refactor(#96)!: K0 — convert to workspace + keri-rs sibling crate by joeldsouzax · Pull Request #126 · devrandom-labs/cesr

joeldsouzax · 2026-07-05T20:24:41Z

Closes #96 (K0 · KERI sans-io core). Converts the single-crate repo into a two-crate Cargo workspace so the KERI sans-io core (K1–K9) can develop in a sibling crate that versions independently — protecting cesr-rs's frozen surface (STABILITY.md #86) from KERI-state churn.

BREAKING (repo layout only): the cesr source moved from repo root into cesr/. The published cesr-rs crate surface is byte-for-byte unchanged (verified: spec review hashed every moved file).

What changed

cesr/ — the entire crate, moved via git mv (100% renames, history preserved). Now inherits [workspace.package]/[workspace.lints]; only manifest metadata changed.
keri/ — new keri-rs (lib name keri), an empty K0 skeleton: no_std, std optional (alloc is a floor), public-API-only dep on cesr, and a link-test proving it binds cesr's public surface. No KERI logic (that's K1+).
Root Cargo.toml — pure virtual [workspace] (members = ["cesr","keri"], resolver = "3"), with nexus-style [workspace.package]/[workspace.dependencies]/[workspace.lints]/[workspace.metadata.crane]. Top-level layout (not nexus's crates/), no workspace-hack.
flake.nix — whole-workspace clippy; wasm/nostd build both crates by package; new cesr-keri-boundary check.
release-plz — keri-rs package added (independent versioning).
Scheduled/release workflows — repointed to cesr/ (see below).

The API boundary is (mostly) free

Rust's per-crate visibility already forbids keri from reaching cesr's non-pub items — it won't compile. The only back-door is enabling cesr's internals/test-utils features, so cesr-keri-boundary is a one-line grep. Bug-probe verified it fails when internals is injected and passes after revert.

Modeled on nexus

Adopted [workspace.dependencies]/[workspace.package]/[workspace.lints]/[workspace.metadata.crane] and whole-workspace clippy+nextest. Diverged deliberately: top-level layout, no workspace-hack.

Prerequisite already done

keri-rs@0.0.1 is reserved + owned on crates.io (via #97's reserve-crate workflow).

Test plan

nix flake check GREEN — all checks across BOTH crates (clippy --workspace, nextest 1777 tests, wasm, no_std, doctest, fuzz-replay, and cesr-keri-boundary).
cargo publish --dry-run clean for both cesr-rs and keri-rs (independent releases).
Source integrity: every moved file byte-identical (blob-hash verified); lint tables verbatim (no relaxation).
Boundary gate bug-probe: fails on internals, passes on revert.
Stale-path sweep of scheduled/release workflows (the local gate can't see these):
- release-plz.yml paths: → cesr/** + keri/** (was src/**, which would have stalled releases)
- keripy-diff.yml / keripy-sync.yml → repointed to cesr/ (were false-green / bogus-report)

Review discipline caught (each invisible to the layer below)

resolver = "3" (edition-2024 default) — code-quality review
keri alloc-floor + dropped premature pub use cesr (semver coupling) — code-quality review
three stale-path CI bugs in scheduled workflows — final holistic review

Design + plan: docs/superpowers/specs/2026-07-05-96-... and docs/superpowers/plans/2026-07-05-96-....

🤖 Generated with Claude Code

Symmetric cesr/ + keri/ workspace; keri-rs already reserved (owned). K0 is pure infra: git-mv cesr into cesr/, empty keri crate under the god-level harness, workspace-shared lints, flake covers both, compiler+grep API gate. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…ayout) Adopt nexus's [workspace.dependencies]/[workspace.package]/[workspace.lints]/ [workspace.metadata.crane] and whole-workspace clippy+nextest. Keep the top-level cesr/ + keri/ layout (not nexus's crates/); no workspace-hack. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Four tasks: workspace + git-mv cesr into cesr/; empty keri skeleton with public-API link test; flake covers both (per-crate wasm/nostd); release-plz + boundary grep-gate + docs. Placeholder-free; verified public API paths. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Root becomes a virtual [workspace] manifest (members = ["cesr"]) with workspace-shared package/deps/lints/crane metadata (nexus patterns, top-level layout). cesr's source moves via git mv with a byte-identical public surface; lints inherited via [lints] workspace = true. Fuzz path deps re-pointed to ../cesr. BREAKING: repo layout only — the cesr-rs crate surface is unchanged. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

resolver 3 is the edition-2024 default (MSRV-aware) — preserves standalone cesr's resolution. The git-mv of README into cesr/ left 7 relative links pointing at root-level files; prepend ../ so they resolve. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

keri-rs (lib name keri) joins the workspace as an infrastructure-only skeleton: no_std scaffolding, std/alloc features forwarding to cesr, a public re-export of cesr, and a test proving it links cesr's PUBLIC API. No KERI logic (that is K1+). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

cesr's core feature requires alloc unconditionally, so keri has no no-alloc mode: request core+alloc from the dep and drop keri's misleading optional alloc feature (only std stays optional). Remove pub use cesr — premature public API that coupled keri-rs's semver to cesr-rs churn; the link test resolves cesr via the extern prelude, not the re-export. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…asm/nostd Whole-workspace clippy; wasm/nostd checks build both cesr-rs and keri-rs by package (--features is illegal at a virtual root). keri now rides the full gate. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…I boundary release-plz gains the keri-rs package (independent versioning). A flake check fails if keri/Cargo.toml enables cesr's internals/test-utils features (the only back-door past the compiler's per-crate visibility). CLAUDE.md notes the workspace. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

The workspace move left three workflows pointing at the old root layout — invisible to nix flake check, they'd fail silently in CI: - release-plz paths: src/** never matched cesr/src/** → releases stalled; now cesr/** + keri/**. - keripy-diff regenerated corpus into root tests/ (false-green parity vs the real cesr/tests/corpus/keripy) → both paths now cesr/tests/corpus/keripy. - keripy-sync --cesr . skipped all relocated files (bogus zero-codes report) → --cesr ./cesr. Also hash cesr/keri Cargo.toml in the nix cache keys. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

codspeed-hq · 2026-07-05T20:26:54Z

Merging this PR will create unknown performance changes

🆕 21 new benchmarks
⏩ 21 skipped benchmarks¹

Performance Changes

	Benchmark	`BASE`	`HEAD`	Efficiency
🆕	`base64_decode[1024B_attachment]`	N/A	6 µs	N/A
🆕	`base64_decode[32B_key_digest]`	N/A	1.6 µs	N/A
🆕	`base64_decode[64B_signature]`	N/A	1.6 µs	N/A
🆕	`base64_encode[1024B_attachment]`	N/A	4.8 µs	N/A
🆕	`base64_encode[32B_key_digest]`	N/A	1 µs	N/A
🆕	`base64_encode[64B_signature]`	N/A	1.1 µs	N/A
🆕	`v1_auto_big`	N/A	2.9 µs	N/A
🆕	`v1_small`	N/A	2.9 µs	N/A
🆕	`controller_idx_sigs_1sig`	N/A	4.1 µs	N/A
🆕	`qb2_to_qb64`	N/A	1.6 µs	N/A
🆕	`qb64_to_qb2`	N/A	1.7 µs	N/A
🆕	`blake3_256_fixed`	N/A	4.6 µs	N/A
🆕	`ed25519n_fixed`	N/A	4.6 µs	N/A
🆕	`strb64_l0_variable`	N/A	4.2 µs	N/A
🆕	`blake3_256_to_qb64`	N/A	3.5 µs	N/A
🆕	`ed25519n_to_qb64`	N/A	3.5 µs	N/A
🆕	`multi_group_controller_witness`	N/A	5.7 µs	N/A
🆕	`stream_parse_scaling[1]`	N/A	4.9 µs	N/A
🆕	`stream_parse_scaling[16]`	N/A	15.3 µs	N/A
🆕	`stream_parse_scaling[256]`	N/A	178.3 µs	N/A
...	...	...	...	...

ℹ️ Only the first 20 benchmarks are displayed. Go to the app to view all benchmarks.

_{Comparing feat-96-workspace-conversion-keri (a9230fa) with main (c8c3e1f)}

21 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

joeldsouzax and others added 13 commits July 5, 2026 20:15

docs(#96): plan uses resolver=3 (edition-2024 default)

ace9535

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

docs(#96): sync plan to keri alloc-floor + no re-export

e360357

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

docs(#96): correct plan note on -p (per-crate features, not illegality)

a9230fa

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

joeldsouzax merged commit 8faf304 into main Jul 5, 2026
6 checks passed

joeldsouzax deleted the feat-96-workspace-conversion-keri branch July 5, 2026 20:33

devrandom-release-plz Bot mentioned this pull request Jul 5, 2026

chore: release #127

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

refactor(#96)!: K0 — convert to workspace + keri-rs sibling crate#126

refactor(#96)!: K0 — convert to workspace + keri-rs sibling crate#126
joeldsouzax merged 13 commits into
mainfrom
feat-96-workspace-conversion-keri

joeldsouzax commented Jul 5, 2026

Uh oh!

codspeed-hq Bot commented Jul 5, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Uh oh!

Conversation

joeldsouzax commented Jul 5, 2026

What changed

The API boundary is (mostly) free

Modeled on nexus

Prerequisite already done

Test plan

Review discipline caught (each invisible to the layer below)

Uh oh!

codspeed-hq Bot commented Jul 5, 2026

Merging this PR will create unknown performance changes

Performance Changes

Footnotes

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant