Skip to content

docs(upstream): record the v0.55.0 release and close the v1.6.2 sync #329

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
sunerpy merged 1 commit into from
Oct 22, 2026
Merged

docs(upstream): record the v0.55.0 release and close the v1.6.2 sync #329

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
213 changes: 198 additions & 15 deletions docs/upstream-sync/UPSTREAM.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,11 +7,13 @@

## Current alignment

- **Tracked colby release:** `v1.6.1`. All portable in-scope behavior audited
through that release is closed: ported, already present, recorded N/A, or
deferred by the owner, as the 2026-10-01 CLOSEOUT entry below lists. The
official codegraph-rs `v0.51.0` binaries passed the release smoke recorded
there.
- **Tracked colby release:** `v1.6.2`. All portable in-scope behavior audited
through that release, and through the `main` commits after it up to the last
discovery audit, is closed: ported, already present, recorded N/A or
KEEP-RUST, excluded by the owner, or moved by the owner to the follow-up pull
request, as the 2026-10-22 CLOSEOUT entry below lists. The official
codegraph-rs `v0.55.0` binaries passed the black-box acceptance in the
2026-10-22 release entry.
- **Last discovery audit:** colby `main` at
`b635dd467f0578926a9c01a37b9d28d2b26689f1` (2026-10-07), audited on
2026-10-10 in [`V1_6_2_AUDIT_2026-10-10.md`](V1_6_2_AUDIT_2026-10-10.md): 268
Expand All @@ -23,15 +25,17 @@
- **Next discovery starts here:**
`b635dd467f0578926a9c01a37b9d28d2b26689f1..origin/main`. Do not re-triage the
audited range.
- **Planned for codegraph-rs `v0.55.0`:** the 2026-10-10 audit's rows under
waves W0 to W6 and W10 of its program. By the owner's decision of 2026-10-16
the rows under W7, W8 and W9 (the UI family F1–F12 with the synthesis stage
and the framework resolvers, VB.NET, `codegraph context`, #1617 and the 26
v1.5.0 performance strategies) move to a follow-up pull request; the audit
marks them `PLANNED (follow-up PR)`, and a row that lands only partly in
`v0.55.0` names both halves. See the 2026-10-16 entry below. The tracked
release stays `v1.6.1` until `v0.55.0` ships and its downloaded binaries are
verified.
- **Shipped in codegraph-rs `v0.55.0`, 2026-10-22:** #322, the 2026-10-10
audit's rows under waves W0 to W6 and W10 of its program, with extraction
version 22. The audit's `LANDED` rows name the commits and #322. See the
2026-10-22 release and CLOSEOUT entries below.
- **Planned for the follow-up pull request:** the rows under W7, W8 and W9
(the UI family F1–F12 with the synthesis stage and the framework resolvers,
VB.NET, `codegraph context`, #1617 and the 26 v1.5.0 performance
strategies), by the owner's decision of 2026-10-16. The audit marks them
`PLANNED (follow-up PR)`, and a row that landed only partly in `v0.55.0`
names both halves. That pull request moves the extraction version from 22 to
23. See the 2026-10-16 entry below.
- **Upstream repository:** <https://github.com/colbymchenry/codegraph>.
- **Shipped in codegraph-rs `v0.52.0`, 2026-10-01:**
- #286, which makes a full `sync` keep what `index` keeps;
Expand Down Expand Up @@ -84,14 +88,40 @@
deferred, scheduled as phases 2–4: F10/F11 first, then F1–F6 with Screens and
Steps, then F7–F9 and F12. The 2026-10-10 program planned all of them for
`v0.55.0`; on 2026-10-16 the owner moved them, with the rest of W7, to the
follow-up pull request. `v0.55.0` lands from them only `950686de`'s test-file
follow-up pull request. `v0.55.0` landed from them only `950686de`'s test-file
classifier and `b1f40c57`'s `e2e/` test path (one predicate with the
resolver's), and the route roots its routes list reads.

The Rust product and upstream release versions are independent. This Current
alignment block is the only place that states release parity; dated log entries
below remain immutable historical evidence.

