From 0f01d28af2384875dc21dbf87c29bc9474f0df78 Mon Sep 17 00:00:00 2001 From: sunerpy Date: Fri, 21 Aug 2026 06:25:35 +0800 Subject: [PATCH] docs(upstream-sync): close the post-1.5.0 range at v0.46.1 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The explore budget epic shipped, so nothing from the 104-commit unreleased range is left unported. The entry's table, the header, and the title move to six tiers through v0.46.1, and the "still open" section becomes a closeout. It keeps the finding that explains why that port took eight review rounds: the whole-file rule returns a complete section before any budget or focus logic runs, and its line and character thresholds are independent, so a named symbol could vanish by two unrelated routes — one of which no clustering-path fix can reach. Both produced the same empty 615-byte answer, which is why the port needed two separate regression tests rather than one. --- docs/upstream-sync/UPSTREAM.md | 33 ++++++++++++++++++++------------- 1 file changed, 20 insertions(+), 13 deletions(-) diff --git a/docs/upstream-sync/UPSTREAM.md b/docs/upstream-sync/UPSTREAM.md index 121ac162..18f6738a 100644 --- a/docs/upstream-sync/UPSTREAM.md +++ b/docs/upstream-sync/UPSTREAM.md @@ -14,8 +14,8 @@ CLOSEOUT entry) - **This project version:** see `Cargo.toml` (`codegraph-rs`, independent line) - **Last sync:** `2026-08-20` — the post-`v1.5.0` **unreleased** range (104 commits - on upstream `main` past the `v1.5.0` tag) was triaged and five tiers shipped - across `v0.43.2` → `v0.46.0`; see the 2026-08-20 entry. Tracked parity stays + on upstream `main` past the `v1.5.0` tag) was triaged and six tiers shipped + across `v0.43.2` → `v0.46.1`; see the 2026-08-20 entry. Tracked parity stays `1.5.0` — none of these releases is a port of an upstream *release* range, and upstream's newest tag is still `v1.5.0` - **Next sync starts here:** `git ls-remote --tags` alone is not enough. Diff @@ -73,7 +73,7 @@ ## Sync log -### 2026-08-20 — post-`v1.5.0` unreleased range TRIAGED, five tiers SHIPPED (`v0.43.2` → `v0.46.0`; tracked parity stays `1.5.0`) +### 2026-08-20 — post-`v1.5.0` unreleased range TRIAGED and CLOSED, six tiers SHIPPED (`v0.43.2` → `v0.46.1`; tracked parity stays `1.5.0`) `git ls-remote --tags` still reports `v1.5.0` as upstream's newest tag, so a version-delta sync had nothing to pull. But `main` had accumulated **104 @@ -96,6 +96,7 @@ port of an upstream *release* range. | `v0.44.1` | #215 | imported members, cross-module Go calls, declaration initializers | #1518, #1521, #1510/PR #1511, #1524, #1512, #1504, #671 | | `v0.45.0` | #217 | Tauri commands bound to their `invoke` call sites | #1543 | | `v0.46.0` | #220 | generated files detected by their content banner | #1500, `57e0854` | +| `v0.46.1` | #223 | a named symbol always renders; real budget accounting | CG-38/36/30/28/26/31 | `v0.43.2` also carried `h2` 0.4.16 for **RUSTSEC-2026-0258** — unrelated to the sync, but it is why that release exists at that moment: the advisory published @@ -155,16 +156,22 @@ Upstream is only now adding GDScript (#1499), which we shipped with a full framework resolver (F1 autoload calls, F2 signal handlers, F3 impact/affected unification) and a byte-stable golden corpus. -#### Still open from this range - -The `explore` budget/selection epic (CG-38/36/30/28/26/31) is planned and under -review but not yet shipped. Its review surfaced a defect worth recording ahead of -the port: `engine.rs`'s whole-file rule returns a complete section before any -budget or focus logic exists, and its line and character thresholds are -independent — so a 63-line file over the character cap falls through to -clustering and can be rejected whole, which makes an explicitly named symbol -render as **615 bytes with zero sections**. That is the same empty answer CG-38 -exists to prevent, reached by a path a clustering-only fix would not cover. +#### Nothing from this range is left unported + +The `explore` budget/selection epic shipped as `v0.46.1`, which closes the range. +Its review took eight rounds, and the reason is worth recording: the whole-file +rule returns a complete section **before** any budget or focus logic exists, and +its line and character thresholds are independent — so a named symbol could +vanish by **two** unrelated routes. A file that passes both gates has a complete +section built and then rejected whole, where no clustering-path fix can reach it; +a file that passes the line gate but fails the character gate falls through to +clustering and is rejected there. Both produced the same **615 bytes with zero +sections** that CG-38 exists to prevent, which is why the port needed two +separate regression tests rather than one. + +CG-27 is recorded above as measured-and-rejected by upstream, and is **not** +ported. The 26 DEFER items and the permanently-N/A `0682137` recorded in earlier +entries are unchanged. The 26 DEFER items and the permanently-N/A `0682137` recorded in earlier entries are unchanged.