Skip to content
Tests

feat(eval): defer runtime-unknown eval/new Function to a throw-on-reach error by default (#5206) #5838

Sign in to view logs

feat(eval): defer runtime-unknown eval/new Function to a throw-on-reach error by default (#5206)

feat(eval): defer runtime-unknown eval/new Function to a throw-on-reach error by default (#5206) #5838

Workflow file for this run

.github/workflows/test.yml at 20c0995
name: Tests
on:
# Run on version tags so `release-packages.yml` can gate publish on a green
# Tests workflow for the exact commit being released. Direct pushes to main
# do NOT trigger tests — the gates that matter are PRs (pre-merge) and tags
# (pre-release).
push:
tags: ['v*']
pull_request:
branches: [main]
# `labeled` is here so the optional opt-in jobs (parity, compile-smoke,
# doc-tests) re-fire when a maintainer or the PR author applies the
# `run-extended-tests` label. Without it, applying the label on an
# existing PR wouldn't re-trigger the workflow.
types: [opened, synchronize, reopened, labeled]
paths-ignore:
- 'docs/src/**'
- '*.md'
- '!CLAUDE.md'
- '!CHANGELOG.md'
# Manual escape hatch for the opt-in jobs. Maintainers (write access)
# can dispatch the workflow against any ref with `run_extended_tests=true`
# to run parity / compile-smoke / package smokes / doc-tests on demand
# without tagging a release.
workflow_dispatch:
inputs:
run_extended_tests:
description: 'Run extended tests (parity, compile-smoke, package smokes, doc-tests)'
type: boolean
default: false
concurrency:
group: test-${{ github.ref }}
cancel-in-progress: true
env:
CARGO_TERM_COLOR: always
MACOSX_DEPLOYMENT_TARGET: "13.0"
jobs:
# ---------------------------------------------------------------------------
# Lint: cargo fmt --check (formatting gate for every PR)
# ---------------------------------------------------------------------------
lint:
# Was macos-14 — moved to ubuntu-latest in v0.5.428 since `cargo fmt
# --check` is portable. The 6 multiplier-min cut is small in absolute
# terms (lint runs in ~30s) but it's free.
runs-on: ubuntu-latest
timeout-minutes: 20
steps:
- uses: actions/checkout@v6
- name: Install Rust toolchain
uses: dtolnay/rust-toolchain@stable
with:
components: rustfmt, clippy
- uses: Swatinem/rust-cache@v2
with:
shared-key: "${{ runner.os }}-perry"
# PRs read from cache; only main writes new entries.
# Avoids cache thrash from short-lived branches.
save-if: ${{ github.ref == 'refs/heads/main' }}
- name: Check formatting
run: cargo fmt --all -- --check
# File-size gate (v0.5.1019): fails the PR if any tracked source
# file exceeds the LOC threshold (5000 initially; eventual target
# is 2000). Big single-file modules are hard to read, slow IDE +
# cargo-check incrementality, and hide regressions in code review.
# Allowlist + exclusions live in the script.
- name: File size limit
run: ./scripts/check_file_size.sh
# GC write-barrier store-site inventory: every raw heap-slot store in
# perry-codegen / perry-runtime / perry-stdlib must be barriered or
# carry a justified GC_STORE_AUDIT(...) marker (or a justified entry
# in scripts/gc_store_site_allowlist.txt). Catches new unbarriered
# old->young store paths before they become nondeterministic segfaults.
- name: GC store-site inventory
run: |
python3 scripts/gc_store_site_inventory.py --self-test
python3 scripts/gc_store_site_inventory.py
# Handle-vs-pointer address classification audit: POINTER_TAG payloads
# can be heap pointers OR small registry handles (fetch/zlib/proxy/...),
# and code must classify by magnitude through value/addr_class.rs before
# dereferencing. Catches new hand-typed band literals (0x100000 etc.)
# and new `as *const GcHeader` casts outside addr_class.rs / gc/ before
# they become Linux-only segfaults (#1843, #4004, #4665, #4800 class).
# Allowlist: scripts/addr_class_allowlist.txt.
- name: Address-classification audit
run: |
python3 scripts/addr_class_inventory.py --self-test
python3 scripts/addr_class_inventory.py
# ---------------------------------------------------------------------------
# API docs drift gate (#465)
#
# Regenerates `docs/src/api/reference.md` and `docs/api/perry.d.ts` from
# the compile-time manifest in `crates/perry-api-manifest/src/entries.rs`,
# then `git diff --exit-code`s the result. Fails when a code change updated
# the manifest without re-committing the artifacts. Closes the "Release
# workflow regenerates docs automatically (no drift from code)" criterion
# — committing the diff is the easiest way to keep the docs in sync,
# since the diff is reviewable in the PR that introduces it.
# ---------------------------------------------------------------------------
api-docs-drift:
runs-on: ubuntu-latest
timeout-minutes: 30
steps:
- uses: actions/checkout@v6
- name: Install Rust toolchain
uses: dtolnay/rust-toolchain@stable
- uses: Swatinem/rust-cache@v2
with:
shared-key: "${{ runner.os }}-perry"
save-if: ${{ github.ref == 'refs/heads/main' }}
- name: Regenerate API docs
run: ./scripts/regen_api_docs.sh
- name: Check for drift
run: |
if ! git diff --quiet -- docs/src/api/reference.md docs/api/perry.d.ts; then
echo ""
echo "::error::API docs drift detected. The compile-time manifest in"
echo "::error::crates/perry-api-manifest/src/entries.rs changed but the"
echo "::error::generated artifacts under docs/ weren't regenerated."
echo ""
echo "Fix by running:"
echo " ./scripts/regen_api_docs.sh"
echo " git add docs/src/api/reference.md docs/api/perry.d.ts"
echo " git commit -m 'docs: regenerate API reference + .d.ts'"
echo ""
echo "Diff:"
git --no-pager diff --stat -- docs/src/api/reference.md docs/api/perry.d.ts
echo ""
git --no-pager diff -- docs/src/api/reference.md docs/api/perry.d.ts | head -200
exit 1
fi
echo "✅ API docs match the manifest."
# ---------------------------------------------------------------------------
# Rust unit tests (266+ tests across all crates)
# ---------------------------------------------------------------------------
# Note: a separate `build` job that produced runtime/stdlib/compiler
# artifacts USED to live here. It only fed `binary-size` (which now does
# its own quick build) — every other job did `cargo build --release`
# itself anyway, so `needs: build` was a serializing barrier with no
# cache benefit. Removed in v0.5.387 along with the `actions/cache@v4`
# blocks (replaced by `Swatinem/rust-cache@v2`, which handles target/
# pruning intelligently and avoids the disk-pressure issue that
# required the manual simulator-runtime wipe + cache=registry-only
# workaround). Each downstream job below builds in parallel directly.
cargo-test:
# Was macos-14 — moved to ubuntu-latest in v0.5.392 to drop the 10×
# billing weight. cargo-test's `--exclude` list already filters out
# every macOS-specific UI crate (perry-ui-{ios,visionos,tvos,
# watchos,gtk4,android,windows} per the comment block below), so
# the platform-independent test set runs identically on Linux. The
# macOS-host coverage we lose here is negligible — these tests
# don't exercise any platform behavior; they're pure logic +
# codegen.
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v6
- name: Install Rust toolchain
uses: dtolnay/rust-toolchain@stable
- uses: Swatinem/rust-cache@v2
with:
shared-key: "${{ runner.os }}-perry"
