Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
22 commits
Select commit Hold shift + click to select a range
16ee93c
feat(moq-video): advertise the catalog rendition before the first key…
kixelated Aug 12, 2026
d75d412
fix(moq-hls): bind renditions to the broadcast their catalog came fro…
kixelated Aug 12, 2026
0910372
feat(net): announce namespaces unasked, with a SETUP opt-out (#2748)
kixelated Aug 12, 2026
71ea143
build(nix): package the Go toolchain so `just go check` actually runs…
kixelated Aug 12, 2026
9403501
fix(net): keep a live-edge subscriber at the live edge across a takeo…
arielmol Aug 12, 2026
976890b
feat(native): start dialing before the AAAA answer lands (#2749)
kixelated Aug 13, 2026
e643ced
fix(net): surface a dropped resume producer instead of stalling reade…
kixelated Aug 13, 2026
6d3c51d
fix(js/net): count advertisements per namespace instead of flagging t…
kixelated Aug 13, 2026
6db246b
chore: release (#2724)
moq-bot[bot] Aug 13, 2026
4622e86
fix(wasm): advertise ALPN so the browser can negotiate a version (#2811)
kixelated Aug 13, 2026
f91e3bb
chore: bump non-Rust package versions (#2821)
kixelated Aug 13, 2026
67c66d6
feat(libmoq): declare the catalog container for manually authored ren…
arielmol Aug 13, 2026
4a9c2ef
ci: stop compiling the tree three times, and cache what main builds (…
kixelated Aug 13, 2026
502fff5
fix(net): decode largest object in subscribe ok (#2837)
Glenn444 Aug 13, 2026
78b4b97
fix(moq-mux): anchor the TS table cadence to the media timeline (#2825)
kixelated Aug 13, 2026
8bbfd6c
feat(cli): run multiple import/export stages over one connection (#2809)
kixelated Aug 13, 2026
022319b
fix(net): wake a capped subscriber when its parked group is evicted (…
kixelated Aug 13, 2026
8ddfd78
feat(bindings): fetch and decode a retained media group (#2827)
kixelated Aug 14, 2026
4e59bf1
Merge remote-tracking branch 'origin/main' into dev
kixelated Aug 14, 2026
98719b1
Merge remote-tracking branch 'origin/main' into dev
kixelated Aug 14, 2026
20798fd
Merge remote-tracking branch 'origin/dev' into claude/merge-main-into…
kixelated Aug 14, 2026
403acc7
fix(merge): restore CMAF buffering in two tests and bound the JS anno…
kixelated Aug 14, 2026
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
1 change: 1 addition & 0 deletions .github/workflows/alert.yml
Original file line number Diff line number Diff line change
Expand Up @@ -22,6 +22,7 @@ on:
# on every push to every PR. `just gh check` enforces both halves of that.
workflows:
- apt-repo
- Cache
- Cachix
- Docker
- libmoq
Expand Down
110 changes: 110 additions & 0 deletions .github/workflows/cache.yml
Original file line number Diff line number Diff line change
@@ -0,0 +1,110 @@
name: Cache

# Populates the Rust build cache that check.yml restores, and the dev shell
# closure that every nix job substitutes.
#
# Actions scopes cache reads to the current branch plus the default branch, so
# only a cache written here on `main` is reachable from a pull request. `main`
# is therefore the single writer and pull requests are readers (`save-if` in
# check.yml): one cache everyone can reach, rather than per-PR entries that fill
# the repository's 10 GB budget while remaining private to the PR that wrote
# them.
#
# This runs the unscoped suite rather than the diff-scoped one: the point is to
# leave artifacts for whatever a future PR happens to touch, not to check `main`
# (the PR that became `main` was already checked). It blocks nobody.

permissions:
contents: read

on:
push:
branches: [main]
# A docs-only merge can't change the build graph, so don't spend a runner.
paths:
- "rs/**"
- "Cargo.toml"
- "Cargo.lock"
- "rust-toolchain.toml"
- "flake.nix"
- "flake.lock"
- ".github/workflows/cache.yml"
# Priming a cold cache by hand, e.g. after a toolchain bump evicts everything.
workflow_dispatch:

# Newest `main` wins; an older warm is worth less than the runner it holds.
concurrency:
group: cache-${{ github.ref }}
cancel-in-progress: true

jobs:
warm:
name: Warm
runs-on: ubuntu-24.04-arm
timeout-minutes: 60

steps:
- name: Free disk space
uses: jlumbroso/free-disk-space@54081f138730dfa15788a46383842cd2f914a1be # main
with:
tool-cache: false

- name: Checkout
uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
with:
persist-credentials: false

