What WasmEdge/WasmEdge shipped
Generated by FoxPlug from public activity — not affiliated with Wasmedge.
Week of September 14, 2026
What shipped
- WasmEdge 0.17.2-rc.3 released with llama.cpp upgraded to v0.4.1 in the WASI-NN GGML backend and system blake3 support in CMake. Release
- WASI-NN GGML backend upgraded to llama.cpp v0.4.1 with hardened metadata handling to validate GGML format integrity. Pull request #5356
- WASI-NN RPC implementation fixed to use dynamic buffer sizing instead of hardcoded 64 KiB for get_output operations. Pull request #5031
- Uninstaller crash fixed when shell profile files contain multiple WasmEdge environment source lines. Pull request #5345
- WASI getaddrinfo call now validates output buffer bounds and initializes Winsock before Windows address resolution. Pull request #5344
- Executor fixed to type null locals of abstract reference types correctly with bottom heap types. Pull request #5347
- LLVM backend optimized to lower v128.store using masked stores for better performance. Pull request #5329
- Memory leak in APICoreTest.ModuleDeletion fixed by properly freeing LoaderContext and ValidatorContext. Pull request #5359
- macOS CI jobs fixed to gate test execution on the matrix tests flag to handle CMake 4.4 behavior. Pull request #5353
- Contributing guidelines updated to use Assisted-by trailer for AI disclosure instead of Co-Authored-By. Pull request #5354
Why it matters
0.17.2-rc.3 brings memory safety improvements with buffer bound checking in WASI operations, upgraded GGML support for llama.cpp v0.4.1, and fixes for installer robustness and reference type handling. Several critical memory leaks and output truncation issues are resolved, making this release more reliable for production use.
Changelog entry
- 0.17.2-rc.3: llama.cpp upgraded to v0.4.1, system blake3 support in CMake, loader native shared-library fixes Release
- WASI-NN GGML backend upgraded to llama.cpp v0.4.1 with GGML metadata validation Pull request #5356
- WASI-NN RPC get_output and get_output_single now use dynamic sizing instead of 64 KiB hardcoded limit Pull request #5031
- Fixed uninstall.sh sed crash when multiple WasmEdge environment source lines exist in shell profiles Pull request #5345
- WASI getaddrinfo validates output buffer bounds and initializes Winsock on Windows Pull request #5344
- Executor corrected to type null locals of abstract reference types with bottom heap types Pull request #5347
- LLVM backend optimized v128.store using masked stores Pull request #5329
- Fixed memory leak in APIUnitTest.cpp APICoreTest.ModuleDeletion test Pull request #5359
WasmEdge 0.17.2-rc.3 is out. WASI-NN GGML upgraded to llama.cpp v0.4.1, dynamic buffer sizing for RPC outputs, WASI getaddrinfo bounds checking, and various stability fixes.
WasmEdge 0.17.2-rc.3 is available. This release includes upgraded WASI-NN GGML backend with llama.cpp v0.4.1, fixed output buffer handling in RPC to support larger data, validated getaddrinfo output bounds in WASI, corrected abstract reference type handling in the executor, and resolved memory leaks. These improvements enhance safety and compatibility for production deployments.