From b4e1367242bf4b66f7485350a82e9a55f5eb690d Mon Sep 17 00:00:00 2001 From: Paul Fidika Date: Mon, 17 Aug 2026 17:21:55 -0600 Subject: [PATCH] pgw#1323: attribute the silent release note to the wheel that actually shipped it MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit pgw#1323's code shipped in 0.121.0 with no fragment, no consumed.tsv row and no mention in that section — a silent release note, verified four ways when it was found. Two cutters then tried the documented repair and backed it out, because `--version ` swept every pending fragment into the older section: 5 fragments for the 0.122.0 cutter, 16 for me on the 0.123.0 attempt. pgw#926 fixed the assembler, and this is the repair it makes safe. Run on fc254059's successor, it does exactly what docs/releasing.md always promised: pgw1323.md -> 0.121.0 (late; its code shipped in v0.121.0) 2 fragment(s) pending, not in 0.121.0: <- HELD, not swept pgw1346-b3a-families.md ... unshipped work cannot be dated into it pgw1355.md ... unshipped work cannot be dated into it Verified rather than assumed: exactly ONE `## 0.121.0` heading (the old bug prepended a duplicate), exactly one new ledger row `0.121.0 pgw1323`, the bullet inside the real section under the "*Attributed after the cut (pgw#1226)*" note, and `pyproject.toml` untouched — this is a repair, not a cut, and it bumps nothing. --- CHANGELOG.md | 11 +++++++++++ changelog.d/consumed.tsv | 1 + changelog.d/pgw1323.md | 8 ++++++++ 3 files changed, 20 insertions(+) create mode 100644 changelog.d/pgw1323.md diff --git a/CHANGELOG.md b/CHANGELOG.md index 8c3d5eb88..15d585c32 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -1184,6 +1184,17 @@ requires a generatable identifier because its bindings are generated. sink and the handler's own Asset; the fourth pins field 9's wire bytes, which is the half a typed assertion structurally cannot see. +*Attributed after the cut (pgw#1226): the change below shipped in this version's tag and was not assembled into the section at cut time.* + +- **The cross-repo consumer fence stopped reading its biggest consumer, silently.** + `inference-endpoints` was renamed to `serverless-endpoints`, and the author-surface fence still + named the old path — so the sweep that proves no public symbol goes unreached quietly stopped + reading 29 of the fleet's 32 endpoint packages and kept passing. `scripts/lint_unreached_surface.py` + now names the repo it expects, with `author_surface_allowlist.txt` and `skip_census.txt` + following it. The hazard this closes is not the rename but the REPAIR for it: GitHub redirects the + retired clone URL, so re-cloning under the dead name turns the fence green over a checkout no lane + updates. + ## 0.120.0 (2026-08-16) — **the transfer plane is OURS — gen-worker consumes tensorfs natively, and the mount client is gone.** - The tensorhub grant transfer plane is first-party at `gen_worker.transfer`. It was vendored diff --git a/changelog.d/consumed.tsv b/changelog.d/consumed.tsv index fb0f53362..bd923892f 100644 --- a/changelog.d/consumed.tsv +++ b/changelog.d/consumed.tsv @@ -91,3 +91,4 @@ 0.123.0 pgw1354 0.124.0 pgw1353 0.124.0 pgw1356 +0.121.0 pgw1323 diff --git a/changelog.d/pgw1323.md b/changelog.d/pgw1323.md new file mode 100644 index 000000000..ca85150f6 --- /dev/null +++ b/changelog.d/pgw1323.md @@ -0,0 +1,8 @@ +- **The cross-repo consumer fence stopped reading its biggest consumer, silently.** + `inference-endpoints` was renamed to `serverless-endpoints`, and the author-surface fence still + named the old path — so the sweep that proves no public symbol goes unreached quietly stopped + reading 29 of the fleet's 32 endpoint packages and kept passing. `scripts/lint_unreached_surface.py` + now names the repo it expects, with `author_surface_allowlist.txt` and `skip_census.txt` + following it. The hazard this closes is not the rename but the REPAIR for it: GitHub redirects the + retired clone URL, so re-cloning under the dead name turns the fence green over a checkout no lane + updates.