save-if: ${{ github.ref == 'refs/heads/main' }}
- name: Run cargo test
# Exclude UI backends that don't build on the ubuntu-latest CI image:
# - perry-ui-macos / perry-ui-ios / perry-ui-tvos / perry-ui-watchos
# / perry-ui-visionos: depend on `objc2` which only compiles on
# Apple platforms (`compile_error!` in objc2/src/lib.rs:219).
# - perry-ui-gtk4: needs system pango/gtk via pkg-config; runner
# image doesn't have libgtk-4-dev installed by default.
# - perry-ui-android: needs Android NDK.
# - perry-ui-windows: needs win32 headers.
# - perry-ui-windows-winui: re-exports perry-ui-windows, so it inherits
# the same win32 / webview2-com dependency that won't build on Linux.
env:
# Rust's Ubuntu target can drive `cc` with `-fuse-ld=lld`; on the
# shared runner this has repeatedly terminated large test links with
# SIGBUS. Use the system linker for the cargo-test gate.
CARGO_TARGET_X86_64_UNKNOWN_LINUX_GNU_RUSTFLAGS: "-C linker-features=-lld"
# Keep test artifacts small enough for the shared runner disk. The
# cargo-test job does not need line tables, and debug info was enough
# to make later package archives hit ENOSPC after several package
# test builds accumulated in target/debug.
CARGO_PROFILE_TEST_DEBUG: "0"
CARGO_PROFILE_DEV_DEBUG: "0"
run: |
(
while sleep 60; do
echo "cargo-test still running at $(date -u +%Y-%m-%dT%H:%M:%SZ)"
done
) &
cargo_test_heartbeat_pid=$!
trap 'kill "$cargo_test_heartbeat_pid" 2>/dev/null || true' EXIT
# #1444: perry-runtime's tests share process-global state — the
# per-thread arena/GC, the timer queues, and the `NOTIFIED` flag are
# process singletons. Running them across the default test-harness
# thread pool lets one test's `js_notify_main_thread` / timer
# scheduling perturb another's wait budget (the event_pump timing
# flakes) and races the GC/threading tests into intermittent SIGSEGV.
# Run perry-runtime single-threaded so the tests can't interfere.
RUST_TEST_THREADS=1 cargo test -p perry-runtime
find target/debug/deps -maxdepth 1 -type f -perm -111 ! -name '*.so' -delete
# The remaining workspace includes large `perry` / `perry-stdlib`
# test binaries. Keep Cargo build jobs serialized so the runner
# does not link several of those large test binaries at once, then
# run packages one at a time and prune linked test executables so
# target/debug/deps does not exhaust the runner disk mid-job.
export CARGO_BUILD_JOBS=1
workspace_packages="$(
cargo metadata --no-deps --format-version 1 | python3 -c '
import json
import sys
excluded = {
"perry-runtime",
"perry-ui-macos",
"perry-ui-ios",
"perry-ui-visionos",
"perry-ui-tvos",
"perry-ui-watchos",
"perry-ui-gtk4",
"perry-ui-android",
"perry-ui-windows",
"perry-ui-windows-winui",
"perry-doc-fixture-my-bindings",
}
metadata = json.load(sys.stdin)
workspace_members = set(metadata["workspace_members"])
for package in metadata["packages"]:
if package["id"] in workspace_members and package["name"] not in excluded:
print(package["name"])
'
)"
for package in $workspace_packages; do
echo "::group::cargo test -p $package"
cargo test -p "$package"
echo "::endgroup::"
cargo clean -p "$package" || true
find target/debug/deps -maxdepth 1 -type f -perm -111 ! -name '*.so' -delete
done
# ---------------------------------------------------------------------------
# GC write-barrier stress (optional / non-blocking)
#
# `crates/perry/tests/gc_write_barrier_stress.rs` runs compiled binaries
# under the slowest GC configuration (PERRY_GC_FORCE_EVACUATE +
# PERRY_GC_VERIFY_EVACUATION) to hunt a *rare* corruption window (#5029).
# Those tests are ~200s each and nondeterministic by nature, so they are a
# poor fit for the blocking per-PR `cargo-test` gate (one flake blocked
# every unrelated PR). They are `#[ignore]`d there and run here instead.
#
# Opt-in + informational: `continue-on-error` so a flake never fails the
# workflow; triggered by the `run-extended-tests` PR label, a
# `workflow_dispatch` with `run_extended_tests=true`, or a tag push.
# ---------------------------------------------------------------------------
gc-stress:
continue-on-error: true
if: >-
github.event_name == 'push' ||
(github.event_name == 'workflow_dispatch' && inputs.run_extended_tests) ||
(github.event_name == 'pull_request' && contains(github.event.pull_request.labels.*.name, 'run-extended-tests'))
runs-on: ubuntu-latest
timeout-minutes: 30
steps:
- uses: actions/checkout@v6
- name: Install Rust toolchain
uses: dtolnay/rust-toolchain@stable
- uses: Swatinem/rust-cache@v2
with:
shared-key: "${{ runner.os }}-perry"
save-if: ${{ github.ref == 'refs/heads/main' }}
- name: Install clang
run: |
sudo apt-get update
sudo apt-get install -y clang
- name: Run GC write-barrier stress tests
env:
# Match the cargo-test gate's linker workaround (lld SIGBUS on the
# shared runner during large test links).
CARGO_TARGET_X86_64_UNKNOWN_LINUX_GNU_RUSTFLAGS: "-C linker-features=-lld"
run: cargo test -p perry --test gc_write_barrier_stress -- --ignored
# ---------------------------------------------------------------------------
# Compiler-output regression gate
#
# Retains HIR, pre/post-opt LLVM IR, assembly, benchmark output, runtime
# counters, vectorization remarks, benchmark timing summaries, and FP
# contraction evidence for the primary CPU benchmark plus numeric fixtures.
# Fails when hot-loop structural contracts regress.
# ---------------------------------------------------------------------------
compiler-output-regression:
runs-on: ubuntu-latest
timeout-minutes: 45
steps:
- uses: actions/checkout@v6
- name: Install Rust toolchain
uses: dtolnay/rust-toolchain@stable
- uses: Swatinem/rust-cache@v2
with:
shared-key: "${{ runner.os }}-perry"
save-if: ${{ github.ref == 'refs/heads/main' }}
- name: Install clang
run: |
sudo apt-get update
sudo apt-get install -y clang
- name: Build compiler
run: cargo build -p perry
- name: Run harness unit tests
run: python3 -m unittest tests.test_compiler_output_regression
- name: Gate native-region proof compiler output
run: |
python3 scripts/compiler_output_regression.py suite \
--suite native-region-proof \
--perry target/debug/perry \
--benchmark-mode smoke \
--runs 1 \
--perf-counters off \
--print-summary
- name: Gate positive vectorization compiler output
run: |
python3 scripts/compiler_output_regression.py capture \
--perry target/debug/perry \
--workload vectorized_buffer_transform \
--benchmark-mode smoke \
--runs 1 \
--perf-counters off \
--gate \
--print-summary
- name: Gate HIR fact rewrite compiler output
run: |
python3 scripts/compiler_output_regression.py capture \
--perry target/debug/perry \
--workload hir_fact_rewrite \
--benchmark-mode smoke \
--runs 1 \
--perf-counters off \
--gate \
--print-summary
- name: Gate FP contraction modes
run: |
python3 scripts/compiler_output_regression.py capture \
--perry target/debug/perry \
--workload fma_contract \
--benchmark-mode smoke \
--runs 1 \
--perf-counters off \
--gate \
--fp-contract=on \
--clang-arg=-march=haswell \
--expect-fma=on \
--out-dir target/compiler-output-regression/fma_contract-fp-contract-on
python3 scripts/compiler_output_regression.py capture \
--perry target/debug/perry \
--workload fma_contract \
--benchmark-mode smoke \
--runs 1 \
--perf-counters off \
--gate \
--fast-math \
--fp-contract=off \
--clang-arg=-march=haswell \
--expect-fma=off \
--out-dir target/compiler-output-regression/fma_contract-fast-no-contract
- name: Upload compiler-output artifacts
if: always()
uses: actions/upload-artifact@v7
with:
name: compiler-output-regression
path: target/compiler-output-regression/
# ---------------------------------------------------------------------------
# Parity tests (Perry output vs Node.js)
# ---------------------------------------------------------------------------
parity:
