GRI-2052: keep gtk-renderer, accept the three advisories with a dated reason #202

Merged
vivy merged 3 commits from security/GRI-2052-cargo-deny-advisories into dev 2026-11-14 23:39:31 +00:00
Member

Closes forge #200.

Summary

dev was red on cargo deny check advisories (four RUSTSEC advisories). The
product owner's rule evolved twice on forge #200/#201/#2051 — the final
decision (GRI-2051, 23:15 UTC, his words): "Because our code does not call
that security risk, I will allow it for this. But as soon as there is a
update, you need to update it."

  • RUSTSEC-2026-0459 (bundled SQLite CVE via sqlx's sqlite feature):
    already cleared on dev by #203 — the client's state
    (crates/client/src/state.rs) is one JSON file, written through a
    temporary file and renamed into place for an atomic replace. No change
    here.
  • RUSTSEC-2024-0429 (glib unsoundness, VariantStrIter),
    RUSTSEC-2024-0436 (paste, unmaintained) and RUSTSEC-2024-0370
    (proc-macro-error, unmaintained), all reached only through
    gtk-renderer: no fix exists upstream. An earlier version of this PR
    moved the whole gtk-renderer stack out of the workspace, which would
    have dropped grimoirix-signage-client — the full-screen player the
    .deb installs and the campus production test needs. Per the product
    owner's final call, gtk-renderer stays, and all three advisories are
    accepted in deny.toml with a dated reason (2026-11-14) and a link to
    forge #201, which tracks the removal condition for each.
  • cargo deny check advisories -D advisory-not-detected is now the
    required, blocking gate in CI and in scripts/local-gates.sh (no more
    continue-on-error). -D advisory-not-detected turns cargo-deny's lint
    for "an ignore entry that no longer matches any reported advisory" into
    an error, so the day a fix lands for one of the three, the entry stops
    matching and the gate goes red until it is removed — enforced by the
    gate, not by memory. Proved locally: added a fake ignore entry, ran the
    check, got error[advisory-not-detected] / advisories FAILED; removed
    it and confirmed clean again.
  • ADR 0013 (advisories block a merge, superseding ADR 0008) gets a dated
    amendment recording this exception and its enforcement.

Test plan

  • cargo deny check -D advisory-not-detected — advisories ok, bans ok, licenses ok, sources ok
  • proof: a fake ignore entry fails with error[advisory-not-detected];
    removed, confirmed clean
  • cargo check --workspace --all-targets --features client/gtk-renderer
    — clean, gtk-renderer builds
  • cargo test --workspace --all-targets --locked and
    cargo test --workspace --doc --locked — passed
  • cargo clippy --workspace --all-targets --locked -- -D warnings —
    clean
  • cargo fmt --all --check — clean
  • ./scripts/local-gates.sh --rust-only — every gate passed
  • repo-boundary --staged — every path belongs

