F by briansrls · Pull Request #657 · gunb-ai/gunbc

briansrls · 2026-04-22T17:42:43Z

Opened from session-dashboard for session smart-dove-561.

Investigation of the four error_type_templates shows the three __EMIT_BUG_{0}__ sentinels are already the native fail-closed form for Python/Go/dag (undefined identifier → target compile/import error). Rust's compile_error! is a message-carrying bridge that Python/Go/dag have no equivalent for in type position. Hits STOP-AND-ESCALATE clause #1 from the brief: the capability asymmetry is a target limitation, not something to paper over with a rename. The real dissolution is upstream (halt with a Diagnostic at 05_emit.dag:996,1046 rather than emit any error-type sentinel); that is C1 territory and out of this lane's scope. Deliverable: findings doc only. No template changes. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

briansrls · 2026-04-22T17:45:09Z

Review metadata

Provider / model: claude / claude-opus-4-7
Commit: 891f764b · Trigger: schedule
Thinking: 68s wall

Verdict: APPROVE — Documentation-only PR adding a single brief at docs/briefs/emit-bug-sentinel-convergence-2026-04-22.md. The brief explicitly delivers a STOP-AND-ESCALATE finding artifact, makes no template/code changes, and argues for upstream dissolution (Diagnostic + halt at 05_emit.dag:996, 1046) rather than per-target sentinel renames. Spot-checked the cited line references against the worktree — all accurate (dag:54, go:107, python:115, rust:141; consumer at 05_emit.dag:996 and 1046). No invariant or modeling-discipline violation in the diff; the recommendation is correctly aligned with the "dissolve, don't bridge" framing in the referenced feedback docs.

briansrls · 2026-04-22T17:46:51Z

Review metadata

Provider / model: codex / gpt-5.4
Commit: 891f764b · Trigger: schedule
Thinking: 122s wall

Findings

docs/briefs/emit-bug-sentinel-convergence-2026-04-22.md:42 violates INVARIANTS.md P1’s “Documentation Describes Live State” and misstates the P3 fail-closed receipt for Python. The brief says all four sentinels “guarantee the downstream compile/import step fails,” but the Python emitter unconditionally adds from __future__ import annotations (src/v2/05_emit_python.dag:301, mirrored in src/v2/stage0/src/v2_compiler_emit_python.rs:528), and src/v3/compiler/tests/boundary/m1_4_emit_python_test.rs:553 explicitly locks in deferred annotation evaluation. In live behavior, __EMIT_BUG_* inside a Python type annotation imports successfully and is stored as a string, so the acceptance criterion is not “already satisfied” for Python.

Verdict
REQUEST_CHANGES. The PR only adds this finding artifact, and the artifact’s central Python conclusion is currently inaccurate. Once the brief reflects the actual deferred-annotation behavior, the rest of the writeup looks consistent.

