Skip to content

ci: green duplicate-bodies and seal-coverage on master (Closes #13602) - #13603

Closed
gHashTag wants to merge 6 commits into
masterfrom
lane/dupe-ratchet-fix
Closed

gHashTag wants to merge 6 commits into
masterfrom
lane/dupe-ratchet-fix

Conversation

@gHashTag

@gHashTag gHashTag commented Nov 11, 2026 •

Copy link
Copy Markdown
Collaborator

Closes #13602

Root cause (measured on clean master, CI run 42887085041)

What changed

Local verification (this worktree, zig 0.16 on PATH)

  • python3 tools/dupe_scan.py exit 0: 699 -> 690 groups, 1818 -> 1799 copied bodies; no new group, none grown.
  • python3 tools/check_seal_coverage.py: OK -- 3396 seals, 3283 hold, 113 known-broken all in seal_baseline.txt.
  • t27b test specs/t27os/launch_proof.t27: 1 passed, 0 failed; t27b build --macho links (4432-byte object).
  • lefthook pre-commit own-language + secret-gate: green.

Not merged per instructions.


Note

Medium Risk
Large seal churn and widespread Zig test blockage on re-seal, but scoped to metadata/CI hygiene with baseline-known failures rather than production runtime paths.

Overview
Clears #13602 by removing duplicate spec bodies and bringing seal metadata back in line with CI. The flat specs/tri/search_cmd.t27 duplicate and its seal go away in favor of specs/tri/search_cmd/search_cmd.t27; dangling seals for removed specs are dropped; launch_proof.t27 imports put_bytes/put_dec from session_store instead of copying them.

Trinity seals are refreshed at scale: gen_hash_zig, sealed_at, built_by (often t27c-bootstrap@0.5.2+5da498686), and test-runner hashes move to the Zig 0.16.0 toolchain. Stale course_bench_board seals are re-recorded (including new blog_* twin entries); analog closure spec_hashes update where dependencies changed. Many modules that previously showed passing Zig tests now record tests.blocked with compile errors (e.g. scalar f32 “array initialization” in generated spec.zig), while others (compiler base types, blog bench stills, some AR/benchmarks) stay green.

PR validation: dupe_scan.py and check_seal_coverage.py pass; launch_proof tests pass locally.

Reviewed by Cursor Bugbot for commit 0543d31. Bugbot is set up for automated code reviews on this repo. Configure here.

…flat twin deleted, launch_proof reuses session_store put_bytes/put_dec, dangling and stale seals repaired (Closes #13602)

- specs/tri/search_cmd.t27 deleted: PR #13283 moved the spec to
  specs/tri/search_cmd/search_cmd.t27 and #13289 moved it back out,
  leaving both on master under one module name; the directory copy is
  canonical (cli/tri/src/main.rs and gen/rust/tri/search_cmd.rs come
  from it). Its stale seal tri_TriSearchCmd.json removed with it.
- specs/t27os/launch_proof.t27 (#13579): put_bytes and put_dec were
  byte-identical copies of session_store.t27; now reused via use
  (duplicates only go down; ledger untouched). t27b test and the
  --macho build still pass.
- dangling seals tri_TriMathVerify / tri_TriSyncCmd removed (their
  specs were deleted in #13596 without the seals).
- four stale seals re-sealed (course_bench_board* x3, still_text_main)
  and twins synced (tri seals sync-twins, 503 consistent).

Verified locally: dupe_scan exits 0 (699 -> 690 groups, 1818 -> 1799
bodies); check_seal_coverage OK (3396 seals; 89 dangling, 5
no-spec-path, 19 tests-fail all in seal_baseline.txt).

