Peel the non-planner planning and archive work onto its own branch - #95
Merged
Merged
Conversation
Everything on mikegrier/deferred-namespace-ops that is not the topology-planner planning work, separated so it can be reviewed and merged without waiting on a component that is still only a plan. 21 markdown files and .gitignore; no code, no Cargo.toml, no build impact. crates/topology-planner/ and the granularity.rs rustdoc link into it stay behind on the source branch. CHECKLIST-io-domains.md and the cache-locality design session were first classified as planner work on the strength of a cross-component handoff line. Measuring both directions reversed the call: io-domains is cited 16 times by the files moving here against once by the planner, and the design session 7 against once. Both moved. Four defects the peel surfaced, none of them in the work being peeled: - 32 doubled apostrophes (crates'', == ''push'') -- PowerShell single-quote escaping written out without unescaping. 27 in CHECKLIST-ship-topology-and-queues.md, 5 already on main in windows-topology-sys/COMPLETED-CHECKLIST.md. Swept the whole repository; 0 remain. - SH-4.8's last finding described an open defect in queue_contention's clock ordering that main has already fixed, and named a file:line matching neither the binary nor the module. Rewritten as done, with the line number dropped rather than re-pointed. - CHECKLIST.md linked src/bin/queue_contention.rs; that probe is a directory on main. Now points at queue_contention/main.rs. - CHECKLIST-ship-topology-and-queues.md linked ../.github/ from a root-level file, escaping the repository. ci.yml's only change was two comment lines accidentally joined into one, so it is reverted to main's text rather than carried here. The damage remains on the source branch and needs fixing there. 14 links to crates/topology-planner/ do not resolve on this branch, 6 of which already do not resolve on main. They resolve when the planner lands. Gate: encoding check 664 files clean; added lines 7-bit ASCII; LF throughout; every relative link resolves except that planner set. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Review round on the peel commit. Six findings, all real; two further claims declined, with reasons. - CHECKLIST-ship-topology-and-queues.md's "Where this stands" called PR #56 open as a draft, while the same file at SH-3.1.2 recorded isDraft=false. #56 was in fact closed unmerged on 2026-09-15, from mikegrier/deferred-namespace-ops -- the branch this content came from. Rewritten to say so, and that the content is landing in peeled pieces, the first of which merged as PR #94. The "221 commits ahead" figure went with it. Three surviving instances are explicitly dated or sit inside archived items, and are left as historical narrative. - M34.3 said CHECKLIST-io-domains.md holds "twelve checked items" and is "757 lines". It holds seven and is 693. Both tallies dropped rather than corrected, per the rule against pasting numbers that are not themselves the finding. - M34.3 queued only io-domains, but CHECKLIST-placement-tool.md and CHECKLIST-ship-topology-and-queues.md carry unarchived completed bodies too, and not one checked item across the three is a stub. Broadened to name all three. - COMPLETED-CHECKLIST.md linked SH-3.1.1 as the same-document fragment #m3-land-the-branch; that heading lives in CHECKLIST-ship-topology-and-queues.md. Now points at the file. - windows-file-watcher's M-inf.4 stub linked #m152, the anchor of the M15.2 entry that closed it. M-inf.4 now has its own anchor on that heading. Sweeping every stub in the change set found no other mismatch, and caught my own first attempt spelling it #minf4 when the convention keeps the hyphen (FL-D9.6 -> fl-d96). - Six inline-code references to repository files a reader is meant to open are now relative links. Left as inline code: bare file:line citations, which a link cannot carry; generic mentions of lib.rs and Cargo.toml; and io-domains' src/mpsc.rs, which names a superseded file and is already followed by a link to its replacement. Declined, with reasons: - Missing UTC offsets on added Moved headings and completion stamps. The repository's timestamp rule says not to go back and repair bare dates already written; these were authored earlier on the source branch and the instants are unambiguous from the commits. - The fourteen unresolvable crates/topology-planner/ links. Six already do not resolve on main; the component lands in its own pull request, and rewriting them here would conflict with it and lose the links when it arrives. Gate: encoding check 664 files clean; added lines 7-bit ASCII; LF throughout; no unresolved link outside the planner set. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Nine findings, eight real and fixed, one declined; a second claim about an earlier commit was checked and did not hold up. - DESIGN-NOTES.md listed the shipping queue traits as Bounded, Consumer, Drain, Observable, Producer, Reserving and Waitable. Drain is not a trait -- it is the iterator struct Consumer::drain returns -- and the list omitted Claim, which is one. The seven names appear to have come from the pub use line, which mixes both. - PLANS.md still said M34.3 was to "archive twelve completed item bodies". The previous commit removed exactly that tally from CHECKLIST.md, citing the rule against pasting a number that is not itself the finding, and broadened the item from one file to three -- and then left the restatement one file away untouched. Restatement drift introduced by the fix for restatement drift. - windows-ioring-sys/CHECKLIST.md said M20.5 was in flight on a separate branch and "That branch has since merged, and M20.5 arrived dissolved". The branch is mikegrier/deferred-namespace-ops, whose PR #56 was closed unmerged, and M20.5 is added by this diff. This is the same defect the previous commit fixed in the ship checklist; its sweep matched only draft-state wording, so it missed a merge-state claim in another file. Re-swept every added line for merge claims about the source branch. - M-inf.2 asked for the archival of the eight completed groups in CHECKLIST-thread-ambient.md, which this branch performs: M22-M29 are in COMPLETED-CHECKLIST.md and only the parked M26+ remains. Checked off, moved to the archive under its own anchor, and stubbed in place. - The M34 preamble asserted that the three root checklists share one milestone space with CHECKLIST.md holding M19-M21, thread-ambient M22-M29 and io-domains M30-M33. CHECKLIST.md already held M30 on main, so adding io-domains puts M30 -- and M30.2 through M30.5 -- on two different pieces of work inside a space this branch declares partitioned. The three feature-scoped checklists also number from M1 independently of it. Preamble corrected to describe both; M34.4 queues the collision, since a contradiction needs an item and not prose. - CHECKLIST-io-domains.md said thread-ambient held M22-M27. Everything else on the branch says M22-M29, and eight groups were archived. - The M3 row of the ship checklist said "4 of 5 open". M3 has nine SH-3.* items with two open, and its stated gate, SH-3.1.1, is checked. - PLANS.md said windows-ioring-sys M1-M18 are archived; this branch archives M19. Declined: - Reducing the four new `## Moved` headings to date-only. The repository instruction prescribes the full timestamp with its offset on that heading; the date-only reading came from M-inf.2, which is local and is now archived. Two same-minute groups are recorded 18 seconds out of order against their position; left, since the instants are unambiguous. Checked and not upheld: - That the previous commit was wrong to say the surviving "221 commits" figures are dated or sit inside completed items. Line 84 carries "As of 2026-09-02"; 259 sits in SH-3.1 and 287 in SH-3.1.2, both checked. The neighbouring "282" at 268 is likewise inside a checked item. The two figures do disagree, which is what a record written at two different times looks like. Writing M34.4 I said three sub-items collide when four do; caught by checking the claim before committing rather than after. Gate: encoding check 664 files clean; added lines 7-bit ASCII; LF throughout; every stub anchor matches its work-item ID and resolves; no unresolved link outside the disclosed planner set. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Third review round. Six findings, all real, plus a rendering defect the round did not name and one over-reach of my own that checking caught. - **The release had already happened.