> **1.6.1 → 1.6.2 sync: COMPLETE** (codegraph-rs `v0.54.0` → `v0.55.0`).
> Every portable, in-scope behavior of the audited range — the `v1.6.2`
> release and the 88 `main` commits after it, through `b635dd46` — landed in
> #322, except the exceptions below, and the official `v0.55.0` binaries passed
> the black-box acceptance in the 2026-10-22 release entry below. The explicit
> exceptions are:
>
> - **COBOL is OWNER-EXCLUDED** (2026-10-10): the extraction (#1161), the CICS
> resolver, #2272 and #2358. crates.io's `tree-sitter-cobol` is a placeholder,
> and the grammar manifest forbids vendoring a grammar.
> - **G16 and G17 are OWNER-EXCLUDED** (2026-10-12): two older C/C++
> extraction gaps found while porting #2446 and only recorded. A variable at
> namespace scope, and a struct or union defined inside a function body, get
> no node.
> - **Refusing a schema newer than the binary (OD-19) is PLANNED (follow-up
> PR)**, with the design constraints the audit records.
> - **The rows under W7, W8 and W9 are PLANNED (follow-up PR)** (2026-10-16):
> the UI family F1–F12 with the synthesis stage and the framework resolvers,
> VB.NET, `codegraph context`, #1617 and the 26 v1.5.0 performance
> strategies.
> - **KEEP-RUST divergences** are recorded row by row in the audit. Three of
> them add behavior upstream does not have: property type references for
> Kotlin and PHP, C# extension methods reached from an outside receiver
> type, and C/C++ alias signatures.
> - The `1.6.0 → 1.6.1` and older caveats below carry forward unchanged.

> **1.6.0 → 1.6.1 sync: COMPLETE** (codegraph-rs `v0.50.3` → `v0.51.0`).
> Every audited portable, in-scope behavior landed in six porting PRs (#275,
> #277–#279, #281, #282) on the #271 engineering baseline, and the official
Expand Down Expand Up @@ -186,6 +216,159 @@ below remain immutable historical evidence.

## Sync log

### 2026-10-22 — CLOSEOUT: `v1.6.1 → v1.6.2` sync COMPLETE in codegraph-rs `v0.55.0` (tracked parity advanced to `v1.6.2`)

The 2026-10-10 audit covered `f4ddf508..b635dd46`: 268 commits, of which the
first 180 make up `v1.6.2`. #322 landed every row the program kept for
`v0.55.0`, waves W0 to W6 and W10, in 206 commits squashed into `0796c8a`.
The audit's status column now reads `LANDED` with those commits and #322,
`ALREADY-HAVE`, `KEEP-RUST`, `N/A`, `OWNER-EXCLUDED`, or `PLANNED (follow-up
PR)`; no row is still `PLANNED` for this release.

The tracked release moves to `v1.6.2`, the newest formal tag before
`b635dd46`. Everything portable through it has landed, except the owner's
exceptions in Current alignment, and the downloaded `v0.55.0` binaries are
verified in the release entry below. The rows `PLANNED (follow-up PR)` get
their own entry when that pull request lands. The next discovery starts at
`b635dd467f0578926a9c01a37b9d28d2b26689f1..origin/main`.

### 2026-10-22 — `v0.55.0` RELEASED: #322 shipped

The implementation PR #322 merged as `0796c8a`, with the tree of its reviewed
head `f5287d0` (`93c051b…`). The final kirocodex goal review failed on the
previous head `d21291b` with two blocking items and passed on `f5287d0` with
none. Main CI run
[`39599579698`](https://github.com/sunerpy/codegraph-rust/actions/runs/39599579698)
passed.

Release-please PR [#325](https://github.com/sunerpy/codegraph-rust/pull/325)
merged as `5f96a741504e656fd9f6bed138b288ab22a490a0`, with the tree of its head
`67a9a9d`, `029198b…`.

- The owner approved it as code owner. One Linux test failed in its CI run
[`39605156993`](https://github.com/sunerpy/codegraph-rust/actions/runs/39605156993)
(`plain_output`, a temporary directory two tests can share); the re-run of
the failed jobs passed, so attempt 2 is all green.
- GitHub never delivered the push event of that merge. The commit has no
workflow run and no check suite, and the repository activity log has no
merge entry for it, though every earlier merge has one. Release Please runs
only on push, and the exact-SHA gate needs a CI push run on the release
commit, which `ci.yml` cannot produce later. The release stalled.
- #327 merged as `b02b563`. A manual Release dispatch may now name the release
PR's own CI run; it is accepted only while the release commit has no CI push
run, and only for a run on the same tree, and a CI push run that appears
later must pass before publication. That merge's push was delivered: Release
Please created tag `v0.55.0` at `5f96a74` and the draft. Its run
[`39698159119`](https://github.com/sunerpy/codegraph-rust/actions/runs/39698159119)
was cancelled once the draft existed, because its exact-SHA gate could not
pass.

Release workflow run
[`39698634632`](https://github.com/sunerpy/codegraph-rust/actions/runs/39698634632),
dispatched on `main` with `tag_name=v0.55.0` and `ci_run_id=39605156993`,
passed all twelve jobs on its first attempt.

- The CI gate accepted #325's run `39605156993` (attempt 2, head `67a9a9d`,
tree `029198b…`), since `5f96a74` has no CI push run; the publish-time check
found none either.
- Every build job checked out tag `v0.55.0` and verified that it resolves to
`5f96a74`. The attestations name the dispatch run and its commit `b02b563`.
- The release was published at 2026-10-22T00:34:16Z UTC.
- The official `codegraph-0.55.0-x86_64-unknown-linux-musl.tar.gz` digest is
`c3503ff2f91c863b65d9d0c8c148eea047c92fd1553cfd0eb4be62e22c0357cc`.

Black-box acceptance of the downloaded release, against the official `v0.54.0`:

- **Artifacts.** `SHA256SUMS` and `gh attestation verify` passed for all six
archives. The binary reports `codegraph 0.55.0`, and its `.comment` section
names rustc 1.98.0.
- **Upgrade.** On this repository's `14c4f93` tree, an index built by
`v0.54.0` (extraction version 21) is `outdated` to `v0.55.0`. One `sync`
makes it current at extraction version 22 with nothing pending, and its
`export` is byte-identical to a fresh `init` (20,241 nodes, 67,962 edges).
`v0.54.0` reports the `v0.55.0` index as `future`, refuses `sync` and
`search`, and leaves every file of the index byte-identical.
- **Daemon.** A `v0.54.0` session and its daemon run on a `v0.54.0` index.
After `codegraph sync`, a `v0.55.0` session replaces that daemon ("replaced
the daemon an older codegraph install left running"), and a write reaches the
index. Without the `sync`, see the known issue below.
- **Cold start.** With the idle timeout at 4 s, the daemon of a cold
`v0.55.0` session is still alive after 8 s, and an edit then reaches the
index. `v0.54.0`'s daemon idle-exits under the open session, and the edit
never arrives.
- **Bulk rename.** With a live session, 600 files renamed at once leave
nothing pending after 4.9 s, and the index equals a fresh `init` of the
renamed tree.
- **Graph.** 20 seeds from the program's inputs hold on `v0.55.0`; 14 of them
do not hold on `v0.54.0`. The 20 include a C# base list, a Vue
`<script setup>` call, a Svelte instance call, Go embedding, a TS value
reference, static member reads in Java and Rust, a Kotlin property type, an
import that binds its file, `it()` no longer binding a project property, a
`super` call that is not a self call, Python bases, overloads by arity, a
type parameter that stays in scope, and a sync that retries a failed TS
import or PHP `require`.
- **Contract.** On the same tree, the `--json` output of `search`, `callers`,
`callees` and `impact` has the same fields and types as `v0.54.0`'s;
`export` adds `metadata.valueRef` on edges and changes nothing else.
- **MCP.** A second `serve --mcp` session beside a running daemon, and one
launched in `$HOME`, both answer MCP 2026-07-28 `server/discover` and
`tools/list` with `resultType: "complete"`. A 2025-06-18 `initialize` gets
2025-06-18 back. `v0.54.0` answers `server/discover` with "Method not
found".
- **Performance.** The published Linux binary exports the same graph as the
benchmarked `f5287d0` candidate. At 4 threads on this repository its wall
time and peak RSS are 11.41 s and 261 MiB, against 11.39 s and 271 MiB
(medians of 3 runs).

**Known issue, older than this release.** In the default daemon mode, an
index an older extraction version built is never upgraded on its own. The
session defers to the daemon, and the daemon refuses to start on an
`outdated` index, so MCP tool calls answer "state-gated Store open rejected
index state outdated" until `codegraph sync` runs. `v0.54.0` does the same on a
`v0.53.0` index. With `CODEGRAPH_NO_DAEMON=1`, the session upgrades the index
itself. `v0.55.0` moves the extraction version from 21 to 22, so every
`v0.54.0` index needs one `codegraph sync <project>`. The owner decided on
2026-10-22 to fix this in a patch release.

**Benchmark.** Baseline: the official `v0.54.0` musl binary. Candidate: the
`f5287d0` musl build with the Release workflow's toolchain (zig 0.15.1,
cargo-zigbuild 0.23.0, rustc 1.98.0). The cells are this repository's
`14c4f93` archive and typescript-large at 32, 4 and 1 threads, and the nine
other registry corpora at 32 and 4. Each binary ran `codegraph init` six times
per cell, interleaved with the other, on a 32-core Linux host (Xeon 6975P-C,
Ubuntu 24.04, kernel 7.0). The first run is dropped; the figures are the median
and range of the other five. Every kept run exited 0 and produced the same
canonical graph as its binary's other runs. All 24 cells are within the
budget: peak RSS at most 1.15x, and wall at most 1.10x or rows per second at
least 0.90x.