- uses: DeterminateSystems/nix-installer-action@1d87d45818068401a10cf16bdc5f00b24994a83f # main
with:
determinate: false
# Trust the flake's cachix substituter. `nixConfig` in flake.nix is
# untrusted by default, so nix logs "ignoring untrusted flake
# configuration setting 'extra-substituters'" and rebuilds from
# source anything missing from the public nixpkgs cache. That was
# costing 678s per job for uniffi-bindgen-go alone, in every job that
# enters the dev shell.
extra-conf: |
extra-substituters = https://kixelated.cachix.org
extra-trusted-public-keys = kixelated.cachix.org-1:CmFcV0lyM6KuVM2m9mih0q4SrAa0XyCsiM7GHrz3KKk=

# cachix.yml only fires on release tags and only pushes the tagged
# package, so the dev shell itself was never cached. Anything in it that
# nixpkgs doesn't serve prebuilt (uniffi-bindgen-go, a
# buildRustPackage from a git tag) was therefore compiled from source by
# every job that entered the shell. Push it once here so pull requests
# substitute it instead.
- uses: cachix/cachix-action@5f2d7c5294214f71b873db4b969586b980625e71 # v17
with:
name: kixelated
authToken: ${{ secrets.CACHIX_AUTH_TOKEN }}

- name: Cache the dev shell
run: |
# `--profile` realises the shell's full build closure, which is what
# the runners actually need; `nix build .#devShells...` alone leaves
# out the build-time inputs.
nix develop --profile /tmp/devshell --command true
cachix push kixelated /tmp/devshell

# `shared-key` replaces the automatic per-job key, which is what lets
# check.yml's differently-named jobs restore what this one saves.
- name: Rust Cache
uses: Swatinem/rust-cache@e18b497796c12c097a38f9edb9d0641fb99eee32 # v2
with:
shared-key: rust
cache-on-failure: true

# Both halves, so the saved cache carries the lint artifacts (clippy rmeta)
# and the test artifacts (codegen + linked test binaries). They share
# little, and a PR usually needs both.
- name: Check
run: nix develop --command just check-all
env:
MOQ_STRICT: 1

- name: Test
if: ${{ !cancelled() }}
run: nix develop --command just test all
env:
MOQ_STRICT: 1
NEXTEST_PROFILE: ci
67 changes: 60 additions & 7 deletions .github/workflows/check.yml
Original file line number Diff line number Diff line change
Expand Up @@ -17,14 +17,18 @@ concurrency:
cancel-in-progress: true

jobs:
# `check` and `test` are separate jobs because they share almost no build
# artifacts: clippy emits rmeta without codegen, while nextest codegens and
# links every test binary. Running them in sequence meant paying both in
# series for no reuse, so they run concurrently and the job takes the slower
# of the two rather than their sum.
check:
name: Check
# Skip the actual work on close/merge: this run exists only to cancel the
# superseded in-flight build via the concurrency group above.
if: github.event.action != 'closed'
runs-on: ubuntu-24.04-arm
# Backstop a hang (e.g. a wedged test) instead of holding the runner for the
# 6-hour default. Generous enough for a cold first build.
# Backstop a hang instead of holding the runner for the 6-hour default.
timeout-minutes: 60

steps:
Expand All @@ -43,17 +47,28 @@ jobs:
- uses: DeterminateSystems/nix-installer-action@1d87d45818068401a10cf16bdc5f00b24994a83f # main
with:
determinate: false
# Trust the flake's cachix substituter. `nixConfig` in flake.nix is
# untrusted by default, so nix logs "ignoring untrusted flake
# configuration setting 'extra-substituters'" and rebuilds from
# source anything missing from the public nixpkgs cache. That was
# costing 678s per job for uniffi-bindgen-go alone, in every job that
# enters the dev shell.
extra-conf: |
extra-substituters = https://kixelated.cachix.org
extra-trusted-public-keys = kixelated.cachix.org-1:CmFcV0lyM6KuVM2m9mih0q4SrAa0XyCsiM7GHrz3KKk=
- uses: DeterminateSystems/flake-checker-action@de924abd783455e8429c858962b9e43062d19da1 # main

# Cache Rust dependencies and build artifacts via the Actions cache.
# Restore only. cache.yml is the single writer, on `main`; see the comment
# there for why. `shared-key` has to match the one it saves under.
- name: Rust Cache
uses: Swatinem/rust-cache@e18b497796c12c097a38f9edb9d0641fb99eee32 # v2
with:
cache-on-failure: true
shared-key: rust
save-if: false

