nclack · Oct 4, 2026 · Oct 4, 2026 · Oct 4, 2026
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
@@ -97,10 +97,10 @@ jobs:
 
       - name: Test
         shell: micromamba-shell {0}
-        run: >
-          ctest --test-dir build
-          --output-on-failure
-          -E "(s3)"
+        run: |
+          uv sync --script tests/validate_zarr.py
+          uv sync --script tests/validate_ngff_readback.py
+          ctest --test-dir build --output-on-failure -E "(s3)"
 
   windows-tests:
     runs-on: windows-latest
@@ -143,6 +143,8 @@ jobs:
         run: |
           eval "$(micromamba shell hook --shell bash)"
           micromamba activate ci
+          uv sync --script tests/validate_zarr.py
+          uv sync --script tests/validate_ngff_readback.py
           ctest --test-dir build --output-on-failure -E "(s3)"
 
   coverage:

diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -111,7 +111,7 @@ install(
 # integration path. cudart_static / nvcomp are baked into the chucky archive
 # but a bare -l flag cannot locate them without -L into the toolkit.
 set(CHUCKY_VARIANT_DESCRIPTION "CPU only")
-set(CHUCKY_PC_LIBS_PRIVATE "-lm -llz4 -lzstd -laws-c-s3 -ldl -lrt -lpthread")
+set(CHUCKY_PC_LIBS_PRIVATE "-lm -lblosc -llz4 -lzstd -laws-c-s3 -ldl -lrt -lpthread")
 if(CHUCKY_ENABLE_GPU)
     set(CHUCKY_VARIANT_DESCRIPTION "GPU + CPU")
     string(APPEND CHUCKY_PC_LIBS_PRIVATE " -lcudart_static -lnvcomp")

diff --git a/Dockerfile b/Dockerfile
@@ -64,6 +64,10 @@ ENV CMAKE_PREFIX_PATH="/opt/nvcomp:/opt/aws"
 WORKDIR /src
 COPY . .
 
+# Install independent readback readers before CTest runs.
+RUN uv sync --script tests/validate_zarr.py \
+    && uv sync --script tests/validate_ngff_readback.py
+
 RUN cmake --preset docker -G Ninja \
         -DCMAKE_BUILD_TYPE=Release \
     && cmake --build build
diff --git a/Dockerfile.coverage b/Dockerfile.coverage
@@ -50,5 +50,9 @@ ENV CMAKE_PREFIX_PATH="/opt/aws"
 WORKDIR /src
 COPY . .
 
+# Install independent readback readers before CTest runs.
+RUN uv sync --script tests/validate_zarr.py \
+    && uv sync --script tests/validate_ngff_readback.py
+
 RUN cmake --preset coverage \
     && cmake --build build
diff --git a/Dockerfile.cpu b/Dockerfile.cpu
@@ -45,6 +45,10 @@ ENV CMAKE_PREFIX_PATH="/opt/aws"
 WORKDIR /src
 COPY . .
 
+# Install independent readback readers before CTest runs.
+RUN uv sync --script tests/validate_zarr.py \
+    && uv sync --script tests/validate_ngff_readback.py
+
 RUN cmake --preset cpu-only \
         -DCMAKE_BUILD_TYPE=Release \
     && cmake --build build
diff --git a/README.md b/README.md
@@ -52,9 +52,8 @@ up to 32 LOD levels. Internal layout supports up to 64 dimensions.
 
 - **aws-c-s3** — Amazon S3 client library for S3 storage backend
 - **lz4**, **zstd** — compression libraries
-- **c-blosc** (optional) — enables CPU Blosc-LZ4/Blosc-Zstd and the GPU
-  Blosc interoperability tests. GPU Blosc encoding uses nvCOMP and does not
-  require c-blosc.
+- **c-blosc** — CPU Blosc-LZ4/Blosc-Zstd compression and GPU Blosc
+  interoperability tests. GPU Blosc encoding uses nvCOMP.
 - **CMake** (3.18+) + **Ninja** — build system
 