Python emitter unconditionally prepends `from __future__ import annotations` (05_emit_python.dag:301, locked by m1_4_emit_python_test.rs:553), so under PEP 563 a __EMIT_BUG_UNRESOLVED_* sentinel in annotation position is stored as a string and never evaluates. The previous writeup claimed acceptance was already met for all four targets; in fact Python has a live P3 gap today. Go/dag remain fail-closed via undefined-identifier; Rust via compile_error!. Recommendation still routes through upstream Diagnostic+halt (C1), which closes the Python gap by construction. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…scalations Pro's APPROVE_WITH_COMMENTS on #652 sha 457da80 flagged that the three new debt rows had end-state dissolutions but no named triggers, failing the tracked-debt discipline bar (per INVARIANTS P5 and modeling-discipline §95-106). Plus: the D/F/G lane escalations (merry-owl-732 stopped pre-PR, smart-dove-561 PR #657 findings, sunny-bee-588 PR #654 findings) revealed that two of my three rows were framed wrong — not just missing triggers, but naming the wrong dissolution target entirely. Updates: **Row: __EMIT_BUG sentinels (F rescope).** - Was: "converge three target specs on native compile-time error forms" - Now: all four are already fail-closed in native sense (F finding). Real dissolution is upstream at src/v2/05_emit.dag:996, 1046 — refactor to produce Diagnostic + halt instead of emitting target-specific sentinels. - Named trigger: when those two call sites are refactored to Diagnostic + halt, error_type_template becomes unreachable. - Cross-ref: PR #657 investigation receipt. **Row: algebra.dag declaration-vs-template (G reclassification).** - Was: "derive *_templates() from the type declaration" - Now: templates carry per-method metadata (size_effect, cost_shape, callback_element_position) declarations can't (G finding). Not redundant authorities — separate structural layers. - Named trigger: when Lane E substrate-carrier-port design lands a chosen option for per-method metadata carriers, *_templates() converges to the chosen form. - Cross-ref: PR #654 investigation receipt + Lane E coordination note. **Row: browser.dag typed-carrier collapse.** - Was: "route service ops through declared carriers" - Now: same description, but named trigger added — "when transport layer can consume typed carriers end-to-end." Lane H PR #658 pilots the pattern; success unblocks LLM + GitHub dissolutions that consume the same pattern. All three rows now have checkable dissolution triggers per INVARIANTS.md P5 tracked-debt discipline. Two rows additionally have their dissolution paths correctly pointed upstream rather than at surface symptoms. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…637 (#652) * WIP: gunbc Director * docs(roadmap): capture 3 novel debt rows + amend pipeline_authority post-#637 2026-04-22 exploratory analysis (gpt-5-4-pro against main@c9758c8) surfaced 3 genuinely novel debt items not in ROADMAP. Verified each against HEAD before adding: 1. **__EMIT_BUG_{0}__ fabrication sentinels** — 3 target emit specs (dag, go, python) use sentinel-string fabrication for error-ref emission; rust correctly uses compile_error!(). Target behavior diverges between fail-closed and fabricated modes. P3 / C-6 violation. Distinct from generic "TODO" debt — this is specific same-PR target-behavior inconsistency. 2. **algebra.dag declaration-vs-template parallel authority** — OrderedRing<T> declaration (16 fields) vs ordered_ring_templates() (9+clamp) are two authorities for the same surface, already drifted. Same pattern for PartialFunction. DISTINCT from the already-tracked signature/comment mismatch row — that's prose vs declaration (doc drift); this is code vs declaration (parallel authority). P2 Boundary Discipline. 3. **browser.dag typed-carrier service-boundary collapse** — same class as already-tracked LLM flattening + GitHub auth bypass: typed carriers declared in-file, raw String used at operational boundary. P1 Modeling Faithfulness / M8 violation. Plus amended the pipeline_authority row: #637 closed the stage- bindings portion (structural PipelineStageBinding consumption added), but pipeline_compile_order_names() at :225-275 still text-slices the compile-fn body for stage ORDER. That residual is blocked on substrate capability for ordered-body reflection — named the trigger explicitly rather than leaving it as generic "unassigned" follow-up. ResourceHandle forgeability (cited by the analysis) verified as already tracked at ROADMAP:138 — no new row needed. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * docs(roadmap): correct pipeline_authority amendment — #637 fully resolved it Codex BLOCKING review on #652 sha 457da80. My earlier amendment claimed #637 left a stage-order text-slicing residual blocked on substrate capability for ordered-body reflection. That was wrong. Verified against src/v3/compiler/src/pipeline_authority.rs: - ordered_pipeline_stages() IS the authority function (explicit in its doc comment: "the declaration order of the bindings in pipeline.dag is the ordering authority") - It reads PipelineStageBinding declarations structurally from the Dag (no text-slicing) - pipeline_compile_order_names() still exists but is called from reconcile_with_compile_body() as a DRIFT CHECK, not an authority path - The drift check fails closed if the text-sliced compile-body stage order disagrees with the structural authority So #637 FULLY closed the original P2 debt — authority moved to PipelineStageBinding declaration order. What remains is a defensive fail-closed drift check, which is a legitimate belt-and-suspenders pattern per INVARIANTS P3, not a violation. Fix: mark the row as RESOLVED. Earlier characterization of a substrate-capability blocker was a misread — I conflated "text- slicing exists in this file" with "text-slicing is the authority path." The two are different shapes. Lesson captured: authority location matters more than text-form location. Verify which function IS the authority before claiming a capability gap. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * docs(roadmap): sharpen pipeline_authority RESOLVED row with correct function names Pro's APPROVE_WITH_COMMENTS on #652 sha 457da80 caught a residual inaccuracy in my correction commit c7b33ca. I got the framing right (authority moved, drift check remains) but named the wrong function as the text-slicing site. Verified against pipeline_authority.rs post-#637: - pipeline_compile_order_names() at :185 is a thin wrapper around ordered_pipeline_stages(dag).map(.stage_name) — NO text slicing - compile_body_stage_names_from_source() at :130-180 is the actual text-slicing function - Called only from reconcile_with_compile_body() at :110-128 as fail-closed drift check Updated row to point at the correct function names and line ranges. Traceability preserved for future readers who grep. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * WIP: gunbc Director * docs(roadmap): add dissolution triggers + reclassify rows per D/F/G escalations Pro's APPROVE_WITH_COMMENTS on #652 sha 457da80 flagged that the three new debt rows had end-state dissolutions but no named triggers, failing the tracked-debt discipline bar (per INVARIANTS P5 and modeling-discipline §95-106). Plus: the D/F/G lane escalations (merry-owl-732 stopped pre-PR, smart-dove-561 PR #657 findings, sunny-bee-588 PR #654 findings) revealed that two of my three rows were framed wrong — not just missing triggers, but naming the wrong dissolution target entirely. Updates: **Row: __EMIT_BUG sentinels (F rescope).** - Was: "converge three target specs on native compile-time error forms" - Now: all four are already fail-closed in native sense (F finding). Real dissolution is upstream at src/v2/05_emit.dag:996, 1046 — refactor to produce Diagnostic + halt instead of emitting target-specific sentinels. - Named trigger: when those two call sites are refactored to Diagnostic + halt, error_type_template becomes unreachable. - Cross-ref: PR #657 investigation receipt. **Row: algebra.dag declaration-vs-template (G reclassification).** - Was: "derive *_templates() from the type declaration" - Now: templates carry per-method metadata (size_effect, cost_shape, callback_element_position) declarations can't (G finding). Not redundant authorities — separate structural layers. - Named trigger: when Lane E substrate-carrier-port design lands a chosen option for per-method metadata carriers, *_templates() converges to the chosen form. - Cross-ref: PR #654 investigation receipt + Lane E coordination note. **Row: browser.dag typed-carrier collapse.** - Was: "route service ops through declared carriers" - Now: same description, but named trigger added — "when transport layer can consume typed carriers end-to-end." Lane H PR #658 pilots the pattern; success unblocks LLM + GitHub dissolutions that consume the same pattern. All three rows now have checkable dissolution triggers per INVARIANTS.md P5 tracked-debt discipline. Two rows additionally have their dissolution paths correctly pointed upstream rather than at surface symptoms. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…G row Codex non-blocking on #652 sha 7362672: my rescoped EMIT_BUG row flattened Lane F's finding into "all four sentinels are already fail-closed" — but F's own investigation (PR #657) specifically notes Python's sentinel degrades under deferred annotations: > "Under `from __future__ import annotations` (default in many modern > Python codebases) annotations are stored as strings and never > evaluated at import, so both (a) and (b) defer failure to runtime > type-checkers or typed reflection — weaker than Rust's guarantee." My flattening contradicted F's own receipt (PR #657). Fix: preserve the nuance. Rust/Go/dag ARE fail-closed in native sense; Python is weaker under deferred annotations. The upstream-dissolution argument still holds — if the emit decision moves to Diagnostic + halt at 05_emit.dag:996, 1046, the Python deferred-annotations gap also evaporates since the sentinel never reaches target compilation at all. Added that note explicitly. Accurate summarization of F's finding restored. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