** This file opened with "The release has not happened", which I rewrote two commits ago to correct the pull-request state while leaving that sentence unexamined. crates.io says `windows-topology-sys` 0.2.0 and `windows-waitable-queues` 0.1.0 were both published on 2026-09-05 and neither is yanked. Status corrected. SH-4.15 now owns reconciling M4, and names the three items confirmed to contradict what shipped: SH-4.3 reasons from "a crate that has never been published" about the version that shipped; SH-4.2 instructs the ioring release that SH-2.2, its own cited authority, concluded does not arise; SH-15.9 schedules a `reserving_mpsc_wide` peer that reserving_mpsc.rs says was never built, the wide word having shipped as a layout under D-37 as amended by D-41. Queued rather than edited item by item: deciding which are done, void, or altered is a re-plan and the engineer's call. - **M34.4 was assigned twice**, once by me. The previous commit added a new M34.4 for the M30 collision without checking that M34.4 and M34.5 were taken, so one ID named a completed guard and an open re-plan at once. Renumbered to M34.6, and the gate now sweeps every checklist for duplicate item IDs rather than relying on having looked. - **M34.2 claimed a restriction that does not exist.** It said a probe's `render()` sits in a `bin` target "which nothing can import", and concluded that testing one requires moving every renderer into the library. A bin cannot be imported from outside, but it can carry its own test module, and this crate already does: `queue_contention`'s `main.rs` declares `mod tests;` and its tests.rs renders through `render_observation` into a String and asserts on it. The claim would have bought a workspace-wide refactor to remove an obstacle that is not there. - **The placement tool's "gate of record" had been withdrawn.** The header bullet gated PT-5.3 on SH-4.1 and SH-4.3, while PT-5.6 declares those prerequisites void and the crate carries `publish = false`. The same file's next paragraph still called both dependencies 0.1.0 with an unreleased breaking change. Both corrected. PT-5.6 also asserts that the reciprocal notes exist in the ship checklist; I read that as false on a first grep and it is true -- SH-4.1 and SH-4.3 both carry them. - **The status table had lost the blank line before it**, so it would have rendered as part of the preceding paragraph rather than as a table. Introduced two commits ago by the same splice that fixed the status text: a splice here consumes the blank line following its range, which had already cost two repairs, and this is the instance I did not think to check because it sat before a table rather than between items. Gate: encoding check 664 files clean; added lines 7-bit ASCII; LF throughout; no duplicate item IDs; every stub anchor resolves; no unresolved link outside the disclosed planner set. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Fourth review round. Two findings: one fixed, one declined. - COMPLETED-PLANS.md's new row linked the whole 3800-line archive rather than the entry it names, where the two sibling rows describing an archived feature checklist both carry anchors. Anchored, and the archived heading now carries an explicit `<a id>` rather than relying on a generated slug -- which is what the sibling at COMPLETED-CHECKLIST.md:1520 does, and what the convention requires, since VS Code's markdown engine resolves the HTML anchor and not the heading's auto-slug. The heading in question is added by this branch, so no archived history was touched to do it. Swept the rest of the change set for the same shape. The only other links naming an item while pointing at a whole file are two `MMT-1.2` references that are pre-existing on main and have no anchor to point at, so they are out of scope here. Declined: - That editing two crate-level COMPLETED-CHECKLIST.md files breaks the append-only rule. Every in-place change is either a bare-text file reference converted to a link -- which the cross-reference convention requires of any document you edit -- or a repair of the doubled- apostrophe corruption. The rule's stated purpose is that an archived summary and its link target cannot drift; no summary, heading text or link target changed. The one heading touched, in windows-file-watcher's archive, gained a second `<a id>` beside the existing one, so the old target still resolves and a new one does too. Two checks this round added to the gate, both because a grep of mine was too narrow rather than because a reviewer caught it: - Anchor resolution. Every `file.md#fragment` link in the change set is now checked for an `<a id>` or a matching heading slug, instead of only checking that the file exists. This is what should have caught the finding above before review did. - I read the sibling row's anchor as broken, on a grep for "review baseline" that could not match the hyphenated id in its `<a id>`. The anchor resolves. That is the second near-miss report this session from a pattern that was too specific; the mechanical check does not have that failure mode. Gate: encoding check 664 files clean; added lines 7-bit ASCII; LF throughout; no duplicate item IDs; every anchor resolves; no unresolved link outside the disclosed planner set. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Fifth review round. Six findings, all upheld -- including one I declined last round and was wrong to. **The append-only rule, reversed from last round.** I declined this in the fourth round on the reading that the rule forbids drift in a summary or a link target, not a typo repair. This round supplied the evidence I had not looked for: crates/windows-platform-probes/CHECKLIST.md, on main, states that the archive "is append-only and its entries are immutable" and calls it "a file that may not be edited to repair" references. That is exactly what my second-round edit did. Worse, the link it repaired was not broken: the M-inf.4 stub pointed at #m152, that anchor is on main, and it resolved. So the edit rewrote real history to buy nothing. - Reverted the anchor from windows-file-watcher's archive. The stub points at #m152 again and now says the entry is M15.2's, which closed both. That file is a pure append once more. - Reverted windows-topology-sys's archive to main entirely. Its whole diff was fourteen in-place edits with nothing appended: nine bare-text references turned into links and five doubled-apostrophe repairs, the latter mine from the first commit. Both kinds are precisely the "repairing references" the rule forbids. It also cut six unresolvable topology-planner links out of this branch, since converting bare text into a link is what made them broken links in the first place. Every completed archive in the change set is now a pure append, which the gate checks by requiring zero deletions against main. The other five: - PLANS.md described three checklists in states they have left: the io-domains row called the bounded MPSC outstanding when M31 built it and only its loom item remains, re-homed as M30.4; it counted M-inf at three items when there are five, so the count is gone rather than corrected; the shipping row read as though the release were ahead; and the placement-tool row called the tool publishable after PT-5.6 withdrew registry publication. - M26+ was marked parked on a gate that has lifted -- both windows-namespace-request-sys and windows-thread-ambient-sys are on main, so its three items are pending. Corrected in the checklist, its heading and PLANS.md; M34.7 owns graduating the id, since the convention's graduation has no number to graduate to when the unblocking event was a branch landing. - The cache-locality session still declared itself OPEN and on PR #56's critical path. The MMT-* plan that implemented it is archived and topology 0.2.0 published the new model. Status and that paragraph are marked for what they are; the transcript below them is untouched. - Three added file references in CHECKLIST.md are now links. - CHECKLIST.md had lost its final newline, and a list ran straight on from a paragraph. Both are mine, from splices: the M-inf.2 stub replaced the last line of the file, and the milestone preamble grew into the item below it. The gate now checks final-newline termination. Gate: encoding check 664 files clean; added lines 7-bit ASCII; LF throughout and every file newline-terminated; no duplicate item IDs; every anchor resolves; archives append-only; no unresolved link outside the disclosed planner set. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Sixth review round. Seven findings: five fixed, two declined. - **The same milestone was written up twice with different numbers.** Root PLANS.md said the streaming-report adapter left 332 `writeln!` sites untouched, measured against a probe killed 300 ms into a 0.8 s run; the new crate-level COMPLETED-PLANS.md said 504 sites, and a probe killed 8 s into a 65 s run keeping 114 bytes. The crate's own DESIGN-NOTES settles it -- 332 is what was measured on 2026-09-09, 504 is the superseded figure -- and that decision already records the cost of keeping such a census by hand: the edit that revised those figures is the edit that left a sentence in the sink's doc comment reading "is arithmetic" with nothing before it. So neither row now carries the digits. Both state the claim and link the decision that owns the numbers, per the rule about pasting measured values. - **The M16 gate had discharged in four places and said so in none.