Skip to content

RB-24: the driver as language-1.2 source (#271) - #419

Merged
humuhumu33 merged 20 commits into
release/v0.4from
claude/issue-271-rb24-driver
Nov 13, 2026
Merged

humuhumu33 merged 20 commits into
release/v0.4from
claude/issue-271-rb24-driver

Conversation

@humuhumu33

@humuhumu33 humuhumu33 commented Nov 3, 2026 •

Copy link
Copy Markdown
Contributor

Part of #271: the RB-24 command-root driver as language-1.2 source. This delivers the registered driver behavior and partially delivers #426; #426 remains open for the direct RuntimeHost compiler/pinned-Lean evidence described below. #271 and #269 remain open for broader compiler integration.

The implementation adds compiler/src/RuntimeHost.lex.tex and the Rust host runtime/renderer support. RuntimeHost supplies seven RuntimeSyntax crates; the resulting driver is validated before printing. The command-driver fixture binds entry and main.rs provenance and exercises file/directory/lock and child-process operations, continuation, streams, status and registered refusals/plants. The renderer's language-1.2 semantics change and coordinated official products are included.

Current delivery evidence:

  • Final source head 0b77d7253e11f8ca33df204e7614adbc47907246 adopts producer 43311807929. The reviewed product delta contains 1,463 paths, with sampled config/input/compiler/semantic bindings; this is not an independent byte audit of every output.
  • The repaired BS-09 route covers 128 named plus 600 seeded vectors, 49 rules and eight plants, with exact native/evaluator/rendered-Rust correspondence and export bindings. The old 575/576 aggregate diagnosis is not a further release prerequisite; its specific historical causal explanation is not established.
  • Full CI43329099701 passed all 70 jobs and 772 steps, including same-run honesty. Exact-head reproducibility43329037236 passed. Canceled runs are excluded.
  • The external merge b8b6e86a94533d87dd89455d5a5bf9400bda96ba has ordered parents 26e5e0ace691ddf17fe3f8de30d1182e7890a193 and 0b77d7253e11f8ca33df204e7614adbc47907246. Its tree cb61de9eb5d6f58ff235faeebdd504e8ed99e409 exactly equals the qualified synthetic checkout. The branch-head tree differs by the inherited floating-point proposal.

Remaining #426 evidence: RB-24 checks RuntimeHost shipped/embedded byte equality and parses/decodes/validates its Rust data. Those checks, and real Lean verification of the command-root fixture, do not by themselves establish compiler checking and pinned-Lean verification of the actual RuntimeHost module. RuntimeHost is absent from the inspected compiler Main import and committed manifest closure. The existing owner should first link any retained exact-source-bound RuntimeHost result and map the two originally disclosed missing twins to their evidence. If that result does not exist, the owner must supply the smallest missing qualification; no blanket producer or completed-gate replay is requested.

This evidence summary was reconciled after the external landing. Earlier regeneration43209826758 and its timing/count narrative are historical preparation, not the final adoption above.

humuhumu33 and others added 2 commits November 3, 2026 12:13
A rust-std root of the type Option S -> Host::Event -> S * Host::Request
gets src/main.rs: a loop printed from the new language-1.2 module
RuntimeHost (seven RuntimeSyntax crates; the files and process groups
exist as arms that carry out and as arms that refuse, picked by the
effects the root admits). The manifest gains entry, the provenance gains
entry and the digest of src/main.rs, RuntimeSyntax gains cfgItem, the
closed table of the standard library grows from 51 to 121 names, and the
validator gains hidden-effect, cfg-condition, cfg-item and cfg-pair.

The driver text is generated from a reference file that was compiled,
linted under the package's lint set and run against a scripted step.
Not done: the Lean twins of the validator (BS-09) and of the package
renderer (BS-15), the formatting of the two new .lex.tex files, the
regeneration, and every cargo build and test.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…n's encoder

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
afflom pushed a commit that referenced this pull request Nov 4, 2026
…context

Port and integration provenance: RB-24 from PR #419 through b70b546, on the Host interface from PR #396. Final generated artifacts and the full acceptance gate remain pending in draft PR #439.

