Skip to content

docs: correct false renumber claim in preflight comment (#387) - #578

Merged
ranxianglei merged 1 commit into
masterfrom
2026-09-06_fix-preflight-renumber-comment
Sep 7, 2026
Merged

ranxianglei merged 1 commit into
masterfrom
2026-09-06_fix-preflight-renumber-comment

Conversation

@ranxianglei

@ranxianglei ranxianglei commented Sep 6, 2026 •

Copy link
Copy Markdown
Owner

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 refs claim 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):

Re-run the pipeline each round: every successful compress renumbers the surviving refs, so the previous round's range refs are stale.

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

  • Comment-only change (verified by diff): 1 file, 4 insertions, 2 deletions, zero code changes.
  • After this lands, grep -r renumber src/ tests/ returns nothing in this repo.

Pre-flight

  • No code changed; no test/typecheck delta possible (diff is entirely within one comment block).

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.
@github-actions

github-actions Bot commented Sep 6, 2026

Copy link
Copy Markdown
Contributor

📦 Built Package Artifact

Branch: 2026-09-06_fix-preflight-renumber-comment (59cc908)

Option A — Install from npm PR tag (recommended)

npm install -g billion-context@pr-578

Each push to this PR publishes a new version under the pr-578 npm tag.

Option B — Download artifact

  1. Download the artifact from the Actions run
  2. Extract the tarball and install:
tar xzf billion-context-pr578.tgz
npm install -g package

This comment is automatically updated on each push.

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 ework agent · qwen3.8-27b

[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):

  • ✅ Comment-only: 1 file changed, 4 insertions(+), 2 deletions(-) — only src/preflight.ts, zero code changes. Matches the spec precisely.
  • ✅ The removed line was the repo's only renumber occurrence, so post-merge grep -rn renumber src/ tests/ returns nothing.

New comment text:

Re-run the pipeline each round: a successful compress hides its range behind a new block, changing the visible view; refs stay stable per-session snapshots (#387), but which ranges are compressible under them does not.

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 typecheck + build + the full test suite on an equivalent commit: green, except one unrelated launcher test (resolveClientCommand: codex/claude resolve to themselves) that fails locally only where a codex CLI happens to be installed — it fails identically without this change and passes on CI runners.

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

@ranxianglei

Copy link
Copy Markdown
Owner Author

Review vs master 3cd85b6 (kernel 0.0.56): ✅ merge-ready.

No follow-ups needed. (tracking: #595)

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 ework agent · qwen3.8-27b

[bot] 🏷 Re-verified against the live tree — agree, merge-ready.

Checked just now (master has since ticked forward to 1330fc5, just past your 3cd85b6; acp-kernel still pinned at 0.0.56):

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.)

@ranxianglei
ranxianglei merged commit 3c8e6bb into master Sep 7, 2026
6 checks passed
@ranxianglei ranxianglei mentioned this pull request Sep 7, 2026
@ranxianglei
ranxianglei deleted the 2026-09-06_fix-preflight-renumber-comment branch October 15, 2026 05:32
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

docs: preflight.ts comment falsely claims compress renumbers surviving refs

1 participant