 The default build targets SM 100 (Blackwell). For other GPUs, set

diff --git a/bench/CMakeLists.txt b/bench/CMakeLists.txt
@@ -95,7 +95,7 @@ foreach(src IN LISTS BENCH_STREAM_SOURCES)
 endforeach()
 
 find_package(Python3 COMPONENTS Interpreter QUIET)
-if(Python3_Interpreter_FOUND AND HAVE_BLOSC)
+if(Python3_Interpreter_FOUND)
     add_test(
         NAME test-bench-cli
         COMMAND
@@ -128,17 +128,15 @@ if(Python3_Interpreter_FOUND)
         list(APPEND MEASUREMENT_BACKENDS gpu)
     endif()
     foreach(backend IN LISTS MEASUREMENT_BACKENDS)
-        if(HAVE_BLOSC)
-            add_test(
-                NAME test-bench_image_readback_${backend}
-                COMMAND
-                    ${Python3_EXECUTABLE}
-                    ${CMAKE_CURRENT_SOURCE_DIR}/test_bench_image_readback.py
-                    $<TARGET_FILE:bench_stream_microscopy> ${backend}
-            )
-            set_tests_properties(test-bench_image_readback_${backend}
-                PROPERTIES TIMEOUT 600)
-        endif()
+        add_test(
+            NAME test-bench_image_readback_${backend}
+            COMMAND
+                ${Python3_EXECUTABLE}
+                ${CMAKE_CURRENT_SOURCE_DIR}/test_bench_image_readback.py
+                $<TARGET_FILE:bench_stream_microscopy> ${backend}
+        )
+        set_tests_properties(test-bench_image_readback_${backend}
+            PROPERTIES TIMEOUT 600)
         add_test(
             NAME test-bench_images_${backend}
             COMMAND

diff --git a/cmake/Blosc.cmake b/cmake/Blosc.cmake
@@ -2,8 +2,7 @@
 #
 # Tries CONFIG mode first (covers vcpkg, Conan, system CMake configs),
 # then falls back to the bundled FindBlosc.cmake module.
-# If neither finds blosc, HAVE_BLOSC is OFF and CPU Blosc is unavailable.
-# GPU Blosc uses nvCOMP and remains available in GPU builds.
+# Required for CPU compression and GPU Blosc interoperability tests.
 
 find_package(blosc CONFIG QUIET)
 if(blosc_FOUND)
@@ -30,12 +29,5 @@ if(blosc_FOUND)
 endif()
 
 if(NOT TARGET Blosc::Blosc)
-    find_package(Blosc QUIET)
-endif()
-
-if(TARGET Blosc::Blosc)
-    set(HAVE_BLOSC ON)
-else()
-    set(HAVE_BLOSC OFF)
-    message(STATUS "Blosc not found — CPU blosc codecs disabled")
+    find_package(Blosc REQUIRED MODULE)
 endif()
diff --git a/cmake/chuckyConfig.cmake.in b/cmake/chuckyConfig.cmake.in
@@ -22,9 +22,7 @@ endmacro()
 
 _chucky_find_dep_config_or_module(lz4 Lz4)
 _chucky_find_dep_config_or_module(zstd Zstd)
-if(@HAVE_BLOSC@)
-    _chucky_find_dep_config_or_module(blosc Blosc)
-endif()
+_chucky_find_dep_config_or_module(blosc Blosc)
 
 find_dependency(aws-c-s3 CONFIG)
 

diff --git a/docs/guide.md b/docs/guide.md
@@ -53,7 +53,7 @@ internal conventions, and key concepts.
 | Target | Source | Purpose |
 |--------|--------|---------|
 | `transpose_cpu` | `src/cpu/transpose.cpp` | OpenMP scatter |
-| `compress_cpu` | `src/cpu/compress.c`, `src/cpu/compress_blosc.c` | zstd / lz4 and optional c-blosc compression (CPU) |
+| `compress_cpu` | `src/cpu/compress.c`, `src/cpu/compress_blosc.c` | zstd / lz4 / c-blosc compression (CPU) |
 | `aggregate_cpu` | `src/cpu/aggregate.c` | Shard packing (CPU) |
 | `lod_cpu` | `src/cpu/lod.cpp` | CPU LOD scatter + reduce |
 | `stream_cpu` | `src/cpu/stream.c` | CPU pipeline orchestrator |
@@ -128,7 +128,7 @@ Both backends implement the same pipeline stages behind the same
 | Stage | GPU | CPU |
 |-------|-----|-----|
 | Scatter | CUDA kernel | OpenMP parallel loop |
-| Compress | nvCOMP LZ4/Zstd, raw or Blosc-framed | libzstd / liblz4 / optional c-blosc |
+| Compress | nvCOMP LZ4/Zstd, raw or Blosc-framed | libzstd / liblz4 / c-blosc |
 | Aggregate | CUDA kernel | Sequential packing |
 | LOD | CUDA kernels (templated on dtype) | C++ templates + OpenMP |
 | Orchestration | 4 CUDA streams, double-buffered | Single-threaded pipeline |

diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt
@@ -109,16 +109,15 @@ list(
 