briansrls · 2026-04-22T18:04:33Z

Review metadata

Provider / model: claude / claude-opus-4-7
Commit: 3d520f17 · Trigger: schedule
Thinking: 42s wall

Verdict: APPROVE — Doc-only PR adding a single findings brief per the lane's STOP-AND-ESCALATE clause. No code under review.

The brief's claims check out against the cited lines: the four error_type_template forms (dag:54, go:~107, python:115, rust:141) are as described, the consumer sites at src/v2/05_emit.dag:996, 1046 use apply_type_template1(...error_type_template, "UNRESOLVED_"|"ANONYMOUS_COPRODUCT"), and the from __future__ import annotations prepend at src/v2/05_emit_python.dag:301 does in fact make the Python sentinel inert at import (PEP 563 reasoning is correct).

The recommendation is correctly oriented: dissolve upstream (produce a Diagnostic + halt at the detection site) rather than paper over templates — which aligns with the modeling-discipline principle of removing the bridge rather than ratcheting it. The Python gap is legitimately called out as a live P3, not hand-waved.

Exploratory observation (non-blocking): the brief asserts Go's undefined: __EMIT_BUG_FOO__ is fail-closed in type position, which is true — but a user could shadow by declaring that identifier (the brief notes this under "legibility"). If the C1 dissolution slips, adding a syntactic poison pill that also carries the label (e.g., a deliberately-malformed comment-prefixed form) would harden Go/dag without waiting for upstream. Noted as a fallback, not an ask.

briansrls · 2026-04-22T18:06:30Z

Review metadata

Provider / model: codex / gpt-5.4
Commit: 3d520f17 · Trigger: schedule
Thinking: 156s wall

Findings

docs/briefs/emit-bug-sentinel-convergence-2026-04-22.md:29 and :75 violate INVARIANTS.md’s “Documentation Describes Live State” by overstating the Python limitation. The current __EMIT_BUG_* identifier does import cleanly under from __future__ import annotations, but “no Python construct fails at import” / “no template change can close the gap on its own” is too strong: even with deferred annotations, a syntactically invalid annotation template still fails at parse/import time with SyntaxError. If the intended claim is narrower, e.g. “no valid annotation expression” or “no acceptable template-only fix,” the brief should say that explicitly.

Verdict
REQUEST_CHANGES. The overall upstream-halt recommendation looks reasonable, but this brief currently turns that recommendation into an incorrect impossibility claim for Python. Tighten that wording and I don’t have other concerns.

Reviewer correctly noted that "no Python construct fails at import" is too strong — a syntactically-malformed template would fail at parse time even under PEP 563. Narrowed the claim to "no valid-expression template fails at import," acknowledging the parse-failure escape hatch exists but noting it is coarse (file-level parse failure, label doesn't carry through) and not the right dissolution. Recommendation unchanged. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

briansrls · 2026-04-22T18:21:11Z

Review metadata

Provider / model: claude / claude-opus-4-7
Commit: f7e2066b · Trigger: schedule
Thinking: 38s wall

Verdict: APPROVE — docs-only PR adding a single finding brief. The cited file:line references (dag/emit.dag:54, go/emit.dag:107, python/emit.dag:115, rust/emit.dag:141, python/emit.dag:111, 05_emit.dag:996/1046, 05_emit_python.dag:301) all check out against the tree. The analysis correctly identifies Python's from __future__ import annotations policy as defeating any annotation-position sentinel, points to the upstream Diagnostic halt as the real dissolution rather than papering over with a rename, and explicitly scopes this PR to the finding artifact per the STOP-AND-ESCALATE clause.

This is exactly the tracked-debt pattern the invariants endorse: documented gap, bounded scope ("no template changes in this PR"), named dissolution trigger (open a C1 lane to dissolve error_type_template into a Diagnostic halt). No code under review to violate anything.

briansrls · 2026-04-22T18:22:24Z

Review metadata

Provider / model: codex / gpt-5.4
Commit: f7e2066b · Trigger: schedule
Thinking: 110s wall

Findings

NON-BLOCKING — P1: Modeling Faithfulness / Documentation Describes Live State: docs/briefs/emit-bug-sentinel-convergence-2026-04-22.md:51 cites feedback_fail_closed_discipline.md and feedback_construction_over_ratchets.md as if they are repo files, but I could not find corresponding markdown files in the tree. Existing docs reference those concepts without the .md suffix, so this currently reads like a dead citation rather than a live receipt.

Verdict
APPROVE_WITH_COMMENTS. The substantive brief looks accurate: the line references check out, and the Python from __future__ import annotations claim matches live behavior. My only concern is the dead-looking .md citation at line 51.

