Repository navigation
docs: correct false renumber claim in preflight comment (#387) - #578
Conversation
The kernel has no ref-renumbering mechanism (verified in #387): refs are stable per-session snapshots. The real reason the pipeline must re-run each preflight round is that a successful compress hides its range behind a new block, changing which ranges remain compressible under the same refs.
📦 Built Package ArtifactBranch: Option A — Install from npm PR tag (recommended)npm install -g billion-context@pr-578Each push to this PR publishes a new version under the Option B — Download artifact
tar xzf billion-context-pr578.tgz
npm install -g packageThis comment is automatically updated on each push. |
[bot] 🏷 Already implemented & verified — PR #578 correctly does this; I did not push a duplicate. A prior daemon run had already opened PR #578 for exactly this fix. Since its branch name is taken and a second PR would be a pure duplicate, I verified the existing PR against the stated scope instead of re-doing it: Scope (authoritative, via the GitHub API):
New comment text:
This is faithful to the corrected rationale in the issue: the range is hidden behind a new block; refs stay stable (per-session snapshots); what changes is which ranges are compressible — hence the pipeline re-runs each round. Pre-flight: comment-only ⇒ no typecheck/test delta is possible (as the issue notes). For completeness I ran Status: mergeable (base is simply ahead of head — normal). Ready for review/merge. Per project rules I don't merge PRs myself — please merge when you're happy: #578 |
|
Review vs master 3cd85b6 (kernel 0.0.56): ✅ merge-ready.
No follow-ups needed. (tracking: #595) |
[bot] 🏷 Re-verified against the live tree — agree, merge-ready. Checked just now (master has since ticked forward to
Nothing changes your read, and the extra commits on master didn't disturb the target. Nothing further from my side — it's yours to merge whenever. (tracking #595 noted.) |
Closes #595 — tracking issue for this PR.
What
Follow-up to #387. The kernel fix (acp-kernel PR #175, merged 2026-08-31, shipped in acp-kernel — now pinned at 0.0.54) removed the false
a prior compress renumbered the remaining refsclaim from all user-visible text. One remnant of the same false claim survived in this repo as a code comment:src/preflight.ts(preflight loop):Why it matters
Not user-visible, but it propagates the exact misconception that caused #387 to any future reader of the preflight loop. The corrected rationale: a successful compress hides its range behind a new block, changing which ranges remain compressible under the same (stable, per-session-snapshot) refs — that is why the pipeline must re-run each round.
Scope
grep -r renumber src/ tests/returns nothing in this repo.Pre-flight