magicciv/.forgejo/workflows/ci.yml

# Magic Civilization — CI gate on push-to-main.
#
# Implements objective p2-10 (`.project/objectives/p2-10-regression-ci-gate.md`):
# every push to `main` must pass the full regression suite on the self-hosted
# apricot runner before the commit is considered "green" on the Forgejo commit page.
#
# Runner registration instructions: `.forgejo/RUNNER_SETUP.md`.
# Forge: http://10.0.0.11:3000/magicciv/magicciv (Forgejo Actions, drone-compatible).
#
# Budget: median completion <= 15 min. If a stage consistently blows this,
# demote it to a nightly workflow (see RUNNER_SETUP.md § "Budget overruns").

name: ci

on:
  push:
    branches:
      - main
  # Allow a user to kick the gate manually from the Forgejo Actions UI or
  # via `POST /api/v1/repos/magicciv/magicciv/actions/workflows/ci.yml/dispatches`.
  # Useful for a first-run smoke on a freshly-registered runner without
  # needing a sacrificial commit.
  workflow_dispatch:

# One CI run per commit on main; a later push supersedes an in-flight run.
concurrency:
  group: ci-main
  cancel-in-progress: true

env:
  # Keeps cargo/rustup/pip caches inside the repo workspace so the runner
  # can reuse them across runs without polluting $HOME on apricot.
  CARGO_TERM_COLOR: always
  RUST_BACKTRACE: 1

jobs:
  regression:
    name: regression gate
    # Self-hosted apricot runner — see .forgejo/RUNNER_SETUP.md for registration.
    # Labels must match forgejo-runner config on apricot exactly.
    runs-on: [self-hosted, linux, apricot]
    timeout-minutes: 25

    steps:
      - name: Checkout
        uses: actions/checkout@v4
        with:
          fetch-depth: 1

      # ── Stage 1: Rust workspace tests ────────────────────────────────
      # Covers all mc-* crates + api-wasm + api-gdext surface tests.
      # GPU-feature tests are attempted but don't fail the build if the
      # gpu feature fails to compile on this runner (CUDA/Vulkan env may
      # not be present); a clean "skipped" message is preferable to a
      # misleading red pipeline on an environment gap.
      - name: cargo test --workspace
        working-directory: src/simulator
        run: cargo test --workspace --locked

      - name: cargo test --workspace --features gpu (best-effort)
        working-directory: src/simulator
        # `cargo check` gates the attempt — if gpu deps won't even compile
        # on this runner, skip the full test run loudly but don't fail.
        run: |
          set -uo pipefail
          if cargo check --workspace --features gpu 2>/dev/null; then
            echo "gpu feature compiles; running gpu tests"
            cargo test --workspace --features gpu --locked
          else
            echo "::warning::gpu feature failed to compile on this runner — skipping gpu tests"
          fi

      # ── Stage 2: GDScript lint (advisory) ────────────────────────────
      # gdtoolkit must be installed on the runner; see RUNNER_SETUP.md.
      #
      # Hard-gated 2026-04-25 (p2-10a). Backlog cleared via file-splitting:
      # unit_renderer_draw.gd, city_rust_bridge.gd, game_state_serialization_helpers.gd,
      # ai_turn_bridge_dispatch.gd, ai_turn_bridge_state.gd, turn_processor_helpers.gd,
      # turn_processor_city_helpers.gd extracted; stale src/generation/auto_play.gd
      # deleted (live autoload at scenes/tests/auto_play.gd was unaffected);
      # StatsTracker.CATEGORY_LABELS renamed to category_labels per
      # class-variable-name rule. `gdlint src/game/engine/src/` exits 0 on main HEAD.
      - name: gdlint
        run: gdlint src/game/engine/src/

      # ── Stage 3: Game data JSON validator ────────────────────────────
      - name: validate game data
        run: python3 tools/validate-game-data.py