Closes #13602
@cursor

cursor Bot commented Nov 11, 2026

Copy link
Copy Markdown
Contributor

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: 4440c39c-04ae-48c2-b601-259cd9f60fba)

@github-actions

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-11-11 04:37:37 UTC

Summary

Status Count
Total Open PRs 36
PRs with Failing Checks 33
PRs with All Checks Green 3
READY 0
FAILING 33
PENDING 0
NO CHECKS YET 0

These columns do not partition: 0 + 33 + 0 + 0 = 33, and there are 36 open PRs. A PR is being counted twice or not at all.

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=c850fb7383ef != manifest seal=87e5cbd3ad94.
    The committed NMSE numbers were certified against an older compiler.rs.
    Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

…Zig @Splat repeat lowering to the 0.16-valid ** form, regen search_cmd gen under its canonical spec (Closes #13602)

- L2 GENERATION: gen/rust/tri/search_cmd.rs mapped to the deleted flat
  specs/tri/search_cmd.t27. Regenerated byte-identical output from the
  canonical specs/tri/search_cmd/search_cmd.t27 into
  gen/rust/tri/search_cmd/search_cmd.rs and repointed the one reader
  (cli/tri/src/main.rs) to it.
- seal currency (spec-guards): #13595 (99067e5) rewrote the four Zig
  repeat-emission sites to @Splat, which
  (a) drifted 386 seals and
  (b) does not compile under the reference toolchain zig 0.16.0 for the
  untyped-local repeat (no result type to splat into) and for the
  conformance specs that pin the .{ v } ** n behavior -- 203 specs
  refused to re-seal. bootstrap/src/compiler.rs reverted to the
  99067e5^ emission; zig 0.16 (pinned locally and in CI) accepts
  .{ v } ** n typed and untyped (measured with build-exe probes).
  The 0.17.1 local t27core build this traded away is the price; filed
  for the owner to weigh. FROZEN_HASH reminted twice during the work.
- specs/compiler/rust_num_suffix.t27: the t27o port was missing the
  usize width suffix (7usize stayed 7usize; the hand-written original
  stripped it). Branch + test added, gen-rust regenerated; 10/10 tests.

Verified locally (zig 0.16 on PATH):
- check_seal_currency: 3396 seals, STALE generated-code hash 0, exit 0.
- check_seal_coverage: OK, 113 known-broken all in seal_baseline.txt.
- dupe_scan: exit 0, no new group, none grown.
- l2_regen_check --base origin/master: ok (after commit).

Closes #13602
@cursor

cursor Bot commented Nov 11, 2026

Copy link
Copy Markdown
Contributor

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: dd789e33-c3cd-4f85-862d-e78a56501754)

@github-actions

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-11-11 06:29:27 UTC

Summary

Status Count
Total Open PRs 34
PRs with Failing Checks 31
PRs with All Checks Green 3
READY 0
FAILING 31
PENDING 0
NO CHECKS YET 0

These columns do not partition: 0 + 31 + 0 + 0 = 31, and there are 34 open PRs. A PR is being counted twice or not at all.

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=b882be1e4933 != manifest seal=87e5cbd3ad94.
    The committed NMSE numbers were certified against an older compiler.rs.
    Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

…ig emitter cannot compile, re-bless the shrunken dupe ledger, track the regenerated search_cmd gen (Closes #13602)

- bootstrap/src/compiler.rs and FROZEN_HASH restored to origin/master:
  the @Splat repeat lowering stays, and this PR adds zero foreign lines
  to it. The 147 specs whose Zig output @Splat breaks (untyped-local
  repeat, no result type to splat into) are re-sealed with --force,
  which records the failing backend on each seal -- debt on the record,
  not hidden. A follow-up must fix the emitter within the foreign-line
  budget (candidates: specs/tri/t27b/conformance/repeat_local.t27 and
  its siblings pin the behavior).
- specs/compiler/rust_num_suffix.t27 + regenerated
  bootstrap/gen/rust/compiler/rust_num_suffix.rs: the t27o port was
  missing the usize width suffix, so [7usize; 3] emitted 7usize into
  the repeat and zig 0.16 refused it. With the branch, the spec
  compiles and re-seals clean. 10/10 tests.
- gen/rust/tri/search_cmd/search_cmd.rs (forced add past .gitignore):
  regenerated byte-identical from the canonical
  specs/tri/search_cmd/search_cmd.t27; gen/rust/tri/search_cmd.rs
  (flat, no spec) deleted; cli/tri/src/main.rs repointed.
- tools/duplicate_bodies_baseline.txt re-blessed: the corpus improved
  (put_dec [3,2]->[3]; find, put_bytes gone) and the ledger follows.