Repo convention cites guiding principles without the .md suffix (see CODING.md:167, design-db16, design-db18). The bare-name form reads as a live principle reference rather than a dead file path. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

briansrls · 2026-04-22T18:36:31Z

Review metadata

Provider / model: claude / claude-opus-4-7
Commit: 0b5946fe · Trigger: schedule
Thinking: 38s wall

Citations check out: template line numbers (54/107/115/141) match, the emit consumer branches at 05_emit.dag:996, 1046 are exactly as described (n_is_type_var || n_is_error path and anonymous-coproduct path both call apply_type_template1(template: spec.error_type_template, ...)), and from __future__ import annotations sits at 05_emit_python.dag:301. The technical claims about compile_error! firing at expansion regardless of position, PEP 563 deferring annotation evaluation to strings, and Go having no type-position escape hatch are all accurate.

Verdict: APPROVE — Doc-only PR adding a findings brief under the STOP-AND-ESCALATE clause. The diff does exactly what the brief's escalation path prescribes: surface the Python fail-closed gap and route the real dissolution upstream (Diagnostic-and-halt at the emit dispatch) rather than paper over per-target templates. Reasoning is grounded, citations are correct, and the "no template changes in this PR" disposition is the right modeling-discipline call — fixing error_type_template strings would be a bridge added in parallel to the missing upstream halt, not a dissolution.

Exploratory observation (non-blocking): The brief asserts Rust's compile_error! "preserves target-compile failure because the upstream halt is missing" and would be deleted alongside the other three once the C1 lane lands. Worth double-checking when that lane is scoped: if any emission path wants to produce a still-parseable artifact with an embedded marker (e.g. for partial-compile workflows), compile_error! semantics are materially different from __EMIT_BUG_*__ and the "delete all four" framing may need revisiting. Not a finding against this diff — just a note for whoever picks up C1.

briansrls · 2026-04-22T18:37:24Z

Review metadata

Provider / model: codex / gpt-5.4
Commit: 0b5946fe · Trigger: schedule
Thinking: 90s wall

APPROVE

Docs-only diff, and the substantive claims in docs/briefs/emit-bug-sentinel-convergence-2026-04-22.md match the live tree. I spot-checked the cited emitter/template lines and verified locally that from __future__ import annotations causes an undefined annotation like __EMIT_BUG_UNRESOLVED_TypeVariable__ to import cleanly as a string annotation, so the brief’s Python fail-closed analysis is accurate. No concrete violations of INVARIANTS.md, docs/modeling-discipline.md, CODING.md, or TESTING.md in this diff.

briansrls

Review metadata

Provider / model: codex / gpt-5.4
Commit: f7e2066b · Trigger: schedule
Thinking: 504s wall

BLOCKING (1)

Root Cause

docs/briefs/emit-bug-sentinel-convergence-2026-04-22.md The analysis checks how the emitted sentinel behaves in target syntax but not whether the sentinel namespace is user-reachable; re-evaluate the target matrix with authored-name shadowing included, then revisit the “Python outlier / no template changes” conclusion.

Non-blocking — Strengths

docs/briefs/emit-bug-sentinel-convergence-2026-04-22.md As a design/docs PR, it is grounded in actual emitter behavior, and the upstream diagnostic-and-halt dissolution still looks like the right long-term fix once the fail-closed analysis is tightened.

⚠️ The upstream-halt direction looks right, but the brief currently overstates Go/.dag fail-closedness by treating user-shadowable sentinels as a readability issue rather than a live correctness gap.

Sign in to view

+
+**Python is the outlier.** The emitter's `from __future__ import annotations` policy (PEP 563) stores annotations as strings, so the sentinel never evaluates and never fails. The acceptance criterion is **not** currently met for Python — this is a real P3 gap, not just a legibility concern.
+
+For Go and dag, what the sentinel violates is legibility, not fail-closedness: the sentinel is a valid identifier, so (a) it reads like regular code in diffs, and (b) a user could shadow it by declaring it themselves.


Reviewer correctly flagged that .dag identifier rules (01_tokenize.dag:516-522) accept leading underscores, and Go identifier syntax likewise admits __EMIT_BUG_*__. A sentinel that any declaration in scope can shadow is not fail-closed — it's a bridge that relies on the negative property "no one has declared this name." That's a C-8 violation. Previous framing split the four targets into "Python gap" vs "Go/dag legibility." The accurate split is: Rust (genuinely fail-closed via compile_error! macro expansion) vs three targets with live C-8 gaps of two shapes (Python deferred-annotation; Go/dag shadowable identifier). All three close by construction under the upstream Diagnostic-and-halt dissolution at 05_emit.dag:996, 1046. Recommendation unchanged; classification of the gaps sharpened. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

briansrls · 2026-04-22T18:52:41Z

Review metadata

Provider / model: claude / claude-opus-4-7
Commit: d5a7ab5a · Trigger: schedule
Thinking: 38s wall

Verdict: APPROVE — docs-only diff (88 lines, one new brief under docs/briefs/). No templates, code, or substrate changed.

