Repository navigation
Conversation
…ormed text, in-memory recall Four defects the review of 9f6847c found, each fixed at the design assumption that made it. S3 (R2-6). An actionable record (a signed application section) is a message as well as a memory: memory keeps one record per cognition key, and the action is identified by its assertion identity wherever it is sent, de-duplicated and delivered. The frame handler's one delivery decision (_surfaceUnstored) is now taken on every branch where memory stored nothing, the admit branch included (a record SVAF admitted under a key the store held reached nobody). remember() sends an action whose cognition memory holds (HEAD's, an earlier action's, a peer's) by its own assertion (_sendAction: nothing stored again, HEAD unmoved, `cognitionHeld: true` and a `delivery` report), and an action whose store write failed; the same assertion given twice is a duplicate. lib/record.js: recordActionable, recordAssertion. B2 (R2-8). aggregateAttestations summarises one named assertion: given, held, or the sole one the attestations under the key name (a node without the record: XMesh's observers); several give no single verdict and byAssertion has each. Adds basis, assertions, byAssertion; the old fields stay. M2 (R2-13). The signed object is UTF-8 bytes, where a lone surrogate can only be U+FFFD: every string a record signs is brought to its well-formed spelling at each boundary (createCMB, a caller's record in remember(), the canonical projection, the Legacy Import route, the public verifier, the store's read) instead of refusing it, so a 0.14.2 author's text cut inside a pair verifies and is held as one string. The signing step still refuses a string no boundary normalised. sym's own cuts (message preview, wire names, SVAF text, context, error texts, decision labels) fall on code point boundaries (lib/core/text.js: cutText). M1 (R2-3). recall() matches on a bounded search index kept apart from the entry cache (each entry's lowercased text, 4,096 characters at most, 32 M in all, the oldest let go first) and reads only the entries it returns; recall(query, { limit, where }) and `total`. A scan reads through the cache, never into it; the cache starts with the newest entries; the timestamp ratchet and the lifecycle are read from the index. The primer, daemon, interior and mesh agent pass limits. Tests: tests/security-0.14.3-records.test.js (13, real nodes on the wire paths) and tests/integration/mixed-0142-records.js (4, against a real 0.14.2 tree, SYM_0142_DIR) fail on 9f6847c and pass here. R2-13's tests and remember-payload's collapse test now state the design as built.
The search-index recall read a non-string query as none and returned every entry; the store's old scan threw on it by accident (query.trim), which tests/daemon-relay-only relies on to see a throwing handler answered. A TypeError now, said on purpose; null, undefined and '' are still no query.
…the review of 9f6847c Each text states the design as built: actionable records sent and delivered by assertion with memory keeping one record per cognition key; attestation summaries about one named assertion; well-formed text normalised at every boundary, not refused; recall matched on a bounded index apart from the entry cache, reading only what it returns, with the numbers measured at 6,000 and 50,000 entries and the memory used.
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 3cfbaf8cde
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| if (t && !seen.has(t) && !content.includes(t)) { seen.add(t); parts.push(t); } | ||
| } | ||
| } | ||
| return parts.join('\n').toLowerCase(); |
There was a problem hiding this comment.
When a multi-word query spans adjacent fields, such as an entry tagged with ['alpha', 'beta'], the previous search matched alpha beta because it combined fields with spaces, while this newline separator makes that query return no results. This silently changes recall() semantics for phrases spanning tags, content, keys, or category fields; retain the prior space separator or normalize whitespace consistently in both indexed text and queries.
Useful? React with 👍 / 👎.
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.
Four defects the independent review of release/0.14.3 head 9f6847c found (S3, B2, M2, M1), reproduced against 0.14.2 and the candidate, each fixed at the design assumption that made it. Into release/0.14.3; not merged, tagged or published.
S3 (R2-6): actionable records are messages, not memories
An actionable record (non-null
application) is identified by its assertion everywhere: memory keeps one record per cognition key (§8.8.2), and the action is sent, de-duplicated and delivered by its assertion identity._surfaceUnstored) now runs on every branch where memory stored nothing, the SVAF admit branch included. 9f6847c surfaced only a directed record there, so an actionable room record SVAF admitted under a key the store held reached nobody.remember()sends an action whose cognition memory holds (HEAD's, an earlier action's, a peer's) by its own assertion (_sendAction: nothing stored again, HEAD unmoved,cognitionHeld: true, adeliveryreport), and an action whose store write failed. The same assertion given twice is a duplicate; a true replay is delivered once.B2 (R2-8): attestation summaries are per assertion
aggregateAttestationssummarises one named assertion: given, held, or the sole one its attestations name (a node without the record: XMesh's cliHostMode observers, 9f6847c total 0). Several: no single verdict,byAssertionhas each. Addsbasis,assertions,byAssertion; old fields unchanged. R2-8 kept: an attestation about another assertion is never a vote on this one.M2 (R2-13): well-formed text is normalised, not refused
The signed object is UTF-8 bytes, where a lone surrogate can only be U+FFFD. Every string a record signs is brought to its well-formed spelling at each boundary (createCMB, a caller's record in remember(), the canonical projection, the Legacy Import route, the public verifier, the store's read). A 0.14.2 author's text cut mid-emoji verifies and is held as one string. The signing step still refuses a string no boundary normalised. sym's own cuts fall on code point boundaries (
lib/core/text.js).M1 (R2-3): recall is not disk-bound
recall() matches on a bounded search index kept apart from the entry cache, and reads only the entries it returns.
recall(query, { limit, where }),total. Scans do not fill the LRU; the cache starts with the newest entries; the ratchet and the lifecycle are read from the index.Tests
tests/security-0.14.3-records.test.js: 13 tests, with real nodes on the wire paths. All 13 fail on 9f6847c and pass here.tests/integration/mixed-0142-records.js: 4 tests against a real 0.14.2 tree (SYM_0142_DIR). All 4 fail on 9f6847c and pass here.Recall, measured (one call, macOS, short entries)
limit: 10recall(''), no limitHeap after load: 8.4 → 10.4 MB at 6,000 entries, and 28.9 → 50.7 MB at 50,000 (0.14.2: 18 and 147 MB). The search index holds 2.4 M and 20.1 M characters.
Dependents
total1,basis: 'sole') instead of 0.emitOutcomereads an action whose cognition memory holds aspublished/sent, andremixed: falseroom deliveries as "delivered only". Recommended:node-host.recallshould pass{ limit }and usetotal, so a broad query on a large store does not read every match.Known