The coverage workflow was failing on Dependabot PRs because
COVERALLS_REPO_TOKEN is not available to PRs from forks for security
reasons. Add a conditional check to skip the upload step when the
token is missing, while still running coverage generation.
This allows PRs to pass CI while retaining coverage reports on master.
Dependabot rejected the previous configuration due to duplicate
package-ecosystem/directory combinations. Merged separate patch/minor/major
update configs into single configs per directory with grouped update types.
Replace railway redeploy with railway up using a dynamically generated
proxy Dockerfile. The proxy pulls the pre-built image from GHCR using
the exact SHA256 digest, avoiding Railpack source builds while ensuring
deterministic deployments.
Changed deployment command from `railway up` to `railway redeploy` because
the Railway service is configured to use an external Docker image from GHCR,
not source code deployment. The `railway up` command was attempting to build
from source with Railpack, which failed since no source code was provided.
The `railway redeploy` command triggers Railway to pull the latest image
from ghcr.io/xevion/pac-man:latest and redeploy the service.
Replaced monolithic dependency grouping with 13 separate update groups
across Cargo (game/server), npm (frontend), and GitHub Actions ecosystems.
Key improvements:
- Separate patch/minor/major updates to prevent breaking changes from
blocking safe updates
- Group by crate (game vs server) for easier review
- Enable auto-merge for patch updates via labels
- Weekly checks with 5 PR limit to avoid spam
- Ignore bevy_ecs 0.17+ and jsonwebtoken 10+ until manual migration
Closed PR #10 (22 jumbled updates causing build failures).
The deployment was failing because the workflow used incorrect syntax
with `uses: docker://` which caused Docker to misinterpret `up` as an
executable rather than an argument to the Railway CLI.
Split deployment into a separate job using `container:` at the job level,
allowing the Railway CLI to execute properly. This follows Railway's
official documentation for GitHub Actions integration.
- Remove frontend build steps from GitHub Actions workflow
- Add frontend-builder stage using oven/bun:1 in Dockerfile
- Build frontend inside Docker for better consistency and portability
- Copy built frontend from frontend-builder stage to runtime image
- Simplify CI/CD pipeline by consolidating build steps
This commit introduces a comprehensive deployment strategy that unifies the frontend and backend into a single Docker container served by the Rust backend, streamlining the deployment process and improving production architecture.
Key changes:
- Split CI/CD workflows: separated build.yaml (for CI/PR checks) and deploy.yaml (for production deployment)
- Implemented unified Docker deployment where the Axum server serves both API routes (under /api) and frontend static files
- Added GitHub Container Registry integration for Docker image distribution
- Updated Railway configuration to use the new healthcheck path (/api/health)
- Enhanced postgres.ts script with named volumes and constants for better container management
- Added API client utilities (web/lib/api.ts) and environment configuration (web/.env.example) for frontend-backend communication
- Configured Vite proxy for local development while supporting same-origin requests in production
- Updated Dockerfile to include frontend static files and proper environment variable handling
This architecture eliminates the need for separate deployments and CORS configuration, as the frontend and API are served from the same origin.
- Add web frontend build step to compile Vike site
- Change artifact upload path from ./dist/ to ./web/dist/
- Install web dependencies before building frontend
cargo-vcpkg installs to target/vcpkg but rust-sdl2's vcpkg-rs crate
auto-detection doesn't reliably find it on all platforms. Set VCPKG_ROOT
explicitly to ensure consistent library discovery across platforms.
Cache restore-keys allowed old vcpkg builds to be restored when baseline
changed, causing persistent failures. Extract vcpkg version from Cargo.toml
using cargo metadata and use it as primary cache key component.
- Extract vcpkg baseline via cargo metadata (structured parsing)
- Unified vcpkg cache with version-isolated keys
- Removed manual macOS vcpkg setup and A/B test steps
- Cache cargo-vcpkg binary for faster builds
Adds cargo-vcpkg step alongside manual vcpkg for macOS to determine if
the fix is due to using manual vcpkg or simply upgrading to vcpkg
2025.10.17. Both methods will run to compare behavior.
- Updated vcpkg baseline from 2024.11.16 to 2025.10.17
- Restored macOS triplets in Cargo.toml
- Added duplicate cargo-vcpkg step for macOS builds
Replaces cargo-vcpkg with direct vcpkg installation for macOS to fix
libogg build failures. Uses vcpkg 2025.10.17 with proper SDL2 package
installation and VCPKG_ROOT environment variable for rust-sdl2
integration. Windows and Linux continue using cargo-vcpkg.
Update GitHub Actions workflows to use correct Cargo.toml/Cargo.lock paths
in root directory instead of pacman/ subdirectory. Fix Justfile web command
to serve from web/dist/client instead of pacman/dist.
This is just because managing both 1.86 and 1.88 is really annoying, so
it's better to just be unified. There's no real point to using 1.88
besides more clippy warnings, which are already impeding my work right
now. So we're downgrading.