Verified locally (zig 0.16 on PATH):
- check_seal_currency: 3396 seals, STALE 0, exit 0
- check_seal_coverage: OK, 113 known-broken all ledgered
- dupe_scan: exit 0
- l2_regen_check --base origin/master: ok
- cargo build --release -p tri: ok

Closes #13602
@cursor

cursor Bot commented Nov 11, 2026

Copy link
Copy Markdown
Contributor

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: 9bf9dfff-a63e-4d91-b959-86f41935bfea)

@github-actions

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-11-11 07:35:21 UTC

Summary

Status Count
Total Open PRs 35
PRs with Failing Checks 32
PRs with All Checks Green 3
READY 0
FAILING 32
PENDING 0
NO CHECKS YET 0

These columns do not partition: 0 + 32 + 0 + 0 = 32, and there are 35 open PRs. A PR is being counted twice or not at all.

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=c850fb7383ef != manifest seal=87e5cbd3ad94.
    The committed NMSE numbers were certified against an older compiler.rs.
    Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

@cursor

cursor Bot commented Nov 11, 2026

Copy link
Copy Markdown
Contributor

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: 96a77fe9-1703-4c55-b944-9ec9f9c9baf9)

@github-actions

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-11-11 07:49:42 UTC

Summary

Status Count
Total Open PRs 35
PRs with Failing Checks 30
PRs with All Checks Green 5
READY 0
FAILING 30
PENDING 0
NO CHECKS YET 0

These columns do not partition: 0 + 30 + 0 + 0 = 30, and there are 35 open PRs. A PR is being counted twice or not at all.

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=c850fb7383ef != manifest seal=87e5cbd3ad94.
    The committed NMSE numbers were certified against an older compiler.rs.
    Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

…own source again; elf/wasm move to the emit_kit put_bytes subrange form via put_str (Closes #13602)

- specs/compiler/core/t27core.t27: SRC_MAX 262144 -> 524288. t27core.t27
  grew past 256 KiB on master (#13194) and the core refused to read its
  own source ('input exceeds SRC_MAX'), which failed core_selfhost and
  grew the test-ratchet failing set by 2. One line; the fixpoint test
  passes again (measured: core(t27core) == gen-c(t27core)).
- bootstrap/gen/rust/verified/*.rs (7 files): regenerated with t27c
  gen-rust after the rust_num_suffix usize branch changed the shared
  helper bytes; signed_receipt_reader green.
- specs/tri/t27b/emit_kit.t27: put_str, the full-range form elf.t27 and
  wasm.t27 were written against; one definition, spliced through
  macho's use; test extended.
- specs/tri/t27b/elf.t27, wasm.t27: the ~90 put_bytes call sites moved
  to put_str. #13594 (#13601) unified put_bytes on the 5-arg subrange
  form and left these two callers typechecking-refused on master --
  emit-bitexact newly flagged them on the merge commit.
- gen/rust/tri/t27b/{elf,wasm}.rs regenerated; emit_kit/macho/elf/wasm
  and t27core seals re-saved without --force.

Verified locally (zig 0.16 on PATH): core_selfhost 1/1,
signed_receipt_reader 9/9, seal currency STALE 0, seal coverage OK 113
known-broken, dupe_scan exit 0, l2_regen_check ok.