** M16 is archived, the session it waited on has concluded, the work became the `MMT-*` plan, and topology 0.2.0 shipped -- yet the status table, the critical-path paragraph, the M3 gate paragraph and PLANS.md all still said M3 waits on M16 and M16 waits on the session. This is the defect that file's own rule names: "a gate that has silently lifted is as harmful as one that has not." The membership was wrong too: three places named four gated items ending in SH-16.10, where the archive names six (SH-16.5/8/9/11/12/13) and lists SH-16.10 among the round's own findings, fixed. One remaining "gated on M16" sits inside a checked item as an account of why that item ran when it did, and is left. - **The archived M16 blockquote contradicted the entries beneath it**, calling the six superseded items "unchecked" where all six are `[x]` carrying `DISCHARGED` markers, and counting "six that remain" where seven do. Both corrected before the file lands, since it is immutable after. - The session's "Status of dependent work" still said SH-16.5 is blocked and unfixed. The header note added last round covers the transcript; a section asserting a dependency is live needs its own marker, as the critical-path paragraph already has. - The new crate-level COMPLETED-PLANS.md was unreachable: the ten other crates with one link it from their PLANS.md, and this one did not. Declined: - That the five timestamped `## Moved` headings breach the archive format. The instruction governing a completed group prescribes `## Moved YYYY-MM-DD HH:MM:SS +-hh:mm`, and explains main's date-only headings in the same breath: "existing headings without one are fine and are not worth going back to change." The date-only reading came from M-inf.2, which was local and is now archived. Two groups recorded 18 seconds apart sit in the opposite order; both instants are accurate records of when each move was made, and matching them would mean relocating the groups rather than correcting a value. - That filing the probes checklist as complete while M4 and M5 are open is a violation. The rows are milestone-scoped, and COMPLETED-PLANS.md's own header sanctions the state: "A checklist reappears in PLANS.md if new work is planned against it; the row here stays as the record of the work that was finished." Gate: encoding check 664 files clean; added lines 7-bit ASCII; LF throughout and newline-terminated; no duplicate item IDs; every anchor resolves; archives append-only; no unresolved link outside the disclosed planner set. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Fourth review round. Two findings: one real and fixed, one declined. The M34 preamble said the three feature-scoped checklists "number from M1 independently and are not part of that space". The M37 preamble two sections below contradicts it: it walks the shared space to find the next free number and skips M36 because M36 belongs to CHECKLIST-placement-tool.md. Both cannot be right, and the M37 one is -- M36 is archived under "placement tool M1-M4 and M36". So placement-tool is a mixed case: it numbers its own milestones from M1 and drew M36 from the shared space as well. My earlier wording turned that into a clean rule by leaving the exception out, which is how the next person picking a number would have picked M36 again. Restated to name every holder, including M35, and to say plainly that numbering from M1 is not evidence a file stays out of the space. Declined: that the M22-M29 archive violates the anchored-stub convention and so M-inf.2 should not be checked. The convention has two paths and this is the other one. A group migration moves a fully complete group under a dated `## Moved` heading and leaves only pending items behind; anchors and stubs belong to the separate rule for a large item moved while its group is still active. Measured: no `## Moved` heading in any of the three archives carries an explicit anchor, and the cross-reference rule says linking to the file is always acceptable when pointing at a heading. CHECKLIST-thread-ambient.md now has zero checked items and zero completed bodies, which is also why M34.3 names the other three files and not it. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Contributor
There was a problem hiding this comment.
🟡 Changes recommended
Several documentation statements conflict with the shipped topology decisions, and one checklist boundary violates the repository formatting convention.
Get a fresh assessment by requesting another Copilot review.
Pull request overview
Peels non-planner planning and design documentation into a standalone branch, updates archival trackers, and records release/status history. No code or build-impacting files changed.
Changes:
- Archives completed checklist work and updates planning references.
- Adds placement-tool planning documentation and preserves intentional planner links.
- Ignores generated probe records and repair backups.
