mul * merge gf_mul back into mining, clean up tests * fix bytes_to_field_elements * linear layer fixed * maybe better? * mds tests * cleanup debug output * found gf_mul bug * fixed gf_mul bug * efficient sbox again * mds matches * cleanup * test internal layer * cleaner * external linear passes * external layer tests * fixed the constants * moar tests * more cleaning * actually test just internal * fairly clean tests now * removed another pointless test * more cleanup * gf_from_const tests * testing * fixed permutations * double hash tests, but they fail * moar tests! * hash-twice works, thanks Gemini 3 Pro! * gpu runs but doesn't mine * better logging * lock workers to gpu * send target to gpu not difficulty * fix verify_nonce * double hash failures * double hash matches * tests pass and it mines now * fix benches * a bit faster now * improved logging * fix benches * simplified gpu code * further simplification * a bit simpler * slight simplification * a bit faster * dynamic thread allocation and work coarsening * more benching * faster * faster * faster * clean up documentation * a bit faster * simplify * simplify * optimized mds * almost par with cpu now * fmt * clippy * support multiple gpus * clippy * unused * fixed some logs --------- Co-authored-by: Nikolaus Heger <nheger@gmail.com>
6.4 KiB
6.4 KiB
Agents and Authors Guide
This document defines the authoring workflow and process conventions for proposal/response documentation and how commits and PRs must link to those docs. It exists to make PR review fast, predictable, and auditable.
If you’re reviewing a PR, the canonical technical narrative lives in:
- This process guide (agents.md) is the single source of truth for authoring and process now.
- Commits and PRs should include clear, self-contained descriptions and link to relevant code diffs.
If you’re authoring changes or operating as an agent, follow this guide.
Purpose
- Keep changes small, explicit, and justified.
- Capture the plan (prompt) and the actual implementation (response) for each iterative step.
- Ensure every commit and PR references the relevant doc(s), so reviewers have full context without hunting.
Where things live
- This process guide: /agents.md (for authors/agents and reviewers)
- Repo overview and user-facing guidance: /README.md
Numbering and naming conventions
Each iteration is a prompt/response pair using a two-digit sequence:
- 08-prompt-.md
- 08-response-.md
Conventions:
- Two-digit sequence numbers (00, 01, …, 10, 11, …) in chronological order.
- short-topic is lowercase, words separated by hyphens. Remove standalone iteration documents. Capture sufficient context directly in commit messages and PR descriptions.
Commit message guidance:
- Title: imperative, concise
- Body: motivation, high-level changes, key code paths, risks/mitigations
- Footer: links to key diffs or files (if helpful)
Authoring workflow
- Plan the change
- Capture background, objectives, scope, non-goals, risks/mitigations, and validation in the PR description template (see below).
- Prefer concise planning; link to tracking issues when relevant.
- Implement the change
- Commit in small, reviewable steps.
- Keep messages crisp; reference the PR and key deltas.
- Document the outcome
- Update the PR description with deviations from plan, validation results, and any follow-ups.
PR description template (copy/paste into GitHub):
- Overview
- What changed (by area/crate)
- Validation (tests, clippy, taplo, manual checks)
- Risks and mitigations
- Follow-ups (if any)
Commit and PR linkage
Every commit must reference the iteration doc. Use this template:
Commit message template
- Title: imperative, max ~72 chars
- Body: details and motivation
- Footer: links to the prompt/response
Example: Fix clippy lints and inline format args across miner-service
- Inline {var} format args in logs
- Remove redundant locals/fields
- No behavior changes
Docs:
- Prompt: docs/implementation/07-prompt-stable-clippy-taplo-and-gpu-enum.md
- Response: docs/implementation/07-response-stable-clippy-taplo-and-gpu-enum.md
PR description template
- What and why (summary)
- Validation (what you ran, results)
- Risks and mitigations
- Links:
- Prompt: docs/implementation/NN-prompt-.md
- Response: docs/implementation/NN-response-.md
Local validation checklist (pre-PR)
Use the stable toolchain.
-
Toolchain and lints
- rustup show active-toolchain -> stable
- cargo fmt --all -- --check
- cargo clippy --workspace --all-targets -- -D warnings
-
TOML formatting (Taplo)
- taplo fmt --check
-
Build and tests
- cargo build --workspace --locked
- cargo test --workspace --locked
-
Runtime sanity (typical)
- cargo run -p miner-cli -- --engine cpu --workers
- Check logs and metrics if enabled (see below)
Notes:
- We standardize on stable (rust-toolchain sets channel = "stable").
- We keep components = ["clippy", "rustfmt"].
- Taplo formatting is authoritative per taplo.toml.
CI expectations (what reviewers look for)
- Stable toolchain
- taplo fmt --check
- cargo fmt --all -- --check
- cargo clippy --workspace --all-targets -- -D warnings
- cargo test --workspace --locked
If PRs introduce new feature flags or targets, document how CI should build them, or gate them until CI can.
Engine selection policy (naming and runtime)
Naming:
- CPU engines use simple, clear names:
- cpu, cpu-chain-manipulator
- GPU engine is implemented and functional:
- gpu for cross-platform GPU mining using WGPU
Runtime behavior:
- All engines are fully implemented and production-ready
- GPU engine automatically detects and optimizes for available hardware
Metrics and observability conventions
- Metrics are feature-gated (metrics feature) and can be toggled at runtime with --metrics-port.
- Per-job and per-thread hash-rate gauges:
- Prefer remove-on-end semantics to avoid scrape-timing artifacts.
- Per-thread EMAs are summed to derive per-job hash rate.
- When changing metrics:
- Update prompt/response docs with metric names and label sets.
- Consider dashboards and alerting downstream.
Style and lint conventions
- Use inline format args in log macros, e.g., "value = {value}".
- Prefer saturating arithmetic for U512 ranges; avoid panics on partitioning.
- Keep tests deterministic; avoid flakiness due to implicit ranges or 0 nonces.
- Apply rustfmt and follow Clippy guidance unless it harms readability/intent.
- TOML is formatted via Taplo per taplo.toml (tabs, key ordering for workspace.dependencies, etc.).
Change management notes
- Avoid renaming CLI options unless clearly justified; when renaming, provide an alias window and document migration in the response doc.
- When adding features (e.g., new engines), default to conservative runtime behavior and explicit logs on unsupported paths.
- For system-level advice (affinity, priorities), document briefly in README and/or example files (e.g., examples/systemd), and link from the PR description if relevant.
FAQ
Q: Why do we keep Cpu-prefixed names in the CLI?
- To prevent ambiguity once GPU engines land and to keep the mental model consistent. We suppress the enum_variant_names lint locally in the CLI.
Q: Why stable and not nightly?
- Consistency and CI portability. Stable + Clippy + Taplo cover our current needs.
Q: Why remove series on job end rather than setting 0?
- To avoid scrape-timing artifacts and enable cleaner aggregations. This is intentional and should be maintained.
Q: Where do I put process/authoring details?
- In this file (/agents.md). This is the single reference for authoring and process guidance.
Quick links
- Process guide (this file): /agents.md
- Workspace README: /README.md