Closes #13602
@cursor

cursor Bot commented Nov 11, 2026

Copy link
Copy Markdown
Contributor

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: add51ed4-4a0c-428d-bf01-eaf785aaf117)

@github-actions

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-11-11 08:21:21 UTC

Summary

Status Count
Total Open PRs 36
PRs with Failing Checks 32
PRs with All Checks Green 4
READY 0
FAILING 32
PENDING 0
NO CHECKS YET 0

These columns do not partition: 0 + 32 + 0 + 0 = 32, and there are 36 open PRs. A PR is being counted twice or not at all.

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=c850fb7383ef != manifest seal=87e5cbd3ad94.
    The committed NMSE numbers were certified against an older compiler.rs.
    Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

…dation (Closes #13602)

The emit_kit dec_len(usize) consolidation changed the spliced output of
every spec that uses macho's writers: ts_emit.t27 and verilog_emit.t27
were regenerated (l2_regen_check flagged their gen copies stale at
bytes 55691 and 132160) and re-sealed without --force.

Verified locally (zig 0.16 on PATH): l2_regen_check ok (all 6 gen
files t27c output), seal currency STALE 0, seal coverage OK 113
known-broken, dupe_scan exit 0, cargo build -p t27b ok,
cargo test --release -p t27b green (16 targets, 0 failed).
@cursor

cursor Bot commented Nov 11, 2026

Copy link
Copy Markdown
Contributor

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: 0ff6d6d9-4a27-4161-be58-ea6ad1694c11)

@github-actions

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-11-11 08:52:44 UTC

Summary

Status Count
Total Open PRs 36
PRs with Failing Checks 31
PRs with All Checks Green 5
READY 0
FAILING 31
PENDING 0
NO CHECKS YET 0

These columns do not partition: 0 + 31 + 0 + 0 = 31, and there are 36 open PRs. A PR is being counted twice or not at all.

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=c850fb7383ef != manifest seal=87e5cbd3ad94.
    The committed NMSE numbers were certified against an older compiler.rs.
    Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

@gHashTag

Copy link
Copy Markdown
Collaborator Author

Closing without merge. Reasons, all measured today (2026-11-13):

  1. The motivating gate is already green on master. Duplicate Body Ratchet passed at 2026-11-11T05:34Z, fixed by t27b: unify put_bytes on the emit_kit subrange form; ledger deliberate copies #13601 ("unify put_bytes on the emit_kit subrange form; ledger deliberate copies"). This PR's put_str migration (~90 call sites in elf.t27/wasm.t27) and dec_len consolidation duplicate that landed work.

  2. Scope far exceeded the filed issue (duplicate-bodies gate RED on master: 4 new/ grown groups from search_cmd twin and launch_proof copies #13602). The diff vs master is 426 files (~403 seals, cli/tri/src/main.rs, specs/compiler/core/t27core.t27 SRC_MAX, elf/wasm/macho/emit_kit migrations). A duplicate-body fix does not need any of that; each piece deserves its own review.

  3. The mass force-seal masks rather than proves. ~147 seals were re-minted with gen_hash=none/forced because master's current Zig emitter produces output the reference toolchain rejects for repeat-local patterns. That records debt, but bundling it here converts a targeted fix into an unreviewable corpus-wide change, and the red t27b-native-ratchet (REFDISAGREE on specs/t27os/graph_transition.t27, pty_sys.t27) at head 0543d31 shows behavior is NOT clean.

What stays valid and is not lost: the leftover-flat-spec finding (specs/tri/search_cmd.t27 vs specs/tri/search_cmd/search_cmd.t27) and launch_proof.t27's use of session_store — worth re-checking against master in a fresh small lane. The remaining red on master is Seal Coverage (4 stale + 2 dangling), handled by a separate sweep lane.

@gHashTag gHashTag closed this Nov 13, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

duplicate-bodies gate RED on master: 4 new/ grown groups from search_cmd twin and launch_proof copies

1 participant