Co-authored-by: humuhumu33 <68450279+humuhumu33@users.noreply.github.com>
humuhumu33 and others added 5 commits November 9, 2026 00:53
RB-24's base, #389, landed; this takes the line at f1b70fe. The six
conflicts keep both sides: release/v0.4's text with RB-24's own edits
applied over it three-way, the register at 478 IDs (SPEC.md, the README
in the line's wording, the RP-07 check and its case), both CHANGELOG
entries, and both rows of plants in VERIFICATION.md.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The first compile and run of the driver found these, each fixed here:

- The SPEC §31 row of RB-24 had no sections cell, so RP-07 read 477
  rows; it cites §17.17, where its prose is.
- A unit test of package.rs built a Manifest without `entry`, and
  host.rs dereferenced a name clippy reads as auto-deref.
- The case validated each crate of RuntimeHost alone, but the base
  names the type of each effect's group (`Files`, `Procs`), so only a
  composition is closed. It now validates all eight compositions the
  renderer can make (two steps, files and process each on or off), each
  valid with exactly its effects and refused as hidden-effect without
  one; the cfg-pair, cfg-condition and cfg-item plants go into the full
  composition.
- Every function of RuntimeHost was `public := false`, which the
  validator refuses as function-visibility; they are public, in the
  module and its shipped copy alike.
- The driver hands `drive` to the thread it spawns as a value, which
  neither a variable nor a path could name; a path of one segment that
  names a function of the crate is now that function, and `drive` is
  written as one.
- `spawn` creates the files a child's output and errors are written to,
  so a root admitting `process` without `files` could not render. The
  HOST table now lists the effects that realize each name, and `File`
  is `files`' and `process`'s, as the spawn request names those files
  and a child has its parent's authority. SPEC.md says both.
- The fixture command joined strings with `append` (lists and bytes
  only), named the definition `bad` without applying it, held its state
  as a pair where a command's state is a program type of its own (now
  the structure `Session`), and stated no axiom policy where Lean
  observes propext, Quot.sound and Classical.choice; each is corrected.
- The case's check that a root without an effect names none of its
  host names matched `File` inside the driver's own type `Files`; it
  now matches whole names.

With these, conformance_rb_24 passes (167 s): every composition
validates, the fixture's step is a command, its package certifies,
exports, builds under cargo, and runs to its fixed result, and each
plant is detected. Gates: fmt, clippy -D warnings on the workspace,
validate-model, validate-spec-links (478).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The renderer's sources moved (the driver's host.rs, package.rs and the
validator), so `rust_renderer` is their new digest and the language-1.2
compiler-semantics ID moves, as the CHANGELOG entry states. Every RB
case but RB-07 passes on this tree (RB-01..RB-21, RB-24); RB-07 now
reaches the committed packages, whose provenance carries the previous
ID: the regeneration this member owes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The merge is clean, but both sides added one register row over 477, so
the count moves to 479 in SPEC.md, the README, the RP-07 check and its
case, and the RB-24 entry of the CHANGELOG, now the newest, states it.
The locks of the standard library, the bootstrap and compiler projects
and the two std models are rewritten: they follow the standard library's
new member and the compiler-semantics ID this member moves. CONFORMANCE.md
is written from the model.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…cs ID

`lexlean lock` on this tree writes the moved `compiler_semantics` digest
into each example's lock; nothing else in them changes.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
@humuhumu33
humuhumu33 changed the base branch from claude/issue-271-effects-boundary to release/v0.4 November 9, 2026 15:56
humuhumu33 and others added 6 commits November 12, 2026 10:05
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…mantics ID

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every fixture lock that is current on release/v0.4 is relocked (298); the
seven negative fixtures whose locks are stale by design stay as they are.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…moves

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Regenerated by regenerate.yml with UOR Atlas (run 43209826758): every
project's export, build oracles, fixtures, verify records (Atlas whole) and
model documents. No count of the bootstrap count cases moved, and every
recursive package keeps its stack row.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@humuhumu33
humuhumu33 marked this pull request as ready for review November 12, 2026 15:54

@afflom afflom left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Concrete final-gate blocker at bca709: run43252902597 job131309489680 (bdd-bootstrap-runtime), normative step7, fails conformance_bs_09 at runtime_validate.rs:736. Actual counts are (117,600,717,575); expected (117,600,717,576). Named crates, seeded mutants and total vectors agree, but one fewer vector was refused.

Please identify the changed vector and reconcile its result with the intended runtime/driver contract before changing any expected count. This log alone does not establish whether the implementation changed incorrectly or the expected count is stale; blindly rewriting576 to575 would conceal that distinction. Reuse the retained run and its uploaded gate evidence rather than rerunning the complete producer to diagnose this single discrepancy.

Windows print/moves also failed at lane-execution step5; their causes were not inferred. The overall full run is still active and cannot be accepted. The owner-reported regeneration including Atlas supersedes the earlier 'not run' status, but the bounded source/product comparison here was capped at300 files and is not complete product clearance. Current release26e5 also needs resulting-tree applicability review before landing. One previously unread runtime job log was archived once; no tests, writers, gate reruns or source patches were initiated.

afflom commented Nov 12, 2026

Copy link
Copy Markdown
Member

Source-only narrowing of review5878646388's BS09 blocker: the575-versus576 refusal assertion counts direct native rt_validate::validate verdicts. It fires before the lowered bootstrap/evaluator/Lean or rustc differential legs, so diagnosing this discrepancy does not require repeating those proof/producer stages.

Between465b and exactbca709, the BS09 case's named count117, seeded count600, seed0x0B55_0009 and count-generation logic are unchanged; its only case-file change is Cfg encoding that would panic if reached and therefore does not explain this completed assertion. This does not prove identical inputs: RuntimeSupport/RuntimeProbe, decoder and plant-helper inputs also bind the corpus. Meanwhile the native validator expands its type/function/method admission and function-value handling.

The narrow missing evidence is a changed vector's stable name/index, exact decoded AST/input identity and old/new refusal reason. Please establish those before deciding whether to repair admission or update a count. The net count difference alone does not identify a single changed vector or justify576→575. Existing archived log and exact source were reused; no new execution, instrumentation patch or log fetch was performed.

… AU-04 follow the driver

- RuntimeSemantics: findFree and findMethod skip a cfg item as they skip a
  module item (the runtime libraries carry none; only the driver's host
  module does), so SP-25's module verifies again.
- BS-09: cfgItem joins RtItem's constructors, so the seeded draw refuses 575
  of the 717 (was 576).
- tests/fixtures/command-driver: its source in the canonical authoring
  notation (AU-04).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@humuhumu33
humuhumu33 marked this pull request as draft November 12, 2026 20:55
humuhumu33 and others added 2 commits November 12, 2026 20:55
RuntimeValidate (BS-09's twin) now carries what RB-24 added to
rust::rt_validate::validate: the cfg item (cfg-condition, cfg-item, cfg-pair,
with the last function of a name standing for it), the host table refused as
hidden-effect at a use, a named type, a path and a call, the new types, calls
and methods, and a crate function named alone as a value. Its tables are one
inList test per value, which keeps validateRuntime's certificate A at
5,418,470 bytes, under the project's max_file_bytes. BS-09 adds eleven crates
for them: 128 named, 728 vectors, 584 refused, 49 rules.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The stack-bound step guessed a package's root as the camelCase of its name,
so a root named otherwise was not found (the IR slices renamed theirs to fit).
Each corpus's roots.json names the root a package exports; the step reads it
and keeps the guess only for a package it does not list.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

@afflom afflom left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Bounded readback at fb47544: this is a substantive repair, not just a new expected count. RuntimeValidate now follows the driver's cfg/host/table/function-value handling, RuntimeSemantics skips cfg items, and eleven named cfg/host cases extend BS09 to128 named +600 seeded vectors,728 total,584 refused and49 rules. The current body still describes the earlier regeneration; distinguish that evidence from this changed twin and its active producer43305086907.

Efficiency correction to the earlier count-diagnostic request: the existing case still compares every vector's exact native outcome/message against rendered Rust, and the normative evaluator's WITHIN=728 assertion requires all728 vectors. All eight plants, exact rule coverage and source/export bindings remain. A complete passing current differential plus final-head acceptance can discharge those obligations without a separate old/new diagnostic gate or replaying historical proofs. Missing per-vector historical evidence limits the specific explanation that adding cfgItem caused576→575; qualify that attribution or cite preserved evidence, but do not add an otherwise redundant release prerequisite merely to reconstruct the old aggregate.