CI verdict: fj ci 202

Closes forge #200. ## Summary `dev` was red on `cargo deny check advisories` (four RUSTSEC advisories). The product owner's rule evolved twice on forge #200/#201/#2051 — the final decision (GRI-2051, 23:15 UTC, his words): *"Because our code does not call that security risk, I will allow it for this. But as soon as there is a update, you need to update it."* - **RUSTSEC-2026-0459** (bundled SQLite CVE via `sqlx`'s `sqlite` feature): already cleared on `dev` by #203 — the client's state (`crates/client/src/state.rs`) is one JSON file, written through a temporary file and renamed into place for an atomic replace. No change here. - **RUSTSEC-2024-0429** (`glib` unsoundness, `VariantStrIter`), **RUSTSEC-2024-0436** (`paste`, unmaintained) and **RUSTSEC-2024-0370** (`proc-macro-error`, unmaintained), all reached only through `gtk-renderer`: no fix exists upstream. An earlier version of this PR moved the whole `gtk-renderer` stack out of the workspace, which would have dropped `grimoirix-signage-client` — the full-screen player the `.deb` installs and the campus production test needs. Per the product owner's final call, `gtk-renderer` stays, and all three advisories are accepted in `deny.toml` with a dated reason (2026-11-14) and a link to forge #201, which tracks the removal condition for each. - **`cargo deny check advisories -D advisory-not-detected`** is now the required, blocking gate in CI and in `scripts/local-gates.sh` (no more `continue-on-error`). `-D advisory-not-detected` turns cargo-deny's lint for "an `ignore` entry that no longer matches any reported advisory" into an error, so the day a fix lands for one of the three, the entry stops matching and the gate goes red until it is removed — enforced by the gate, not by memory. Proved locally: added a fake `ignore` entry, ran the check, got `error[advisory-not-detected]` / `advisories FAILED`; removed it and confirmed clean again. - ADR 0013 (advisories block a merge, superseding ADR 0008) gets a dated amendment recording this exception and its enforcement. ## Test plan - [x] `cargo deny check -D advisory-not-detected` — `advisories ok, bans ok, licenses ok, sources ok` - [x] proof: a fake `ignore` entry fails with `error[advisory-not-detected]`; removed, confirmed clean - [x] `cargo check --workspace --all-targets --features client/gtk-renderer` — clean, `gtk-renderer` builds - [x] `cargo test --workspace --all-targets --locked` and `cargo test --workspace --doc --locked` — passed - [x] `cargo clippy --workspace --all-targets --locked -- -D warnings` — clean - [x] `cargo fmt --all --check` — clean - [x] `./scripts/local-gates.sh --rust-only` — every gate passed - [x] `repo-boundary --staged` — every path belongs CI verdict: `fj ci 202`
GRI-2052: clear the four cargo-deny advisories blocking dev
All checks were successful
CI / the pre-push hook (pull_request) Successful in 15s
CI / cargo fmt (pull_request) Successful in 20s
CI / cargo deny advisories (pull_request) Successful in 1m55s
CI / cargo deny (pull_request) Successful in 2m1s
CI / repository boundary (pull_request) Successful in 57s
CI / cargo clippy (pull_request) Successful in 3m14s
CI / frontend-web (pull_request) Successful in 2m25s
CI / cargo test (pull_request) Successful in 5m8s
CI / Notify gate failed (pull_request) Has been skipped
CI / CI (pull_request) Successful in 15s
a63fd1dcb5
RUSTSEC-2026-0459 (bundled SQLite CVE) cannot be cleared by swapping sqlx's
sqlite feature for rusqlite: rusqlite and sqlx's optional sqlx-sqlite
dependency can never share a libsqlite3-sys version in the same workspace
Cargo.lock, even with sqlx's sqlite feature never turned on, because several
of the server's sqlx features (migrate, chrono, uuid, json) hold a weak
feature edge back to sqlx-sqlite. So the client's state is a plain JSON
file for this release instead, written through a temporary file and
renamed into place for an atomic replace.

RUSTSEC-2024-0429/0436/0370 (glib unsoundness and two unmaintained crates,
all reached only through gtk-renderer) are cleared by taking gtk-renderer
and its dependencies out of the workspace for this release; nothing else
reaches glib or gstreamer. The removed code is archived at
refs/archive/2026-11-14/gtk-renderer-removal-GRI-2052 and forge issue #201
tracks bringing it back.

cargo deny check advisories now blocks CI and local-gates.sh instead of
being continue-on-error, per the product owner's rule that a vulnerability
or unsound advisory is never ignored.
vivy requested changes 2026-11-14 19:55:22 +00:00
Dismissed
vivy left a comment

Hold — do not merge. The advisory work is right and ignore = [] is right, but this PR removes the production client.

grimoirix-signage-client (src/bin/grimoirix_signage_client.rs) is the full-screen player the .deb installs, and it was behind gtk-renderer. With this PR, package-client-deb.sh and mvp-smoke-test.sh exit 1. The campus test (#29) would have no player to install. The direction I gave on #200 assumed gtk-renderer was dev-only; that was wrong, and the fault is mine, not the author's.

The JSON state file instead of SQLite is fine as a change, and the CI change (advisories blocking, in ci needs) is fine.

What happens to the glib / paste / proc-macro-error part is a product-owner decision, asked on GRI-2051. Until he answers, leave this PR open as it is.

**Hold — do not merge.** The advisory work is right and `ignore = []` is right, but this PR removes the production client. `grimoirix-signage-client` (`src/bin/grimoirix_signage_client.rs`) is the full-screen player the `.deb` installs, and it was behind `gtk-renderer`. With this PR, `package-client-deb.sh` and `mvp-smoke-test.sh` exit 1. The campus test (#29) would have no player to install. The direction I gave on #200 assumed `gtk-renderer` was dev-only; that was wrong, and the fault is mine, not the author's. The JSON state file instead of SQLite is fine as a change, and the CI change (advisories blocking, in `ci` needs) is fine. What happens to the `glib` / `paste` / `proc-macro-error` part is a product-owner decision, asked on GRI-2051. Until he answers, leave this PR open as it is.
Owner

Held — do not merge. This PR removes the screen player itself, not a spike.

gtk-renderer gates the binary grimoirix-signage-client (crates/client/Cargo.toml:32-35 on dev). scripts/package-client-deb.sh:32 builds exactly that binary into the .deb, and install.sh:110/425 installs that .deb on each display. There is no other player in the tree (no browser/kiosk path). So after this PR the campus mini PCs have nothing to run, and #29 (on-device acceptance) and the release #136 cannot happen.

The claim on #200 that gtk-renderer is "not part of the campus deploy" was wrong, and I relayed it into the plan without checking. My mistake, not Nyssa's.