File summaries
| File | Description |
|---|---|
PLANS.md |
Updates active plan statuses and release history. |
design-sessions/DESIGN-SESSION-2026-09-02-cache-locality-model.md |
Records the concluded locality-model session. |
DESIGN-NOTES.md |
Adds repository-wide design and queue decisions. |
crates/windows-topology-sys/DESIGN-NOTES.md |
Updates topology cross-references. |
crates/windows-topology-sys/COMPLETED-PLANS.md |
Archives the topology reshape plan. |
crates/windows-topology-sys/CHECKLIST.md |
Links planner adapter ownership. |
crates/windows-platform-probes/PLANS.md |
Adds completed-plan tracking. |
crates/windows-platform-probes/DESIGN-NOTES.md |
Queues workflow validation work. |
crates/windows-platform-probes/COMPLETED-PLANS.md |
Archives probe milestones. |
crates/windows-ioring-sys/COMPLETED-CHECKLIST.md |
Archives completed io-ring work. |
crates/windows-ioring-sys/CHECKLIST.md |
Replaces completed work with archive references. |
crates/windows-file-watcher/COMPLETED-CHECKLIST.md |
Archives watcher milestones. |
crates/windows-file-watcher/CHECKLIST.md |
Replaces completed milestone bodies with stubs. |
COMPLETED-PLANS.md |
Records completed workspace plans. |
COMPLETED-CHECKLIST.md |
Archives thread-ambient checklist groups. |
CHECKLIST.md |
Adds tooling milestones and archive tracking. |
CHECKLIST-thread-ambient.md |
Records that the M26+ gate has lifted. |
CHECKLIST-ship-topology-and-queues.md |
Records publication and PR status. |
CHECKLIST-placement-tool.md |
Adds the placement-tool plan. |
CHECKLIST-io-domains.md |
Updates queue/runtime planning and handoffs. |
.gitignore |
Ignores generated probe records and backup files. |
Review details
- Files reviewed: 19/21 changed files
- Comments generated: 3
- Review effort level: Lite
💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.
Comment on lines
+155
to
+156
| distinction mean something to a reader rather than only to CI. | ||
| ## M6: is a set of "equivalent" processors actually equivalent? |
Comment on lines
139
to
141
| by the representation, which is the subject of the open locality-model work -- see `SH-16.8` in | ||
| CHECKLIST-ship-topology-and-queues.md, where absence | ||
| [CHECKLIST-ship-topology-and-queues.md](../../CHECKLIST-ship-topology-and-queues.md), where absence | ||
| becomes first-class rather than a shape. Recorded here so the gap is not rediscovered, and queued |
Comment on lines
+171
to
+177
| **The whole-object `Provenance` is superseded, and should not be replaced by another | ||
| whole-object scalar.** Derivation: with trust per-relation, an object-level scalar can only be | ||
| the minimum (a topology with ninety-nine measured relations and one synthetic reads | ||
| `SYNTHETIC`, which is useless) or the maximum (which is dishonest). Trust belongs to an | ||
| **answer** -- "A and B share L3, established by these observations" carries its own -- and that | ||
| falls directly out of modeling observed connectivity, since a connectivity model exists to | ||
| answer queries and the query result is the thing needing a label. |
MikeGrier
pushed a commit
that referenced
this pull request
Sep 18, 2026
Brings back PR #95, the second peel from this branch, together with the 21 review fixes made to that content after it left. Merge rather than rebase: this branch is published and 404 commits deep, so a rebase would rewrite every one of them. Twelve conflicts, all in planning documents that had been peeled and then corrected on `main`. Eleven resolved to `main`'s side after checking, for each file, what this branch held that `main` did not -- in every case the branch-only lines were the pre-correction wording: the stale "the release has not happened", the "twelve checked items" tally, the M22-M27 range, the `Drain`-for-`Claim` trait list, the "has since merged" claim about a branch that closed unmerged, and 32 doubled apostrophes. PLANS.md was the exception and took main's version plus this branch's `crates/topology-planner` row, which `main` does not carry because the directory does not exist there. That row's own claim was stale too -- "deferred past PR #56" with #56 still open -- and now records that #56 closed unmerged and the component goes in its own pull request. Three things the conflict list did not name: - ci.yml carried a conflict resolution from merge `1abcaaf` that had welded a step's `if:` and `run:` onto one line, which is not valid YAML. `main` never changed the file, so git kept this branch's damaged copy silently. Taken from `main`, which is the only difference between them. - The ship checklist's status paragraph said "this branch is the second" peel. Written on the peel branch, where that was true; on this branch it names the source of the peels. Rewritten to name PR #94 and PR #95. - windows-platform-probes' design note explained why that paragraph reads differently on `main` -- "because `main` has no `M34` for it to belong to". The peel moved the root checklist to `main`, so `main` now has `M34.5` and both copies say the same thing. The divergence it described no longer exists, and `main`'s own copy still asserts it does. Kept deliberately divergent: windows-topology-sys' COMPLETED-CHECKLIST.md links topology-planner here and leaves the same references as bare text on `main`, because the targets resolve on one side and not the other. Verified: no conflict markers; encoding check clean; no duplicate work-item IDs; LF throughout; all 14 previously-unresolvable topology-planner links now resolve, and the only three broken links in the tree are pre-existing and untouched by this merge. The merge changes no .rs, .toml or .lock file, so the build is unaffected -- the tree's code is byte-identical to the pre-merge tip. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
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
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.