Spot-checked the factual claims against the tree:

Template citations at dsl/extdeps/languages/{dag,go,python,rust}/emit.dag match (dag:54, go:107, python:115, rust:141). ✓
Consumer call sites at src/v2/05_emit.dag:996, 1046 match (the n_is_type_var || n_is_error branch and the anonymous-coproduct branch both call apply_type_template1(template: spec.error_type_template, …)). ✓
The from __future__ import annotations policy at src/v2/05_emit_python.dag:301 and mirror at stage0/src/v2_compiler_emit_python.rs:528 is real and unconditional. ✓

The brief correctly declines to paper-over with a rename, correctly identifies this as a C-8 gap (shadowable/deferred-evaluation sentinel relying on negative properties), and correctly routes the real fix upstream (Diagnostic + halt at the dispatch site) rather than into template policy. That aligns with feedback_fail_closed_discipline and feedback_construction_over_ratchets — addressing the substrate rather than strengthening a bridge. Scope discipline is good: explicitly "no change to the emission dispatch logic," finding artifact only.

No findings.

briansrls · 2026-04-22T18:53:41Z

Review metadata

Provider / model: codex / gpt-5.4
Commit: d5a7ab5a · Trigger: schedule
Thinking: 89s wall

Verdict: APPROVE

Docs-only diff. The new brief’s claims are consistent with the cited live code paths I spot-checked, including the per-target error_type_template definitions, the 05_emit.dag call sites, the .dag identifier rules, and Python’s unconditional from __future__ import annotations emission. I did not find a concrete violation of INVARIANTS.md, docs/modeling-discipline.md, CODING.md, or TESTING.md in the changed lines.

briansrls

Review metadata

Provider / model: codex / gpt-5.4
Commit: d5a7ab5a · Trigger: schedule
Thinking: 278s wall

Non-blocking — Strengths

docs/briefs/emit-bug-sentinel-convergence-2026-04-22.md The brief now matches live repo behavior, stays within docs scope, and tracks the remaining debt in bounded form by naming the upstream diagnostic-and-halt dissolution instead of proposing a new bridge.

✅ Docs-only diff, and I did not find a remaining contradiction with THESIS.md or the active invariants in the changed lines.

briansrls marked this pull request as ready for review April 22, 2026 17:42

briansrls mentioned this pull request Apr 22, 2026

docs(roadmap): 3 novel debt rows + pipeline_authority amendment post-#637 #652

Merged

2 tasks

briansrls commented Apr 22, 2026

View reviewed changes

briansrls merged commit da0855f into main Apr 22, 2026
4 checks passed

briansrls deleted the session/smart-dove-561-emit-bug-finding branch June 1, 2026 18:43


		Python is the outlier. The emitter's `from __future__ import annotations` policy (PEP 563) stores annotations as strings, so the sentinel never evaluates and never fails. The acceptance criterion is not currently met for Python — this is a real P3 gap, not just a legibility concern.

		For Go and dag, what the sentinel violates is legibility, not fail-closedness: the sentinel is a valid identifier, so (a) it reads like regular code in diffs, and (b) a user could shadow it by declaring it themselves.

Conversation

briansrls commented Apr 22, 2026

Uh oh!

briansrls commented Apr 22, 2026

Uh oh!

briansrls commented Apr 22, 2026

Uh oh!

briansrls commented Apr 22, 2026

Uh oh!

briansrls commented Apr 22, 2026

Uh oh!

briansrls commented Apr 22, 2026

Uh oh!

briansrls commented Apr 22, 2026

Uh oh!

briansrls commented Apr 22, 2026

Uh oh!

briansrls commented Apr 22, 2026

Uh oh!

briansrls left a comment

Choose a reason for hiding this comment

Uh oh!

This comment was marked as resolved.

Uh oh!

briansrls commented Apr 22, 2026

Uh oh!

briansrls commented Apr 22, 2026

Uh oh!

briansrls left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant