Skip to content

v2 self host (full) - #5052

Merged
briansrls merged 2 commits into
mainfrom
session/proud-lark-378
Jun 17, 2026
Merged

v2 self host (full)#5052
briansrls merged 2 commits into
mainfrom
session/proud-lark-378

Conversation

@gunbai-bot

@gunbai-bot gunbai-bot Bot commented Jun 17, 2026

Copy link
Copy Markdown
Contributor

Auto-opened by session-dashboard for session proud-lark-378.
Pushing to session/proud-lark-378 advances this PR.

Worker attestation

Before flipping this PR to ready for review, confirm each item:

  • Title describes the change (not the session id or branch).
  • PR body summarises what and why (replace the TODO below).
  • Tests run: name the command (e.g. npm test, cargo test) and the result.
  • If this closes a work item, the body contains a Closes #N directive.
  • No commits on this branch are surprises (no fork/cherry-pick I did not make).
  • No secrets / credentials / large binaries staged.

Summary

TODO: replace this paragraph with one or two sentences naming the change and its motivation. Reviewers read this first.

Test plan

  • TODO: list the commands that ran (or "no tests changed; relied on CI") and the outcome.

briansrls and others added 2 commits June 17, 2026 02:16
@briansrls
WIP: v2 self host (full)
ae906cd
@briansrls @claude
Followup to #5050: clarify self-host test discovery + corpus-removal …
fbf11b4 
…comments

Addresses review comments on #5050 (no behavior change — comments only):
- fixed_point_test.dag header: state explicitly that `test fn` discovery scans
  the whole --source-root (src/v2) recursively, so the file is found at its
  compiler/self_host/ path; --scan-dir only governs the unified_claim_* scan.
  (Rebuts the 'outside scan dir → zero coverage' reading: the gate run shows all
  5 witnesses PASS within the 170-witness corpus.)
- claim_witness_corpus_ci_runner.dag removal comment: the deleted self-host rows
  were never live (their src/v2/test/claim/self_host/... entry paths don't exist);
  real coverage moved to the executed test fn. Stop over-claiming the whole runner
  is dead (its dsl/test/claim/... rows point at files that do exist).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
@briansrls
briansrls marked this pull request as ready for review June 17, 2026 02:48
@briansrls
briansrls merged commit ce2a9cc into main Jun 17, 2026
2 checks passed
@briansrls
briansrls deleted the session/proud-lark-378 branch June 17, 2026 03:01
briansrls added a commit that referenced this pull request Jun 17, 2026
@briansrls @claude
Merge origin/main into session/jolly-bear-882
52b37e9 
Integrate main's v2 self-host (#5052), folder-based .dag naming (#5051), and
cruft sweep (#5053). Conflict resolutions:
- ci_claim_gate (modify/delete): keep the deletion; #5051's filename-hygiene
  check (no __ in .dag basenames) is ported into claim_batch's discovery path,
  so the witness floor still enforces it.
- DESIGN.md: CI floor line points at claim_batch + the gunbc rust gate.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
briansrls added a commit that referenced this pull request Jun 17, 2026
@briansrls @claude
@gunbai-bot @cursoragent
Remove all bespoke Rust CI tooling: claim_batch + gunbc-run .dag gate…
9f1fe60 
…s (deletes tools/) (#5048)

* WIP: v1 -> v2 bootstrap/compile

* fix(bootstrap): restore v1 self-host fixed point (register method_template_projection_source)

`regen_stage0 --verify` was stale at HEAD: a fresh self-compile dropped
`pub mod method_template_projection_source;` from the generated lib.rs, so the
committed stage0 seed could no longer reproduce itself (and a real regen would
orphan the module + break the build, since regen_stage0.rs and tests/bootstrap.rs
reference it).

Root cause: #5025 added the hand-maintained method_template_projection_source.rs;
#5028 hand-added its `pub mod` to the generated lib.rs but never registered the
module in regen's authority, so fresh regen omitted it (a 7th hand-maintained
stage0 file outside both registries).

Fix (mirrors v1_compiler_dag_collect handling):
- register method_template_projection_source.rs in HAND_MAINTAINED_STAGE0_FILES
  so regen copies it and the output-set assertion accepts it
- ensure_method_template_projection_module() injects its `pub mod` into the
  generated lib.rs (the .dag has no source for this module — it holds
  GENERATED_METHOD_TEMPLATE_PROJECTION_DAG as a Rust const)
- sync tests/bootstrap.rs copy_stage0_support_modules to the full hand-maintained
  set (it was missing resolved_graph_cache + v1_compiler_dag_collect)

regen-safe: no .dag change, no generated stage0 file changes (committed lib.rs
already declared the module; this makes regen reproduce it). Validated:
`regen_stage0 --verify` green, regen write produces zero drift, v1 claim gate
still green (158 witnesses).

Gating of this fixed point is provided by PR #5026 (executing .dag bootstrap
claim); this PR is its prerequisite post-rename.

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

* WIP: v1 -> v2 bootstrap/compile

* Clear two clippy lints masked behind the never_loop error

- src/v1/tests/src/lib.rs: remove duplicated #[cfg(test)] on mod pd3_adversarial (clippy::duplicated_attributes)
- tools/ci_claim_gate/src/main.rs: fix doc-comment list markdown (clippy::doc_lazy_continuation); wording unchanged

clippy --all-targets -- -D warnings now clean workspace-wide.