The JSON-file state for RUSTSEC-2026-0459 looks like a sound answer on its own. The glib part needs a product-owner decision; question is on #200.

**Held — do not merge.** This PR removes the screen player itself, not a spike. `gtk-renderer` gates the binary `grimoirix-signage-client` (`crates/client/Cargo.toml:32-35` on `dev`). `scripts/package-client-deb.sh:32` builds exactly that binary into the `.deb`, and `install.sh:110/425` installs that `.deb` on each display. There is no other player in the tree (no browser/kiosk path). So after this PR the campus mini PCs have nothing to run, and #29 (on-device acceptance) and the release #136 cannot happen. The claim on #200 that `gtk-renderer` is "not part of the campus deploy" was wrong, and I relayed it into the plan without checking. My mistake, not Nyssa's. The JSON-file state for RUSTSEC-2026-0459 looks like a sound answer on its own. The `glib` part needs a product-owner decision; question is on #200.
vivy changed title from GRI-2052: clear the four cargo-deny advisories blocking dev to WIP: [DO NOT MERGE until the product owner answers the glib/paste/proc-macro-error question on forge #200] GRI-2052: clear the four cargo-deny advisories blocking dev 2026-11-14 22:12:49 +00:00
GRI-2052: keep gtk-renderer, accept the three advisories with a dated reason
All checks were successful
CI / the pre-push hook (pull_request) Successful in 11s
CI / cargo fmt (pull_request) Successful in 11s
CI / cargo deny (pull_request) Successful in 1m49s
CI / cargo deny advisories (pull_request) Successful in 1m56s
CI / cargo clippy (pull_request) Successful in 2m22s
CI / repository boundary (pull_request) Successful in 47s
CI / frontend-web (pull_request) Successful in 1m39s
CI / cargo test (pull_request) Successful in 3m50s
CI / Notify gate failed (pull_request) Has been skipped
CI / CI (pull_request) Successful in 16s
ec8c3956aa
The product owner's rule on GRI-2051 (23:15 UTC): our code does not call the
unsound glib path RUSTSEC-2024-0429 reaches, so it is accepted alongside the
two unmaintained crates underneath gtk-renderer (RUSTSEC-2024-0436, -0370),
each dated and with the removal condition on forge issue #201. The bundled
SQLite advisory stays cleared the way #203 already merged it -- a plain JSON
state file, no sqlx sqlite feature.

Restores gtk-renderer, its five crates and the two binaries
(grimoirix_signage_client, renderer_spike) from dev, where they were never
removed; only this branch had taken them out. package-client-deb.sh and
mvp-smoke-test.sh build the real player again.

cargo deny check advisories now runs with -D advisory-not-detected, in CI
and in local-gates.sh, so an ignore entry that stops matching a reported
advisory fails the gate the day a fix is available. Proved locally: a fake
ignore entry turns the gate red with error[advisory-not-detected]; removed
after confirming.
nyssa changed title from WIP: [DO NOT MERGE until the product owner answers the glib/paste/proc-macro-error question on forge #200] GRI-2052: clear the four cargo-deny advisories blocking dev to GRI-2052: keep gtk-renderer, accept the three advisories with a dated reason 2026-11-14 23:22:10 +00:00
vivy approved these changes 2026-11-14 23:39:28 +00:00
vivy left a comment

Approved on ec8c3956. My hold is resolved: gtk-renderer and both binaries are back (client Cargo.toml identical to dev), deny.toml ignores exactly RUSTSEC-2024-0429/0436/0370 with dated reasons and #201, advisories job has no continue-on-error and runs with -D advisory-not-detected (proved with a fake entry per GRI-2052). Six required contexts green. Note: Cargo.lock also carries transitive bumps (wry 0.57.2, dom_query 0.29, rustls-webpki 0.103.17, …) from the earlier commit; tests green, accepted.

Approved on ec8c3956. My hold is resolved: `gtk-renderer` and both binaries are back (client Cargo.toml identical to dev), deny.toml ignores exactly RUSTSEC-2024-0429/0436/0370 with dated reasons and #201, advisories job has no continue-on-error and runs with `-D advisory-not-detected` (proved with a fake entry per GRI-2052). Six required contexts green. Note: Cargo.lock also carries transitive bumps (wry 0.57.2, dom_query 0.29, rustls-webpki 0.103.17, …) from the earlier commit; tests green, accepted.
vivy merged commit b69a8d4556 into dev 2026-11-14 23:39:31 +00:00
vivy deleted branch security/GRI-2052-cargo-deny-advisories 2026-11-14 23:39:31 +00:00
Sign in to join this conversation.
No reviewers
No milestone
No project
No assignees
2 participants
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
Grimoirix/grimoire-of-digital-signage!202
No description provided.