The next delivery step stays with the existing owner: finish/adopt the applicable current producer outputs, address actual failures and qualify the resulting final tree with the complete coverage intact. Current draft CI43305100861 failed, repro43305100991 passed; no failure cause is inferred from metadata. No producer, runtime or old log was repeated by this review.

humuhumu33 and others added 2 commits November 12, 2026 21:35
With the driver's cases in RuntimeValidate, Lean's extraction of the bootstrap
project writes more than the 33,554,432 bytes a child process may produce
(LLS8002 in regeneration run 43305086907). Ilya approved 256 MiB on
2026-11-12; a semantic input, so the lock and every package move in the
regeneration that follows.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Regenerated by regenerate.yml (run 43311807929) after the self-hosted
validator and the raised output limit: the bootstrap export, build oracles,
fixtures and verify records. No count moved and every recursive package keeps
its stack row.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@humuhumu33
humuhumu33 marked this pull request as ready for review November 12, 2026 23:28

@afflom afflom left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Current product adoption credited at0b77d725. Complete non-truncated tree comparisons of the changed bootstrap/tests subtrees find1,463 raw changed paths:45 bootstrap build,1,293 verification and125 production-export paths. There is no authored source/config/case change in this delta. Sampled current config and RuntimeValidate input digests agree with the build manifest; validate_runtime provenance agrees on compiler semantics and bootstrap semantic identity. This supersedes the previous 'no new products committed' status, without claiming every generated byte has been independently validated.

Keep the existing complete728-vector/49-rule/8-plant differential, export bindings and full final-head gate as the acceptance route. No historical count-forensics gate or repeat of the completed producer is required. Surviving CI43329099701 is pending; cancelled43329037276 is excluded, and exact-head repro43329037236 passed. Please update the PR body to distinguish this adoption from its older regeneration/count narrative when assembling acceptance. Existing owner retains qualification; no logs, runtime or writers were repeated by this readback.

afflom commented Nov 13, 2026

Copy link
Copy Markdown
Member

Conditional delivery preparation is complete for current0b77d725. Existing run43329099701 has66/69 materialized jobs successful; Intel moves/lower/print are still queued, so this is not final acceptance. Same-run honesty, runtime, rest, compiler, Windows and Atlas/join passed; exact repro43329037236 passed. Cancelled43329037276 remains excluded.

The successful honesty job131581289671 explicitly checked out syntheticb8514297, treecb61de9e, parents actual26e5+0b77. That resulting tree differs from the branch head only by the inherited floating-point proposal; final landing must preserve the qualified resulting tree, not claim head/merge byte equality. One previously unread checkout log is retained; no log replay is needed.

The full #426 body maps to this RB24 delivery route. After complete gate/final review and actual merged-tree equality, close #426 explicitly with the merged PR; #271/#269 remain open for broader integration. Update the PR's evidence narrative to distinguish current producer adoption43311807929 and728-vector/49-rule/8-plant coverage from the older run/count summary. As review5882283043 established, the historical575/576 vector investigation is not a further acceptance prerequisite. No repeated producer, proof or historical count gate is requested.

@humuhumu33
humuhumu33 merged commit b8b6e86 into release/v0.4 Nov 13, 2026
71 of 89 checks passed
@humuhumu33
humuhumu33 deleted the claude/issue-271-rb24-driver branch November 13, 2026 05:19

@afflom afflom left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Postlanding acceptance readback at0b77d725: CI43329099701 passed70/70 jobs and772/772 steps with same-run honesty; exact repro43329037236 passed. The externally merged b8b6e86 treecb61de9e equals the qualified b851 checkout, preserving ordered26e5+0b77 ancestry. Current products and the complete728-vector/49-rule/8-plant differential are credited. The old BS09 aggregate failure is retired as a current blocker; no historical forensic gate or completed producer replay is owed. This is a postlanding review, not a claim of final approval before the external merge.

#426 remains open for its explicit first acceptance clause. The actual RuntimeHost compiler/pinned-Lean evidence is not established by the retained gate mapping: rust_driver.rs105–116 checks source/shipped/embedded equality then decode_host; that route parses source data and validates Rust shapes. Compiler Main imports and its committed22-input manifest exclude RuntimeHost. Real Lean certification of the command-root fixture is credited, but it is not direct qualification of RuntimeHost itself.

Correction to conditional comment6842368308: its broad full-body mapping did not establish this distinction. Existing owner should first link any already-produced exact-source RuntimeHost compiler/Lean receipt and map the original two missing-twin obligations to current evidence. If absent, propose the smallest missing qualification in the existing lane; this does not invalidate completed fixture proofs or call for blanket regeneration. The PR body now records current adoption43311807929, terminal results and the remaining clause accurately. #271/#269 remain open.

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.

2 participants