      # ── Stage 4: Objectives dashboard sync check ─────────────────────
      # Fails if the dashboard in .project/objectives/ drifts from the
      # frontmatter in individual objective files.
      - name: objectives report --check
        run: python3 tools/objectives-report.py --check

      # ── Stage 4.5: Build GDExtension ─────────────────────────────────
      # Godot's GDScript compile depends on `GdClimatePhysics`, `GdGridState`,
      # `GdTechWeb` etc. which come from the Rust GDExtension. The .so is
      # gitignored (built per-host), so a fresh `actions/checkout@v4`
      # workspace has no binary — Godot's parser errors before any test runs.
      # We build release-profile `magic-civ-physics-gdext` into
      # `src/game/engine/addons/magic_civ_physics/libmagic_civ_physics.x86_64.so`
      # which is what `magic_civ_physics.gdextension` points at.
      - name: build GDExtension
        working-directory: src/simulator
        run: bash build-gdext.sh

      # ── Stage 4.75: Godot import — build class cache ─────────────────
      # Fresh checkouts have no `.godot/global_script_class_cache.cfg`;
      # without it Godot's parser cannot resolve GDExtension type names
      # (e.g. `GdClimatePhysics`), so even well-formed GDScript fails to
      # compile. `--headless --import` walks the project once to populate
      # the cache.
      - name: Godot import (build class cache)
        run: |
          flatpak run --filesystem=home org.godotengine.Godot \
            --path src/game --headless --import

      # ── Stage 5: Headless GUT ────────────────────────────────────────
      # Runs via Flatpak Godot on apricot. `tools/gut-headless.sh` wraps
      # the Flatpak invocation and parses JUnit XML exit code (Flatpak
      # swallows Godot's quit() exit code).
      #
      # Hard-gated 2026-04-26 (p2-10b). Original backlog: 40 failures out
      # of 439 tests. All 40 triaged: 32 fixed (stale assertions, wrong
      # API names, type mismatches, missing production_cost fixture field,
      # is_flying() keyword check, diplomacy null-partner guard) and 8
      # spun out as pending with objective refs (p2-10c thru p2-10j).
      # `gut-headless.sh` exits 0 only when JUnit XML reports 0 failures.
      - name: headless GUT
        run: bash tools/gut-headless.sh

      # ── Stage 6: 1-seed T100 smoke batch (advisory) ──────────────────
      # Minimum-viable determinism + no-stall check. Confirms the commit
      # doesn't deadlock or crash the autoplay loop. PARALLEL=1 to keep
      # runtime predictable within the 15-minute budget.
      #
      # Currently advisory — autoplay-batch + flatpak sandbox plumbing
      # has remaining rough edges on fresh CI checkouts (meta.json /
      # turn_stats.jsonl not always landing even when the game completes).
      # The `cargo test` stage already covers the deeper simulation
      # determinism; this stage is the end-to-end dual-language smoke.
      # Flip to hard-fail when the last sandbox path bug is fixed.
      - name: autoplay smoke (seed 1, 100 turns) (advisory)
        continue-on-error: true
        env:
          PARALLEL: "1"
        run: |
          set -uo pipefail
          # Absolute path keeps flatpak's sandbox happy across checkout
          # layouts — relative paths break when the runner workdir is not
          # the CWD the sandbox resolves against.
          out_dir="$GITHUB_WORKSPACE/.local/iter/ci_smoke_${GITHUB_SHA:-unknown}"
          mkdir -p "$out_dir"
          bash tools/autoplay-batch.sh 1 100 "$out_dir"

      # Surface the smoke batch's per-turn stats + events on failure so a
      # red pipeline is debuggable without having to re-run locally.
      - name: Upload smoke artifacts
        if: failure()
        uses: actions/upload-artifact@v3
        with:
          name: ci-smoke-${{ github.sha }}
          path: .local/iter/ci_smoke_*/**
          retention-days: 7