# Release-publish decoupling: aspirational extended suite. Per maintainer
# decision it no longer BLOCKS package publishing — release-packages.yml's
# await-tests gate keys on this workflow's run conclusion, and job-level
# `continue-on-error: true` keeps a red result here from failing that
# conclusion. The job still runs on every tag + shows its own pass/fail as
# an informational signal (and core jobs — cargo-test/lint/api-docs-drift/
# compiler-output-regression — still gate publish).
continue-on-error: true
# Was macos-14 — moved to ubuntu-latest in v0.5.392. Parity tests
# just compare Perry's stdout against `node --experimental-strip-types`'s
# stdout per test file; both run cleanly on Linux. `gtimeout` on
# macOS is `timeout` on Linux (the run_parity_tests.sh wrapper
# detects either). Node 22+ is installed via setup-node@v4 below.
# 10× billing weight cut.
#
# v0.5.1018: gated to tag pushes only (`github.event_name == 'push'`).
# The pull_request trigger above still fires the workflow on PRs for
# the lint / cargo-test / api-docs-drift gates, but parity now only
# runs on release tags. Direct main commits + PR cycle no longer pay
# the ~20 min parity bill; release tagging still catches regressions
# before publish (release-packages.yml await-tests gate waits on
# this job by name for tag events).
#
# Opt-in: apply the `run-extended-tests` label to a PR, or dispatch
# the workflow manually with `run_extended_tests=true`, to run this
# job on demand. PR authors and maintainers can both apply labels.
if: >-
github.event_name == 'push' ||
(github.event_name == 'workflow_dispatch' && inputs.run_extended_tests) ||
(github.event_name == 'pull_request' && contains(github.event.pull_request.labels.*.name, 'run-extended-tests'))
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v6
- name: Install Rust toolchain
uses: dtolnay/rust-toolchain@stable
- uses: Swatinem/rust-cache@v2
with:
shared-key: "${{ runner.os }}-perry"
save-if: ${{ github.ref == 'refs/heads/main' }}
- name: Setup Node.js
uses: actions/setup-node@v6
with:
node-version: '22'
- name: Build compiler
run: cargo build --release
- name: Run parity tests
run: ./run_parity_tests.sh
- name: Check parity threshold
run: |
set +e
python3 scripts/parity_threshold_gate.py \
--check \
--output-json test-parity/reports/parity_threshold_latest.json \
--output-md test-parity/reports/parity_threshold_latest.md
status=$?
set -e
cat test-parity/reports/parity_threshold_latest.md >> "$GITHUB_STEP_SUMMARY"
exit "$status"
- name: Check for new failures
run: |
REPORT="test-parity/reports/latest.json"
KNOWN="test-parity/known_failures.json"
# Filter empty strings — `run_parity_tests.sh` emits `compile: [""]`
# when there are zero compile failures (a printf+sed quirk in the
# JSON generator), and that empty entry would propagate to a
# spurious "NEW FAILURES: - " line and fail this gate.
jq -r '(.failures.parity // []) + (.failures.compile // []) | .[] | select(. != "")' "$REPORT" | sort -u > /tmp/all_fails.txt
if [[ -f "$KNOWN" ]]; then
# Issue #797 — known_failures.json moved from flat strings to
# structured records. Keep the audit metadata (the `_schema`
# key at the top of the file) out of the test-name set so
# CI doesn't try to match a real test against it.
jq -r 'keys[] | select(. != "_schema")' "$KNOWN" | sort -u > /tmp/known.txt
# Schema sanity check — every non-_schema entry must be an
# object with non-empty `category` and `reason`. Fails the
# build on malformed entries so the format doesn't silently
# drift back to the legacy flat-string shape.
bad="$(jq -r 'to_entries | map(select(.key != "_schema")) | .[] | select((.value | type) != "object" or (.value.category // "") == "" or (.value.reason // "") == "") | .key' "$KNOWN")"
if [[ -n "$bad" ]]; then
echo "Malformed known_failures.json entries (missing category/reason or not an object):"
echo "$bad" | sed 's/^/ - /'
exit 1
fi
else
: > /tmp/known.txt
fi
TOTAL=$(wc -l < /tmp/all_fails.txt | tr -d ' ')
comm -23 /tmp/all_fails.txt /tmp/known.txt > /tmp/new.txt
if [[ -s /tmp/new.txt ]]; then
echo "NEW FAILURES (not in known_failures.json):"
sed 's/^/ - /' /tmp/new.txt
exit 1
fi
echo "All ${TOTAL} failures are known/triaged."
- name: Generate parity matrix trend
run: |
python3 scripts/parity_matrix_trend.py \
--check \
--output-json test-parity/reports/parity_matrix_latest.json \
--output-md test-parity/reports/parity_matrix_latest.md
cat test-parity/reports/parity_matrix_latest.md >> "$GITHUB_STEP_SUMMARY"
- name: Upload parity report
if: always()
uses: actions/upload-artifact@v7
with:
name: parity-report
path: |
test-parity/reports/latest.json
test-parity/reports/parity_threshold_latest.json
test-parity/reports/parity_threshold_latest.md
test-parity/reports/parity_matrix_latest.json
test-parity/reports/parity_matrix_latest.md
# Capture per-test compile stderr written by run_parity_tests.sh into
# `test-parity/output/*.compile_error.log` so the long-tail
# macOS-14-only compile failures (tracked as `ci-env` in
# known_failures.json) can finally be diagnosed by reading the actual
# error message rather than inferring from the test family.
- name: Upload compile-error logs
if: always()
uses: actions/upload-artifact@v7
with:
name: parity-compile-errors-${{ runner.os }}
path: test-parity/output/*.compile_error.log
if-no-files-found: ignore
# ---------------------------------------------------------------------------
# Compile smoke test (all 130+ test files must compile)
# ---------------------------------------------------------------------------
compile-smoke:
# Release-publish decoupling (see `parity` above): aspirational extended
# suite, informational only — does not block package publishing.
continue-on-error: true
# Was macos-14 — moved to ubuntu-latest in v0.5.392. The smoke
# compiles every `test-files/*.ts` with the bare `perry foo.ts -o
# out` path; the auto-optimize cache + clang link steps work
# identically on Linux. The v0.5.385 sha256 sidecar is portable
# via `command -v sha256sum || shasum -a 256`. `xargs -P` is
# GNU on Linux (the macos-14 BSD xargs we tuned for behaves the
# same for our usage). Linux runners are 4-vCPU vs macos-14's 3,
# so could try NJOBS=4, but keeping NJOBS=3 + retry conservatively
# for the cargo auto-optimize race (issue tracked separately).
# 10× billing weight cut.
#
# v0.5.1018: gated to tag pushes only (`github.event_name == 'push'`).
# See the parity job comment above for rationale — release-packages.yml
# still requires this job on tag events before publishing.
#
# Opt-in: `run-extended-tests` PR label or `workflow_dispatch` with
# `run_extended_tests=true` runs this job on demand.
if: >-
github.event_name == 'push' ||
(github.event_name == 'workflow_dispatch' && inputs.run_extended_tests) ||
(github.event_name == 'pull_request' && contains(github.event.pull_request.labels.*.name, 'run-extended-tests'))
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v6
- name: Install Rust toolchain
uses: dtolnay/rust-toolchain@stable
- uses: Swatinem/rust-cache@v2
with:
shared-key: "${{ runner.os }}-perry"
save-if: ${{ github.ref == 'refs/heads/main' }}
- name: Build compiler
run: cargo build --release
- name: Issue #945 scalar method IR guard
run: |
PERRY_BIN="$PWD/target/release/perry" \
scripts/run_issue_945_scalar_method_ir_guard.sh
- name: Compile all test files
run: |
set -uo pipefail
export PERRY="$PWD/target/release/perry"
export LOGS_DIR="/tmp/perry_smoke_logs"
# Tests that are known to not compile cleanly under the bare
# `perry foo.ts -o out` smoke path. Sources:
# - test_ui_*: need `--target macos` / `ios-simulator` to pull
# in libperry_ui_*; the no-target compile path doesn't link
# the platform widgets.
# - test_timer: hangs on the runtime event loop under the
# no-arg compile path.
#
# The 11-entry Buffer/typed-array `ci-env` skip family that
# previously lived here (test_gap_buffer_ops, test_buffer_*,
# test_inline_uint8array_param, test_issue_167_*,
# test_issue_227_*, test_gap_fetch_response,
# test_issue_234_blob_methods, etc.) has been removed as of
# PR #239: the actual root cause was a missing
# `module.declare_function("llvm.assume", VOID, &[I1])` in
# `crates/perry-codegen/src/runtime_decls.rs`. Apple Clang
# ≥21 (Xcode 26 / local) auto-recognised the intrinsic;
# Apple Clang 15 (macos-14 runner / Xcode 15.x / LLVM 17)
# errored with `error: use of undefined value '@llvm.assume'`.
# Space-separated skip list (associative arrays require bash
# 4+; macOS-14 ships bash 3.2). Word-boundary match keeps the
# substring lookup safe.
# test_phase2v3_3_show_toast_set_text imports `setText` and
# `showToast` from perry/ui. The cross-platform stubs in
# perry-runtime/src/ui_text_registry.rs ARE meant to make the
# bare-target compile path work (no `--target` flag → host
# build), but the macOS doc-tests + this Linux compile-smoke
# both still fail at link time because the runtime registers
# the macOS-side handler in `perry-ui-macos/src/app_run.rs`,
# which is only linked when `needs_ui = true`. Without
# `--target macos`, perry-ui-macos isn't pulled in, and the
# cross-platform stubs route to a NULL handler → undefined
# symbol. The fix is a separate coordination work item with
# the v3.3 toast/setText worker (PR #322 family); skip-listing
# here unblocks the v9 CI smoke landing without papering over
# the underlying gap.
# test_ramda_user_import intentionally imports a user package
# (`ramda`) through the V8 fallback path. The compile-smoke
# runner does not npm-install optional package fixtures, so the
# strict unresolved-namespace diagnostic is expected here.
# test_take_screenshot imports perry/ui (same `--target gtk4`
# / pre-built libperry_ui_gtk4.a coupling as the test_ui_*
# family above — bare-target compile path doesn't link the
# platform widgets on Linux).
# test_issue_842_side_effect_dynamic_import compiles a barrel
# file that dynamic-imports a sibling helper; the smoke
# harness compiles each .ts standalone with `perry foo.ts -o
# out`, so the helper .o never gets produced and ld fails on
# the unresolved symbol. The test belongs in a multi-file
# integration runner, not the per-file smoke pass.
# test_jose_signverify_roundtrip references `jwtVerify` from
# the jose ext (#1025 sibling work). The bare smoke compile
# path doesn't link the jose-specific runtime symbol — same
# ld-unresolved-reference failure observed on #1038 pre-merge.
# Move to a jose-aware test runner once that crate's CI hook
# is wired up.
# test_ui_on_keydown_smoke / test_issue_1495_image_systemname /
# test_issue_1867_audio_playback / test_issue_2022_canvas_draw_image
# all import `perry/ui` (App/Image/Canvas/loadImage/keydown). Like the
# rest of the test_ui_* / media family above they need `--target
# macos` to link the platform widgets; the bare `perry foo.ts -o out`
# smoke path doesn't pull in libperry_ui_* on Linux, so they fail with
# ld undefined-symbol errors. ci-env, not a Perry codegen bug.
export SKIP_TESTS=" \
test_ui_comprehensive \
test_ui_controls \
test_ui_phase4 \
test_ui_on_keydown_smoke \
test_issue_1495_image_systemname \
test_issue_1867_audio_playback \
test_issue_2022_canvas_draw_image \
test_timer \
test_phase2v3_3_show_toast_set_text \
test_issue_351_media_playback \
test_issue_442_inline_button_bg \
test_issue_538_background_tasks \
test_issue_553_mobile_widgets \
test_issue_556_table_array \
test_issue_556_table_concat \
test_issue_610_foreach \
test_issue_610_smoke \
test_issue_640_navstack_textfield \
test_issue_763_reactive_textfield \
test_issue_764_state_at_module_init \
test_ramda_user_import \
test_take_screenshot \
test_issue_842_side_effect_dynamic_import \
test_jose_signverify_roundtrip \
test_parity_assert \
test_parity_async_hooks \
test_parity_buffer \
test_parity_child_process \
test_parity_cluster \
test_parity_crypto \
test_parity_dgram \
test_parity_diagnostics_channel \
test_parity_decimal \
test_parity_dns \
test_parity_dns_promises \
test_parity_dotenv \
test_parity_events \
test_parity_fs \
test_parity_fs_promises \
test_parity_http \
test_parity_http2 \
test_parity_https \
test_parity_lodash \
test_parity_module \
test_parity_moment \
test_parity_net \
test_parity_path \
test_parity_perf_hooks \
test_parity_process \
test_parity_querystring \
test_parity_readline \
test_parity_readline_promises \
test_parity_stream \
test_parity_stream_consumers \
test_parity_stream_promises \
test_parity_stream_web \
test_parity_sys \
test_parity_test \
test_parity_timers \
test_parity_timers_promises \
test_parity_tls \
test_parity_url \
test_parity_util \
test_parity_validator \
test_parity_worker_threads \
test_parity_zlib "
rm -rf "$LOGS_DIR"
mkdir -p "$LOGS_DIR"
# Worker function. Each test owns a unique marker filename
# (.pass / .fail / .skip) under $LOGS_DIR, so concurrent
# workers never race on the same file. Counts + failure list
# are aggregated below AFTER all workers finish — no shared
# bash-counter state crosses subshell boundaries. Per-test
# stderr is captured to $LOGS_DIR/<name>.compile_error.log so
# the artifact upload step preserves the actual error
# messages (long-tail macOS-14 codegen failures are tracked
# as `ci-env` in test-parity/known_failures.json and are
# otherwise diagnosed by inference, not data).
compile_one() {
local f="$1"
[[ -d "$f" ]] && return 0
local name
name=$(basename "$f" .ts)
if [[ "$SKIP_TESTS" == *" $name "* ]]; then
: > "$LOGS_DIR/${name}.skip"
return 0
fi
local err_log="$LOGS_DIR/${name}.compile_error.log"
# Try once. If perry compile fails, sleep 2s then retry.
# The xargs -P parallel pass can race when two workers both
# need an auto-optimize rebuild of the same feature combo —
# the loser's clang sees a momentarily-missing
# `target/perry-auto-<hash>/release/libperry_runtime.a` and
# bails with `errno=2`. Race window is sub-second so a
# single retry after a brief delay is the cheapest fix
# without serializing the workers entirely. `||` short-
# circuits on first success — the success path is unchanged.
try_compile() {
"$PERRY" "$f" -o "/tmp/perry_smoke_${name}" 2>"$err_log"
}
try_compile && status=0 || status=$?