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

* Resolve merge with main #5038: keep the fail-closed rev-scan fix

main's #5038 silenced clippy::never_loop by reducing parse_compiled_receipt to
lines().next_back(), which only inspects the literal last line and preserves the
original fail-closed bug (trailing output after the compiled: line -> false reject).
This keeps the rev-scan that actually fixes it, guarded by the discriminating test
parse_compiled_receipt_skips_non_matching_trailing_lines.

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

* WIP: v1 -> v2 bootstrap/compile

* Park axiom slice: remove broken prose axiom model (idea deferred)

The prose statement: String axiom model was inert (a human restating
DESIGN.md, not mechanically checked) and had unresolved v2 type errors
(List not matchable with Empty/Cons; generic-lambda params not inferring
field access). Operator parked the axiom-modeling idea; removing the
slice rather than shipping cringey/broken code into std/. The mechanical
reframe (time->cost.dag, safety->Risk dim with Diagnostic as values) is
saved for a future clean take.

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

* WIP: v1 -> v2 bootstrap/compile

* WIP: v1 -> v2 bootstrap/compile

* WIP: v1 -> v2 bootstrap/compile

* WIP: v1 -> v2 bootstrap/compile

* WIP: v1 -> v2 bootstrap/compile

* Slice 2: wire clippy/fmt CI gates (rust_gate_runner) + fix the latent stage0_core breakage they caught

What ships:
- rust_gate_runner (tools/): realizes the v2.workflow.rust_stage0_gates .dag
  selector. Per changed path it calls path_is_rust_monolith (String->Bool, in
  the v1 interpreter) — the .dag is the authority for WHICH gates run, the binary
  is the realizer. On a Rust-touching diff it runs cargo fmt --all --check and
  cargo clippy --all-targets -- -D warnings, fail-closed (undeterminable diff =>
  run the gates). The in-process witness gate stays the always-on floor; this adds
  the two Rust-only dependents. A .dag-only diff selects neither.
- ci.yml: adds the runner step (affected-set via origin/main...HEAD); fetch-depth
  1->0 so the merge-base resolves (runner fail-closes if absent). SCAFFOLD,
  dissolves when the Rust seed reaches zero (DESIGN §7).

Latent breakages the clippy gate caught on first run (both pre-existing on main,
uncaught because CI only ran the witness gate):
- v1-stage0-core did not compile (E0432): the generated CoreCrate module list
  omitted std_emit_model, though the shared emit files import it. Fixed at the
  authority (stage0_crates.dag) + regen_stage0; 'regen --verify' confirms the seed
  is at a fixpoint. This makes 'cargo check --workspace' green again.
- doc_lazy_continuation lint in src/v1/tests/src/pipeline.rs (same class as #5045).

Proven by execution: witness gate 176 green; rust_gate_runner exit 0 (fmt+clippy
both pass) on a .rs path; exit 1 on a planted fmt violation; skips on .dag-only;
git-mode plan resolves the diff vs origin/main.

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

* WIP: v1 -> v2 bootstrap/compile

* WIP: v1 -> v2 bootstrap/compile

* WIP: v1 -> v2 bootstrap/compile

* WIP: v1 -> v2 bootstrap/compile

* WIP: v1 -> v2 bootstrap/compile

* WIP: v1 -> v2 bootstrap/compile

* Merge origin/main into session/jolly-bear-882

Integrate main's v2 self-host (#5052), folder-based .dag naming (#5051), and
cruft sweep (#5053). Conflict resolutions:
- ci_claim_gate (modify/delete): keep the deletion; #5051's filename-hygiene
  check (no __ in .dag basenames) is ported into claim_batch's discovery path,
  so the witness floor still enforces it.
- DESIGN.md: CI floor line points at claim_batch + the gunbc rust gate.

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

* fmt: reflow claim_batch import block after merge

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

* Replace emit_host_runner with gunbc-run emit_host_gate.dag (#5058)

* WIP: Replace emit_host_runner (bespoke tools/ Rust) with a gunbc-run .dag gat

* Replace emit_host_runner with gunbc-run emit_host_gate and delete the crate.

Host MVP smoke rows (rust/python/go/typescript) now run via dsl/tools/emit_host_gate.dag
with POSIX shell transports; CI witnesses the gate by execution. emit_host.dag transport
rows stay fail-closed in src/v2 (no host effects there) until T-22 Realization lands.

Co-authored-by: Cursor <cursoragent@cursor.com>

* fix(emit-host-gate): bootstrap portable Go when absent from CI PATH.

Self-hosted runners have cargo/rust but not always `go`; download the
arch-matched toolchain tarball before the MVP smoke row runs.

Co-authored-by: Cursor <cursoragent@cursor.com>

---------

Co-authored-by: Brian Searls <briansrls@gmail.com>
Co-authored-by: Cursor <cursoragent@cursor.com>

* WIP: v1 -> v2 bootstrap/compile

* WIP: v1 -> v2 bootstrap/compile

* WIP: v1 -> v2 bootstrap/compile

* fmt: reformat interp_dry_run_test.rs (pre-existing violation caught by the new rust gate)

The #5048 rust gate (cargo fmt --all --check) is the first CI fmt enforcement;
this test file accumulated a violation on main before any gate existed. Reformat
so the gate lands green.

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

* WIP: v1 -> v2 bootstrap/compile

---------

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-authored-by: gunbai-bot[bot] <289086189+gunbai-bot[bot]@users.noreply.github.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@briansrls