| corpus | threads | canonical rows | wall | rows/s | peak RSS |
| --- | ---: | --- | --- | ---: | --- |
| repo-14c4f93 | 32 | 142,230 → 166,893 (1.173x) | 8.19 s (7.90–8.36) → 8.67 s (8.39–8.89), 1.058x | 1.109x | 554 MiB (550–557) → 440 MiB (438–448), 0.795x |
| repo-14c4f93 | 4 | 142,230 → 166,893 (1.173x) | 12.89 s (12.79–12.99) → 11.39 s (11.38–11.44), 0.883x | 1.328x | 262 MiB (261–265) → 270 MiB (270–272), 1.032x |
| repo-14c4f93 | 1 | 142,230 → 166,893 (1.173x) | 36.29 s (36.14–36.34) → 29.02 s (28.72–29.07), 0.800x | 1.467x | 228 MiB (226–231) → 241 MiB (240–241), 1.056x |
| typescript-large | 32 | 236,805 → 243,505 (1.028x) | 17.62 s (17.30–17.73) → 15.27 s (14.96–15.77), 0.866x | 1.187x | 684 MiB (679–694) → 546 MiB (544–555), 0.797x |
| typescript-large | 4 | 236,805 → 243,505 (1.028x) | 38.39 s (38.24–38.95) → 20.66 s (20.61–20.71), 0.538x | 1.911x | 392 MiB (386–394) → 336 MiB (336–338), 0.859x |
| typescript-large | 1 | 236,805 → 243,505 (1.028x) | 120.07 s (119.56–120.57) → 47.77 s (47.61–47.86), 0.398x | 2.585x | 358 MiB (356–360) → 313 MiB (313–315), 0.876x |
| fd-small | 32 | 2,470 → 3,003 (1.216x) | 0.21 s (0.21–0.21) → 0.21 s (0.21–0.21), 1.000x | 1.216x | 275 MiB (273–276) → 122 MiB (121–124), 0.443x |
| fd-small | 4 | 2,470 → 3,003 (1.216x) | 0.21 s (0.21–0.21) → 0.21 s (0.21–0.21), 1.000x | 1.216x | 61 MiB (57–61) → 39 MiB (38–40), 0.644x |
| tokio-medium | 32 | 35,448 → 44,937 (1.268x) | 2.02 s (2.00–2.02) → 2.02 s (2.02–2.07), 1.001x | 1.266x | 355 MiB (353–359) → 205 MiB (203–205), 0.577x |
| tokio-medium | 4 | 35,448 → 44,937 (1.268x) | 2.37 s (2.32–2.37) → 2.22 s (2.22–2.22), 0.937x | 1.353x | 95 MiB (94–97) → 80 MiB (80–81), 0.847x |
| express-medium | 32 | 11,846 → 12,046 (1.017x) | 0.97 s (0.97–1.02) → 0.62 s (0.57–0.62), 0.635x | 1.602x | 342 MiB (342–346) → 182 MiB (179–182), 0.532x |
| express-medium | 4 | 11,846 → 12,046 (1.017x) | 2.87 s (2.87–2.87) → 0.52 s (0.51–0.52), 0.179x | 5.666x | 72 MiB (71–75) → 55 MiB (54–57), 0.769x |
| leveldb-small | 32 | 17,260 → 17,479 (1.013x) | 0.97 s (0.97–1.02) → 0.97 s (0.97–0.97), 0.998x | 1.015x | 286 MiB (283–292) → 119 MiB (117–122), 0.416x |
| leveldb-small | 4 | 17,260 → 17,479 (1.013x) | 1.27 s (1.22–1.27) → 1.02 s (1.02–1.02), 0.802x | 1.262x | 77 MiB (75–78) → 62 MiB (62–63), 0.806x |
| gson-medium | 32 | 34,746 → 43,157 (1.242x) | 2.07 s (2.02–2.07) → 2.22 s (2.22–2.23), 1.074x | 1.157x | 371 MiB (369–373) → 248 MiB (242–249), 0.668x |
| gson-medium | 4 | 34,746 → 43,157 (1.242x) | 4.12 s (4.07–4.12) → 4.12 s (4.07–4.17), 1.000x | 1.242x | 100 MiB (98–101) → 94 MiB (93–96), 0.932x |
| newtonsoft-large | 32 | 81,453 → 109,589 (1.345x) | 7.49 s (7.34–7.54) → 8.14 s (7.94–8.14), 1.087x | 1.238x | 416 MiB (414–421) → 360 MiB (355–363), 0.865x |
| newtonsoft-large | 4 | 81,453 → 109,589 (1.345x) | 14.79 s (14.74–14.84) → 12.84 s (12.79–12.94), 0.868x | 1.550x | 161 MiB (161–162) → 175 MiB (173–177), 1.086x |
| alamofire-small | 32 | 7,043 → 10,295 (1.462x) | 0.52 s (0.52–0.52) → 0.72 s (0.72–0.72), 1.388x | 1.053x | 330 MiB (321–334) → 195 MiB (185–195), 0.590x |
| alamofire-small | 4 | 7,043 → 10,295 (1.462x) | 0.62 s (0.62–0.62) → 0.82 s (0.82–0.82), 1.326x | 1.103x | 71 MiB (71–72) → 59 MiB (58–61), 0.833x |
| redis-medium | 32 | 65,393 → 67,446 (1.031x) | 4.07 s (4.07–4.12) → 4.22 s (4.07–4.42), 1.037x | 0.995x | 270 MiB (262–276) → 198 MiB (197–203), 0.734x |
| redis-medium | 4 | 65,393 → 67,446 (1.031x) | 3.07 s (3.02–3.07) → 3.22 s (3.17–3.27), 1.049x | 0.983x | 129 MiB (126–131) → 121 MiB (119–122), 0.936x |
| kong-medium | 32 | 41,516 → 41,516 (1.000x) | 3.02 s (2.97–3.07) → 2.22 s (2.22–2.27), 0.735x | 1.360x | 402 MiB (391–407) → 241 MiB (236–243), 0.598x |
| kong-medium | 4 | 41,516 → 41,516 (1.000x) | 5.53 s (5.48–5.58) → 2.67 s (2.62–2.67), 0.483x | 2.070x | 116 MiB (115–117) → 89 MiB (88–90), 0.765x |

### 2026-10-16 — the codegraph-rs `v0.55.0` program SPLIT into two pull requests (OD-18)

- **Decision:** asked about the program's progress after it had run more than
Expand Down
Loading
Loading