Second peel from the long-lived
mikegrier/deferred-namespace-opsbranch. Everything here isplanning and design documentation that is not
topology-plannerwork; the planner itself landsin its own pull request.
What this is
21 markdown files and
.gitignore. No code, noCargo.toml, no build impact -- so the cargogate has nothing to run against.
crates/topology-planner/stays behind, along with thegranularity.rsrustdoc link into it and thePLANS.mdrow that indexes it.Why the cut fell where it did
Two files were first classified as planner work on the strength of a cross-component handoff line.
Measuring citations in both directions reversed the call:
CHECKLIST-io-domains.mdis cited 16times by the files moving here against once by the planner, and the cache-locality design session 7
against once. The handoff line was a sample, not the population. Both moved.
Review
Four rounds against an unguided brief ("review all the changes"), one model each -- gpt sol, claude
opus, gpt astra, claude sonnet -- finding 6, 9, 6, 2 issues. 21 fixed, 5 declined with reasons,
2 checked and not upheld. The rounds found real defects that earlier rounds had introduced, so the
falling count is not convergence; it is variance between models.
The substantive ones:
crates.io shows
windows-topology-sys0.2.0 andwindows-waitable-queues0.1.0 both published2026-09-05, unyanked. Three items reason from the opposite:
SH-4.3argues from "a crate that hasnever been published" about the version that shipped;
SH-4.2instructs a release thatSH-2.2,its own cited authority, concluded does not arise;
SH-15.9schedules areserving_mpsc_widepeerthat
reserving_mpsc.rssays was never built. Status corrected;SH-4.15queues the re-plan ratherthan checking items off, because deciding which are done, void, or merely altered is the engineer's
call.
self-contradictory (open draft in the header,
isDraft=falseatSH-3.1.2).M34.2claimed a restriction that does not exist -- that a probe'srender()in abintargetcannot be tested without moving it to the library.
queue_contentionalready does exactly that:main.rsdeclaresmod tests;and itstests.rsrenders throughrender_observationinto aString. The claim would have bought a workspace-wide refactor to remove an obstacle that is notthere.
PT-5.6, and the crate carriespublish = false.M30now names two different pieces of work inside the shared milestone space. Queued asM34.6; renumbering is not a reviewer's call.crates'',== ''push'') from PowerShell quote-escaping written outunescaped.
ci.ymlcomment welded onto one line (reverted to main'stext rather than carried), a link to
src/bin/queue_contention.rswhere that probe is a directory,a root-level file linking
../.github/, and a stub pointing at the wrong anchor.Known and disclosed
8 links into
crates/topology-planner/do not resolve here. Six of them already fail to resolveon
maintoday; they resolve when the planner lands. Rewriting them would conflict with that branchand lose the links.
Gate
Encoding check clean across the repository; added lines 7-bit ASCII; LF throughout; no duplicate
work-item IDs; every completed-item stub anchor matches its ID and resolves; every archive
append-only; no unresolved link outside the planner set above.