Bigfoot / Sonarqube (push) Skipped
Bigfoot / Build & Test Debug with ./ConanProfiles/host/clang (Unity Build: OFF) (push) Successful in 6m12s
Bigfoot / Build & Test Debug with ./ConanProfiles/host/clang (Unity Build: ON) (push) Successful in 5m49s
Bigfoot / Build & Test Debug with ./ConanProfiles/host/clang_asan (Unity Build: OFF) (push) Successful in 6m33s
Bigfoot / Build & Test Debug with ./ConanProfiles/host/clang_asan (Unity Build: ON) (push) Successful in 6m50s
Bigfoot / Build & Test RelWithDebInfo with ./ConanProfiles/host/clang (Unity Build: OFF) (push) Successful in 6m48s
Bigfoot / Build & Test RelWithDebInfo with ./ConanProfiles/host/clang_asan (Unity Build: OFF) (push) Failing after 32s
Bigfoot / Build & Test RelWithDebInfo with ./ConanProfiles/host/clang_asan (Unity Build: ON) (push) Failing after 0s
Bigfoot / Build & Test RelWithDebInfo with ./ConanProfiles/host/clang (Unity Build: ON) (push) Successful in 6m40s
Bigfoot / Build & Test Release with ./ConanProfiles/host/clang (Unity Build: OFF) (push) Canceled after 0s
Bigfoot / Build & Test Release with ./ConanProfiles/host/clang (Unity Build: ON) (push) Successful in 15m18s
Bigfoot / Clang Format Checks (push) Failing after 10s
Bigfoot / Build & Test Release with ./ConanProfiles/host/clang_asan (Unity Build: ON) (push) Successful in 16m5s
Bigfoot / Build & Test Release with ./ConanProfiles/host/clang_asan (Unity Build: OFF) (push) Successful in 16m28s
Introduces a dual-backend SerializationManager: Tool mode keeps reading/writing BigFile over SQL as before, Game mode now reads a Sasquatch file instead - a flattened, mmap'd counterpart to a BigFile produced ahead of time by Sasquatch::WriteFrom or the new SasquatchPacker CLI tool. The backend is picked at compile time via a new BIGFOOT_TOOL_MODE option (mirrors the existing BIGFOOT_OPTIMIZED/BIGFOOT_NOT_OPTIMIZED pattern); SaveAsset only exists in Tool mode, since a running game never writes back to its own asset pack. Also adds SerializationManager::LoadAssets (Game-mode only): loads a batch of UUIDs by decoding them across multiple threads instead of one at a time, layered by HardReference depth so a reference target is always fully registered before anything decodes a reference to it. This required making HardReference/SoftReference resolution genuinely thread-safe (it mutates ContainerHandle ref counts and AssetLibrary's UUID index/unreferenced-set from inside Unpack()) via atomic ref counts and targeted mutexes, validated by dedicated concurrency tests exercising real thread contention. BigFile itself is faster too: prepared statements are now pooled per query text instead of recompiled on every Request (pooled rather than cached-as-one because LoadAsset recurses and a naive cache would corrupt an in-flight statement), and PRAGMA mmap_size is enabled. Benchmarked against real Asset<T> objects (AssetsForBenchmarking) through the full Pack/Unpack/HardReference pipeline: Sasquatch's raw storage layer hits ~31 GiB/s (mmap ceiling), and the new parallel LoadAssets reaches ~18 GiB/s end-to-end versus ~1.3 GiB/s for the existing sequential LoadAsset loop. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>