# The same two recipes a developer runs locally. Both diff against
# origin/$GITHUB_BASE_REF and touch only the packages this PR changed plus
# their dependents, so a one-crate PR compiles one crate.
# The same recipe a developer runs locally. It diffs against
# origin/$GITHUB_BASE_REF and touches only the packages this PR changed
# plus their dependents, so a one-crate PR compiles one crate.
#
# MOQ_STRICT turns every "tool missing, skipping" guard into an error.
# Locally an absent formatter means less gets checked; here it would mean
Expand All @@ -64,6 +79,44 @@ jobs:
env:
MOQ_STRICT: 1

test:
name: Test
if: github.event.action != 'closed'
runs-on: ubuntu-24.04-arm
timeout-minutes: 60

steps:
- name: Free disk space
uses: jlumbroso/free-disk-space@54081f138730dfa15788a46383842cd2f914a1be # main
with:
tool-cache: false

- name: Checkout
uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
with:
persist-credentials: false
# Full history so `just test` can diff against origin/$GITHUB_BASE_REF.
fetch-depth: 0

- uses: DeterminateSystems/nix-installer-action@1d87d45818068401a10cf16bdc5f00b24994a83f # main
with:
determinate: false
# Trust the flake's cachix substituter. `nixConfig` in flake.nix is
# untrusted by default, so nix logs "ignoring untrusted flake
# configuration setting 'extra-substituters'" and rebuilds from
# source anything missing from the public nixpkgs cache. That was
# costing 678s per job for uniffi-bindgen-go alone, in every job that
# enters the dev shell.
extra-conf: |
extra-substituters = https://kixelated.cachix.org
extra-trusted-public-keys = kixelated.cachix.org-1:CmFcV0lyM6KuVM2m9mih0q4SrAa0XyCsiM7GHrz3KKk=

- name: Rust Cache
uses: Swatinem/rust-cache@e18b497796c12c097a38f9edb9d0641fb99eee32 # v2
with:
shared-key: rust
save-if: false

# NEXTEST_PROFILE picks up the longer hang timeout in .config/nextest.toml;
# without it a runner under load could trip the local one.
- name: Test
Expand Down
6 changes: 4 additions & 2 deletions CLAUDE.md
Original file line number Diff line number Diff line change
Expand Up @@ -24,11 +24,13 @@ Use the Nix dev shell for project commands so local runs match CI tooling. If Ni

To force a base, `just check origin/dev` and `just fix origin/dev` take it positionally. `just test` can't: it's a module, so `just test origin/dev` looks for a *recipe* named `origin/dev`. Name the recipe to get past that: `just test default origin/dev`.

**CI runs exactly these recipes: `just check` then `just test`, with `MOQ_STRICT=1`.** There is no separate `just ci`, so there is no second definition of "checked" to drift from this one. The split is by cost, not by environment: `check` lints and compiles (plus `tsc -b` and the Python docs, which catch what `--noEmit` and autodoc can't), while `test` links and runs the test binaries, which is the expensive half.
**CI runs exactly these recipes: `just check` and `just test`, with `MOQ_STRICT=1`.** There is no separate `just ci`, so there is no second definition of "checked" to drift from this one. The split is by cost, not by environment: `check` lints and compiles (plus `tsc -b` and the Python docs, which catch what `--noEmit` and autodoc can't), while `test` links and runs the test binaries, which is the expensive half. They run as concurrent jobs in `check.yml`, because clippy emits rmeta without codegen while nextest codegens and links, so there is nearly nothing for one to reuse from the other.

The Rust build cache is written by `main` and read by pull requests, never the other way around (`.github/workflows/cache.yml`). Actions scopes cache reads to the current branch plus the default branch, so a PR-only workflow can never leave anything a *later* PR can restore. Both jobs in `check.yml` therefore restore under `shared-key: rust` with `save-if: false`. Don't add a save to a PR job: each one is gigabytes, the repository budget is 10 GB, and PR-scoped entries evict each other while remaining unreadable to everyone else.

`MOQ_STRICT` is the one thing CI does differently. Every tool the checks use is guarded with `command -v` so an incomplete local toolchain checks less instead of failing; in CI that would be a green run that silently checked nothing, so the variable turns the required set into an up-front precondition (`_tools` in the root justfile). Required is per scope, mirroring what the diff actually dispatches, so a docs-only PR doesn't have to have gradle.

Two tools stay unrequired: `swift` exists only on macOS (swift.yml is its gate), and **`go`/`uniffi-bindgen-go` are not in the dev shell**, so `just go check` skips itself in CI exactly as it always has. `uniffi-bindgen-go` isn't in nixpkgs (it installs from a NordSecurity git tag), so closing that hole means packaging it in the flake.
One tool stays unrequired: `swift` exists only on macOS, so swift.yml is its gate rather than the PR path.

Two gates live outside the PR path, in `.github/workflows/nightly.yml`: `just rs audit` (cargo-deny) because an advisory lands without this repo changing, and `just rs features` (the `--all-features` and `--no-default-features` compiles) because each is a full extra workspace compile that shares almost nothing with the default one. A break there lands on `main` rather than being caught in review, which is the accepted trade; anything that must block a merge belongs in `check`.

Expand Down
Loading
Loading