if [[ $status -ne 0 ]]; then
# Retry-on-race semantics: the xargs -P parallel pass can
# race when two workers both need an auto-optimize rebuild
# of the same feature combo — the loser's clang sees a
# momentarily-missing libperry_runtime.a and bails. A single
# retry after a brief delay is the cheapest fix.
sleep 2
try_compile && status=0 || status=$?
fi
if [[ $status -eq 0 ]]; then
: > "$LOGS_DIR/${name}.pass"
rm -f "/tmp/perry_smoke_${name}" "$err_log"
else
: > "$LOGS_DIR/${name}.fail"
fi
}
export -f compile_one
# ubuntu-latest runners have 4 vCPUs; perry compile is
# CPU-bound for HIR/codegen but waits on the linker (clang)
# for a meaningful chunk of each test. v0.5.384 dropped to
# NJOBS=3 after NJOBS=6 hit a cargo auto-optimize file-lock
# race: two workers rebuilding the same `target/perry-auto-
# <hash>/lib*.a` led to one worker's clang seeing `errno=2`
# mid-link. v0.5.429 closed that race at the source via an
# OS file lock in `commands/compile/optimized_libs.rs::
# build_optimized_libs` (fslock dep — flock on Unix,
# LockFileEx on Windows; serializes per-hash, parallel across
# different hashes). NJOBS=6 is now safe again. Sequential
# baseline was ~26 min; NJOBS=3 → ~10-12 min; NJOBS=6 → ~6-8
# min on a 4-vCPU runner. The retry-once in compile_one stays
# as a belt-and-suspenders safety net for any remaining race
# corner the lock doesn't catch.
NJOBS="${PERRY_SMOKE_JOBS:-6}"
printf '%s\n' test-files/*.ts \
| xargs -P "$NJOBS" -n 1 -I{} bash -c 'compile_one "$@"' _ {}
# Count markers via shopt nullglob + bash array length. Pre-fix
# we used `ls -1 "$LOGS_DIR"/*.fail | wc -l` which fails when
# no matches exist (`ls` exits 1 on missing files), and with
# GH Actions' default `bash -eo pipefail` the failed pipe
# propagates errexit and kills the script BEFORE printing the
# summary line — so a clean run with zero failures still made
# compile-smoke exit 1 (PR #285 / v0.5.379 introduced this).
# nullglob makes an empty glob expand to nothing instead of
# the literal pattern, so the array length is correctly 0.
shopt -s nullglob
pass_files=("$LOGS_DIR"/*.pass)
fail_files=("$LOGS_DIR"/*.fail)
skip_files=("$LOGS_DIR"/*.skip)
PASS=${#pass_files[@]}
FAIL=${#fail_files[@]}
SKIP=${#skip_files[@]}
echo "Compile smoke: $PASS passed, $FAIL failed, $SKIP skipped"
if [[ $FAIL -gt 0 ]]; then
echo "Compile failures:"
for marker in "$LOGS_DIR"/*.fail; do
[[ -e "$marker" ]] || continue
name=$(basename "$marker" .fail)
echo " - $name"
# Surface the head of each failure log directly in the job
# output so a quick scan reveals the underlying error
# without downloading the artifact.
err_log="$LOGS_DIR/${name}.compile_error.log"
if [[ -s "$err_log" ]]; then
echo " --- compile stderr (first 30 lines) ---"
head -n 30 "$err_log" | sed 's/^/ /'
echo " --- end ---"
fi
done
exit 1
fi
- name: Upload compile-smoke error logs
if: always()
uses: actions/upload-artifact@v7
with:
name: compile-smoke-error-logs
path: /tmp/perry_smoke_logs/*.compile_error.log
if-no-files-found: ignore
# Thread-primitive compile-error tests (#146): checks that closures
# passed to perry/thread primitives with outer-variable writes are
# rejected. The runtime thread_primitives.ts example is covered by
# the doc-tests job below.
- name: Thread-primitive compile-error tests
run: ./scripts/run_thread_tests.sh
# perry/ui styling-matrix CI gate (Phase A of issue #185): verifies
# crates/perry-ui/src/styling_matrix.rs is in sync with every backend's
# lib.rs FFI exports, regenerates docs/src/ui/styling-matrix.md, then
# `git diff --exit-code` catches a forgotten-to-commit regeneration.
# Drift fails CI loudly so a future FFI add/remove can't silently
# land without a matrix update.
- name: UI styling matrix
run: |
./scripts/run_ui_styling_matrix.sh
git diff --exit-code -- docs/src/ui/styling-matrix.md \
|| (echo "docs/src/ui/styling-matrix.md regenerated; commit the diff" && exit 1)
# Visual styling test ↔ spec consistency (#185 follow-up).
# `docs/examples/ui/styling/visual_test.ts` is the canonical
# comprehensive visual test app; `visual_test.spec.md` documents
# each cell's expected visible signature for human/LLM-aided
# screenshot verification. The two files must stay in lockstep
# — adding a row to the .ts without updating the spec silently
# breaks the verification flow. The actual cross-platform
# compile-test of visual_test.ts is handled by the existing
# run_doc_tests.sh loop downstream.
- name: Visual styling test ↔ spec consistency
run: ./scripts/run_visual_test_check.sh
# Fastify end-to-end integration (#174): launches a Perry-compiled
# Fastify server as a background process, curls four routes covering
# simple GET, path params, POST with JSON body + reply.code(), and
# 404 fallback. The docs Fastify example is marked no-test because
# app.listen() blocks forever; this script provides the coverage
# that the no-test tag would otherwise hide.
- name: Fastify integration tests
run: ./scripts/run_fastify_tests.sh
# Memory-stability regression suite. Two failure modes microbenchs
# don't catch: (1) slow RSS accumulation across 100k-200k iterations
# of allocate-and-discard (would catch a future block-pinning /
# cache-leak / tenuring-trap regression in the gen-GC work), and
# (2) crashes when gc() is forced aggressively during JSON parse,
# deep recursion, or closure init. Each test runs under default,
# PERRY_GEN_GC=1, and PERRY_GEN_GC=1 PERRY_WRITE_BARRIERS=1 so a
# regression in any GC mode is caught. Linux-only because /usr/bin/time
# availability + RSS reporting differs on Windows runners.
- name: Memory stability tests
if: runner.os == 'Linux' || runner.os == 'macOS'
env:
PERRY_GC_EVIDENCE_DIR: ${{ runner.temp }}/gc-evidence
run: ./scripts/run_memory_stability_tests.sh
- name: Upload GC evidence artifacts
if: always()
uses: actions/upload-artifact@v7
with:
name: gc-evidence-${{ runner.os }}
path: ${{ runner.temp }}/gc-evidence
if-no-files-found: ignore
# ---------------------------------------------------------------------------
# HarmonyOS ArkUI codegen smoke (Phase 2 v9).
#
# The harmonyos compile path produces a 3-part output: the .so (LLVM
# codegen, just like every other backend), the ArkUI Index.ets (emitted
# by perry-codegen-arkts from the harvested perry/ui App({...}) call),
# and the NAPI bridge declarations. End-to-end `perry compile --target
# harmonyos` requires the OpenHarmony SDK (clang + musl sysroot, ~600
# MB) which isn't pre-installed on ubuntu-latest runners and isn't
# worth downloading every CI run.
#
# What CI CAN cover without the SDK is the codegen-side ArkUI emission
# — the part that's most likely to regress as Phase 2 widgets evolve.
# `crates/perry-codegen-arkts/tests/phase2_full_app_smoke.rs` is the
# comprehensive integration test: constructs a single Module that uses
# every Phase 2 widget shape (state<T>, Tabs, Menu, Grid, LazyVStack
# with .map, ForEach via array.map, inline style: { } with animation/
# shadow/textDecoration, @app.media image, Toggle/TextField/Slider with
# multi-arg invokeCallback1 closures) and asserts the emitted Index.ets
# contains every canonical pattern v2-v13 added.
#
# Discrete job (separate from cargo-test) so a regression in just one
# widget surfaces as one red cell, not buried in the workspace test
# output. cargo-test ALSO runs these tests as part of `cargo test
# --workspace` — this job is the named visibility, not a duplicate run.
#
# Linker-side validation is covered by manual on-device runs against
# DevEco Studio's Pura 90 Pro Max emulator (see CLAUDE.md v0.5.399+
# entries for the workflow).
# ---------------------------------------------------------------------------
harmonyos-smoke:
# Aspirational smoke (informational) — must not block package publish.
# release-packages await-tests keys on this workflow's run conclusion;
# continue-on-error keeps a red result here from failing it (same as
# parity/compile-smoke/doc-tests/drizzle/effect-basic-smoke). Core jobs
# (cargo-test/lint/api-docs-drift/compiler-output-regression) still gate.
continue-on-error: true
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v6
- name: Install Rust toolchain
uses: dtolnay/rust-toolchain@stable
- uses: Swatinem/rust-cache@v2
with:
shared-key: "${{ runner.os }}-perry"
save-if: ${{ github.ref == 'refs/heads/main' }}
- name: Run perry-codegen-arkts unit tests
run: cargo test -p perry-codegen-arkts --release --lib
- name: Run Phase 2 full-app integration smoke
run: |
cargo test -p perry-codegen-arkts \
--release \
--test phase2_full_app_smoke \
-- --nocapture
# ---------------------------------------------------------------------------
# drizzle-mysql smoke: runs the tier-3 release fixture
# `tests/release/packages/drizzle-mysql/` against a real MySQL — the CI
# counterpart of #489's local acceptance, closing #804.
#
# The fixture itself stays Docker-free (per the tier-3 "no Docker"
# preference in scripts/release_sweep_tiers/tier03_real_packages.sh —
# locally it skips when no mysqld is reachable on 127.0.0.1:3306). The
# `services: mysql:8` block below is runner-side setup, transparent to
# the fixture. `MYSQL_ALLOW_EMPTY_PASSWORD` matches the fixture's
# root-no-password convention; `MYSQL_DATABASE` auto-creates the test
# DB on init so the fixture's CREATE-IF-NOT-EXISTS is a no-op.
#
# Gated to tag pushes + opt-in (parity with compile-smoke / doc-tests
# / parity). Real-DB setup + perry release build + drizzle +
# @perryts/mysql compile is ~15 min wall — PRs shouldn't pay it by
# default. Opt-in via the `run-extended-tests` label.
# ---------------------------------------------------------------------------
drizzle-mysql-smoke:
# Release-publish decoupling (see `parity` above): aspirational extended
# suite, informational only — does not block package publishing.
continue-on-error: true
if: >-
github.event_name == 'push' ||
(github.event_name == 'workflow_dispatch' && inputs.run_extended_tests) ||
(github.event_name == 'pull_request' && contains(github.event.pull_request.labels.*.name, 'run-extended-tests'))
runs-on: ubuntu-latest
services:
mysql:
image: mysql:8
env:
MYSQL_ALLOW_EMPTY_PASSWORD: "yes"
MYSQL_DATABASE: perry_drizzle_test
ports:
- 3306:3306
options: >-
--health-cmd="mysqladmin ping -h 127.0.0.1 -P 3306 --silent"
--health-interval=5s
--health-timeout=3s
--health-retries=20
steps:
- uses: actions/checkout@v6
- name: Install Rust toolchain
uses: dtolnay/rust-toolchain@stable
- uses: Swatinem/rust-cache@v2
with:
shared-key: "${{ runner.os }}-perry"
save-if: ${{ github.ref == 'refs/heads/main' }}
- name: Setup Node.js
uses: actions/setup-node@v6
with:
node-version: '22'
- name: Install mysql client (for fixture health probe)
run: |
sudo apt-get update -qq
sudo apt-get install -y -qq mysql-client
- name: Wait for MySQL service to accept connections
run: |
set -e
for i in $(seq 1 30); do
if mysql -h 127.0.0.1 -P 3306 -u root -e "SELECT 1" >/dev/null 2>&1; then
echo "mysql ready after $i attempts"
break
fi
sleep 1
done
mysql -h 127.0.0.1 -P 3306 -u root -e "SELECT VERSION();"
- name: Build perry compiler
run: cargo build --release -p perry-runtime -p perry-stdlib -p perry
- name: Run drizzle-mysql fixture
run: |
cd tests/release/packages/drizzle-mysql
PERRY_BIN="$GITHUB_WORKSPACE/target/release/perry" bash fixture.sh
# ---------------------------------------------------------------------------
# ink-link-smoke: runs the tier-3 release fixture
# `tests/release/packages/ink-link-smoke/` as the CI counterpart to #803.
#
# This is intentionally compile/link-only. #348 tracks broader Ink runtime
# and rendering compatibility; this job guards the package-graph +
# compilePackages linker contract that the fixture documents.
#
# Gated to tag pushes + opt-in (parity with drizzle-mysql-smoke /
# compile-smoke / doc-tests). The fixture installs Ink + React and builds a
# release Perry compiler, so PRs shouldn't pay it by default. Opt-in via the
# `run-extended-tests` label or workflow dispatch.
# ---------------------------------------------------------------------------
ink-link-smoke:
# Aspirational smoke (informational) — see harmonyos-smoke. Does not block publish.
continue-on-error: true
if: >-
github.event_name == 'push' ||
(github.event_name == 'workflow_dispatch' && inputs.run_extended_tests) ||
(github.event_name == 'pull_request' && contains(github.event.pull_request.labels.*.name, 'run-extended-tests'))
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v6
- name: Install Rust toolchain
uses: dtolnay/rust-toolchain@stable
- uses: Swatinem/rust-cache@v2
with:
shared-key: "${{ runner.os }}-perry"
save-if: ${{ github.ref == 'refs/heads/main' }}
- name: Setup Node.js
uses: actions/setup-node@v6
with:
node-version: '22'
- name: Build perry compiler
run: cargo build --release -p perry-runtime -p perry-stdlib -p perry
- name: Run Ink link fixture
run: |
cd tests/release/packages/ink-link-smoke
PERRY_BIN="$GITHUB_WORKSPACE/target/release/perry" bash fixture.sh
- name: Upload Ink fixture logs
if: always()
uses: actions/upload-artifact@v7
with:
name: ink-link-smoke-logs
path: |
tests/release/packages/ink-link-smoke/install.log
tests/release/packages/ink-link-smoke/perry-compile.log
if-no-files-found: ignore
# ---------------------------------------------------------------------------
# effect-basic-smoke: runs the tier-3 release fixture
# `tests/release/packages/effect-basic/` as the CI counterpart to #802.
#
# This is intentionally advisory: #802 asks for a live Effect compile/run
# signal even while broader Effect end-to-end compatibility remains in
# progress. Gated to tag pushes + opt-in, matching the named package smokes.
# ---------------------------------------------------------------------------
effect-basic-smoke:
continue-on-error: true
if: >-
github.event_name == 'push' ||
(github.event_name == 'workflow_dispatch' && inputs.run_extended_tests) ||
(github.event_name == 'pull_request' && contains(github.event.pull_request.labels.*.name, 'run-extended-tests'))
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v6
- name: Install Rust toolchain
uses: dtolnay/rust-toolchain@stable
- uses: Swatinem/rust-cache@v2
with:
shared-key: "${{ runner.os }}-perry"
save-if: ${{ github.ref == 'refs/heads/main' }}
- name: Setup Node.js
uses: actions/setup-node@v6
with:
node-version: '22'
- name: Build perry compiler
run: cargo build --release -p perry-runtime -p perry-stdlib -p perry
- name: Run Effect fixture
run: |
cd tests/release/packages/effect-basic
PERRY_EFFECT_BASIC_ADVISORY=1 PERRY_BIN="$GITHUB_WORKSPACE/target/release/perry" bash fixture.sh
- name: Upload Effect fixture logs
if: always()
uses: actions/upload-artifact@v7
with:
name: effect-basic-smoke-logs
path: |
tests/release/packages/effect-basic/install.log
tests/release/packages/effect-basic/perry-compile.log
tests/release/packages/effect-basic/perry-run.log
tests/release/packages/effect-basic/perry-out.txt
tests/release/packages/effect-basic/diff.log
if-no-files-found: ignore
# ---------------------------------------------------------------------------
# Doc-example tests: compile + run every .ts under docs/examples/.
# UI examples launch with PERRY_UI_TEST_MODE=1 so they auto-exit after one
# frame. Gallery screenshots are diffed against per-OS baselines (advisory
# until Linux/Windows baselines stabilize).
#
# Gated to tag pushes + opt-in (parity with parity/compile-smoke). The
# macOS-14 matrix entry takes ~30 min wall and dominates the PR feedback
# loop, so PRs no longer pay the bill by default. Release tags still run
# doc-tests as part of the release-packages.yml gate.
#
# Opt-in: apply the `run-extended-tests` label to a PR, or dispatch the
# workflow manually with `run_extended_tests=true`, to run this job on
# demand. PR authors and maintainers can both apply labels.
# ---------------------------------------------------------------------------
doc-tests:
# Release-publish decoupling (see `parity` above): aspirational extended
# suite, informational only — does not block package publishing.
continue-on-error: true
if: >-
github.event_name == 'push' ||
(github.event_name == 'workflow_dispatch' && inputs.run_extended_tests) ||
(github.event_name == 'pull_request' && contains(github.event.pull_request.labels.*.name, 'run-extended-tests'))
strategy:
fail-fast: false
matrix:
include:
- os: macos-14
ui_backend: perry-ui-macos
shell: bash
# stdlib/http/snippets.ts excluded since v0.5.886: it links
# against js_axios_response_data_parsed +
# js_node_http2_create_secure_server which live in
# perry-ext-axios / perry-ext-http-server. v0.5.885's
# PERRY_NO_AUTO_OPTIMIZE skips the well-known-binding probe
# that would route those .a files into the link surface.
# Proper fix: hoist well-known-binding lookup out of
# build_optimized_libs into link.rs so it runs even when
# auto-optimize is skipped. Tracked separately.
cmd_exclude_gallery: "./scripts/run_doc_tests.sh --verbose --skip-xcompile --filter-exclude ui/gallery.ts --filter-exclude stdlib/http/snippets.ts"
cmd_gallery: "./scripts/run_doc_tests.sh --verbose --skip-xcompile --filter ui/gallery.ts"
# Repeat `--xcompile-only-target=…` per target rather than a
# single comma-delimited value because PowerShell splits even
# `--foo=a,b` at the comma when unquoted (array literal).
# Repetition sidesteps the whole issue on every shell.
#
# web + wasm cross-compile dropped from the macOS blocking gate
# in v0.5.429 — both targets are portable and the ubuntu-24.04
# matrix entry below already verifies them at 1× billing
# weight. Keep ios-simulator here because it requires Apple
# SDK that only macos-14 has.
cmd_xcompile_blocking: "./scripts/run_doc_tests.sh --verbose --xcompile-only --xcompile-only-target=ios-simulator"
cmd_xcompile_advisory: "./scripts/run_doc_tests.sh --verbose --xcompile-only"
# Baseline captured from a clean CI run of 24723671119 (900x970).
gallery_advisory: false
# ubuntu-24.04 / perry-ui-gtk4 doc-tests re-disabled v0.5.873:
# 39 of 88 tests TIMEOUT at the 15s execution budget. The gtk4
# `glib::timeout_add_local_once` → `app.quit()` exit path
# doesn't terminate the main loop cleanly under xvfb-run, so
# PERRY_UI_TEST_MODE-driven self-exit never fires and the
# harness has to SIGKILL each binary. Separate bug from the
# webkit6 / soup / ed25519 fixes that v0.5.864→0.5.871 closed.
# Tracked separately. Re-enable when the gtk4 testkit exit
# path is fixed.
# - os: ubuntu-24.04
# ui_backend: perry-ui-gtk4
# shell: bash
# cmd_exclude_gallery: "xvfb-run -a ./scripts/run_doc_tests.sh --verbose --skip-xcompile --filter-exclude ui/gallery.ts"
# cmd_gallery: "xvfb-run -a ./scripts/run_doc_tests.sh --verbose --skip-xcompile --filter ui/gallery.ts"
# cmd_xcompile_blocking: "./scripts/run_doc_tests.sh --verbose --xcompile-only --xcompile-only-target=web --xcompile-only-target=wasm --xcompile-only-target=ios-simulator"
# cmd_xcompile_advisory: "./scripts/run_doc_tests.sh --verbose --xcompile-only"
# gallery_advisory: false
# windows-2022 doc-tests for perry-ui-windows temporarily disabled:
# 30+ doc-test snippets COMPILE_FAIL on the Windows runner with
# perry exit 1 (mix of platform-banner gaps and #463-unimplemented
# node:* surface). Tracked separately; re-enable once the snippet
# set + perry-ui-windows compile paths are reconciled. macOS
# doc-tests still gate the release.
# - os: windows-2022
# ui_backend: perry-ui-windows
# shell: pwsh
# cmd_exclude_gallery: "./scripts/run_doc_tests.ps1 --verbose --skip-xcompile --filter-exclude ui/gallery.ts"
# cmd_gallery: "./scripts/run_doc_tests.ps1 --verbose --skip-xcompile --filter ui/gallery.ts"
# cmd_xcompile_blocking: "./scripts/run_doc_tests.ps1 --verbose --xcompile-only --xcompile-only-target=web --xcompile-only-target=wasm --xcompile-only-target=ios-simulator"
# cmd_xcompile_advisory: "./scripts/run_doc_tests.ps1 --verbose --xcompile-only"
# gallery_advisory: false
runs-on: ${{ matrix.os }}
defaults:
run:
shell: ${{ matrix.shell }}
steps:
- uses: actions/checkout@v6
# macos-14 ships with ~14 GB free disk after the preinstalled Xcode +
# iOS/tvOS/watchOS simulator runtime images. Several `cargo build
# --release` jobs in this workflow consistently OOM'd at the cache
# restore step (`No space left on device` from the runner's own
# diagnostic writer, before cargo even started). Wiping the simulator
# runtime IMAGES — not the SDKs — reclaims ~15-25 GB without
# affecting cross-compile to `aarch64-apple-ios-sim` (that only needs
# the SDK, which lives inside the active Xcode app).
- name: Free up disk space (macOS)
if: runner.os == 'macOS'
run: |
BEFORE=$(df -h / | tail -1 | awk '{print $4}')
sudo rm -rf /Library/Developer/CoreSimulator/Profiles/Runtimes/*Simulator* || true
sudo rm -rf ~/Library/Developer/CoreSimulator/Caches/* || true
AFTER=$(df -h / | tail -1 | awk '{print $4}')
echo "Disk free: ${BEFORE} -> ${AFTER}"
- name: Install Rust toolchain
uses: dtolnay/rust-toolchain@stable
- name: Set up MSVC environment (Windows)
# Populates LIB / INCLUDE / PATH so link.exe can find user32.lib
# etc. Without this, runner's MSVC install exists on disk but the
# shell session has no clue where, and perry's LNK1181 fatal
# errors looking for Windows SDK libs.
if: matrix.os == 'windows-2022'
uses: ilammy/msvc-dev-cmd@v1
- name: Install GTK4 + GStreamer + Xvfb + PulseAudio headers (Linux)
if: matrix.os == 'ubuntu-24.04'
run: |
sudo apt-get update
# libgstreamer1.0-dev + libgstreamer-plugins-base1.0-dev added in
# v0.5.442 for PR #371 (perry/media streaming playback, #351).
# gstreamer-sys's build script needs `gstreamer-1.0.pc` findable
# via pkg-config; without these two packages doc-tests-gtk4 fails
# at the cargo build step with "Package gstreamer-1.0 was not
# found in the pkg-config search path". gstreamer-base is the
# transitive dep gstreamer-base-sys needs for the playbin element
# that perry/media wraps.
# libwebkitgtk-6.0-dev added for the perry/ui-gtk4 WebView
# feature (Phases 1-5 + v2 follow-ups, #658): the `webkit6` crate
# (0.4 series) and its transitive `javascriptcore6-sys` need
# `webkitgtk-6.0.pc` AND `javascriptcoregtk-6.0.pc` findable via
# pkg-config. Ubuntu 24.04 (noble) ships libwebkitgtk-6.0-dev
# which provides BOTH .pc files (the old libwebkit2gtk-4.1-dev
# name only ships the 4.1 .pc and that's not what webkit6 wants).
sudo apt-get install -y \
libgtk-4-dev libadwaita-1-dev xvfb pkg-config \
libpulse-dev \
libgstreamer1.0-dev libgstreamer-plugins-base1.0-dev \
libwebkitgtk-6.0-dev libshumate-dev
- name: Surface Android NDK location (for cross-compile)
if: matrix.os == 'macos-14' || matrix.os == 'ubuntu-24.04'
run: |
# 1. Discover NDK location on the runner.
NDK=""
if [ -n "$ANDROID_NDK_HOME" ]; then
NDK="$ANDROID_NDK_HOME"
elif [ -d "$ANDROID_HOME/ndk-bundle" ]; then
NDK="$ANDROID_HOME/ndk-bundle"
elif [ -d "$ANDROID_HOME/ndk" ]; then
NDK=$(ls -1d "$ANDROID_HOME/ndk/"*/ 2>/dev/null | sort -V | tail -1)
NDK="${NDK%/}"
fi
if [ -z "$NDK" ]; then
echo "No Android NDK found on runner — android xcompile will skip"
exit 0
fi
echo "ANDROID_NDK_HOME=$NDK" >> "$GITHUB_ENV"
# 2. Point cc-rs + cargo at the NDK's clang wrapper so
# `cargo build --target aarch64-linux-android` picks up the
# right linker/CC/AR instead of the host `cc`.
HOST_TAG=$(uname -s | tr '[:upper:]' '[:lower:]')-x86_64
# macOS NDK uses darwin-x86_64 even on arm64 runners (rosetta).
if [ "$(uname -s)" = "Darwin" ]; then HOST_TAG="darwin-x86_64"; fi
TOOLCHAIN="$NDK/toolchains/llvm/prebuilt/$HOST_TAG/bin"
API=24
CLANG=$(ls "$TOOLCHAIN"/aarch64-linux-android*-clang 2>/dev/null | sort -V | tail -1)
if [ -z "$CLANG" ]; then
echo "Could not locate NDK clang under $TOOLCHAIN — skipping env setup"
exit 0
fi
echo "CC_aarch64_linux_android=$CLANG" >> "$GITHUB_ENV"
echo "AR_aarch64_linux_android=$TOOLCHAIN/llvm-ar" >> "$GITHUB_ENV"
echo "CARGO_TARGET_AARCH64_LINUX_ANDROID_LINKER=$CLANG" >> "$GITHUB_ENV"
echo "Android NDK wired: CLANG=$CLANG"
- name: Install Apple SDK Rust targets (macOS only)
if: matrix.os == 'macos-14'
run: |
rustup target add aarch64-apple-ios-sim
# tvOS-sim is Rust Tier-3 — perry auto-rebuilds with
# `+nightly -Zbuild-std`, which requires the rust-src
# component on the nightly toolchain.
rustup toolchain install nightly --component rust-src --profile minimal || true
- name: Install Android Rust target (macOS + Ubuntu)
if: matrix.os == 'macos-14' || matrix.os == 'ubuntu-24.04'
run: rustup target add aarch64-linux-android
- uses: Swatinem/rust-cache@v2
with:
shared-key: "${{ runner.os }}-perry"
save-if: ${{ github.ref == 'refs/heads/main' }}
- name: Build compiler + UI backend + harness
run: cargo build --release -p perry -p perry-runtime -p perry-stdlib -p ${{ matrix.ui_backend }} -p perry-doc-tests
- name: Pre-build Apple UI libs for cross-compile (macOS only)
if: matrix.os == 'macos-14'
run: |
# iOS-sim is Rust Tier-2 — builds with stable. tvOS-sim is Tier-3
# and needs nightly + -Zbuild-std; perry's auto-optimize handles
# that path itself, so we skip pre-building perry-ui-tvos here.
cargo build --release -p perry-ui-ios --target aarch64-apple-ios-sim
- name: Lint docs/src markdown fences (repo-wide)
if: matrix.os == 'macos-14'
run: cargo run --release --quiet -p perry-doc-tests -- --lint docs/src
- name: Run non-gallery doc-example tests (blocking)
run: ${{ matrix.cmd_exclude_gallery }}
- name: Run gallery screenshot diff
id: gallery
continue-on-error: ${{ matrix.gallery_advisory }}
run: ${{ matrix.cmd_gallery }}
- name: Cross-compile for web + wasm (blocking)
id: xcompile_blocking
run: ${{ matrix.cmd_xcompile_blocking }}
- name: Cross-compile remaining targets (advisory)
# iOS-sim/tvOS-sim/watchos-sim/android still surface real errors
# that the harness logs but that aren't yet tracked issues. Keep
# advisory until each target has a green baseline run.
id: xcompile_advisory
continue-on-error: true
run: ${{ matrix.cmd_xcompile_advisory }}
- name: Upload doc-tests report
if: always()
uses: actions/upload-artifact@v7
with:
name: doc-tests-report-${{ matrix.os }}
path: docs/examples/_reports/latest.json
- name: Upload gallery screenshot + diff artifacts
if: always()
uses: actions/upload-artifact@v7
with:
name: gallery-screenshots-${{ matrix.os }}
path: |
target/perry-doc-tests/gallery_*.png
docs/examples/_baselines/**/gallery.png
# ---------------------------------------------------------------------------
# Binary size tracking (main branch only)
# ---------------------------------------------------------------------------
binary-size:
if: github.ref == 'refs/heads/main'
runs-on: macos-14
steps:
- uses: actions/checkout@v6
- name: Install Rust toolchain
uses: dtolnay/rust-toolchain@stable
- uses: Swatinem/rust-cache@v2
with:
shared-key: "${{ runner.os }}-perry"
save-if: ${{ github.ref == 'refs/heads/main' }}
- name: Build release binaries
run: cargo build --release -p perry -p perry-runtime -p perry-stdlib
- name: Report binary sizes
run: |
echo "## Binary Sizes" > /tmp/sizes.md
echo '```' >> /tmp/sizes.md
ls -lh target/release/perry target/release/libperry_runtime.a target/release/libperry_stdlib.a 2>/dev/null | awk '{print $5, $9}' >> /tmp/sizes.md
echo '```' >> /tmp/sizes.md
cat /tmp/sizes.md
- name: Upload size report
uses: actions/upload-artifact@v7
with:
name: binary-sizes
path: /tmp/sizes.md