 # Conditional deps via generator expressions — empty branches are dropped at
 # generation time without referencing missing targets.
-set(_with_blosc $<BOOL:${HAVE_BLOSC}>)
 set(_with_gpu $<BOOL:${CHUCKY_ENABLE_GPU}>)
 set(_chucky_public_deps
     ${_math_lib}
     Threads::Threads
     Lz4::Lz4
     Zstd::Zstd
+    Blosc::Blosc
     AWS::aws-c-s3
     $<$<PLATFORM_ID:Windows>:psapi>
-    $<${_with_blosc}:Blosc::Blosc>
     $<${_with_gpu}:CUDA::cuda_driver>
     $<${_with_gpu}:CUDA::cudart_static>
     $<${_with_gpu}:nvcomp::nvcomp>

diff --git a/src/cpu/CMakeLists.txt b/src/cpu/CMakeLists.txt
@@ -2,16 +2,9 @@ add_src_lib(transpose_cpu SOURCES transpose.cpp transpose.h
     LINKS index_ops threadpool
 )
 
-add_src_lib(compress_cpu SOURCES compress.c compress.h compress_blosc.h
-    LINKS codec_types threadpool Lz4::Lz4 Zstd::Zstd chucky_log
+add_src_lib(compress_cpu SOURCES compress.c compress.h compress_blosc.c compress_blosc.h
+    LINKS codec_types threadpool Lz4::Lz4 Zstd::Zstd Blosc::Blosc chucky_log
 )
-if(HAVE_BLOSC)
-    target_sources(compress_cpu PRIVATE compress_blosc.c)
-    target_link_libraries(compress_cpu PUBLIC Blosc::Blosc)
-    target_compile_definitions(compress_cpu PUBLIC HAVE_BLOSC)
-else()
-    target_sources(compress_cpu PRIVATE compress_blosc_stub.c)
-endif()
 
 add_src_lib(aggregate_cpu SOURCES aggregate.c aggregate.h
     LINKS index_ops threadpool chucky_log

diff --git a/src/cpu/compress_blosc_stub.c b/src/cpu/compress_blosc_stub.c
diff --git a/tests/CMakeLists.txt b/tests/CMakeLists.txt
@@ -162,11 +162,7 @@ add_test_exe(test_ngff_flush_cascade test_ngff_flush_cascade.c stream_cpu writer
 add_test_exe(test_multiarray_cpu test_multiarray_cpu.c multiarray_cpu stream_cpu writer test_shard_sink chucky_log)
 
 # GPU multi-array stream test
-add_gpu_test_exe(test_multiarray_gpu test_multiarray_gpu.c LINKS CUDA::cuda_driver CUDA::cudart_static multiarray_gpu stream writer test_shard_sink)
-if(CHUCKY_ENABLE_GPU AND HAVE_BLOSC)
-    target_compile_definitions(test_multiarray_gpu PRIVATE HAVE_BLOSC)
-    target_link_libraries(test_multiarray_gpu PRIVATE Blosc::Blosc)
-endif()
+add_gpu_test_exe(test_multiarray_gpu test_multiarray_gpu.c LINKS CUDA::cuda_driver CUDA::cudart_static multiarray_gpu stream writer test_shard_sink Blosc::Blosc)
 # A cross-array ordering cycle would otherwise strand the GPU test runner.
 if(CHUCKY_ENABLE_GPU)
     set_tests_properties(test-test_multiarray_gpu PROPERTIES TIMEOUT 60)
@@ -178,14 +174,24 @@ add_test_exe(test_multi_append test_multi_append.c stream_cpu writer test_shard_
 # CPU stream + zarr store integration (exercises async IO fencing)
 add_test_exe(test_cpu_zarr_fs test_cpu_zarr_fs.c stream_cpu test_zarr_helpers test_counting_sink writer Zstd::Zstd test_platform test_shard_verify)
 
-# Zarr readback validation via zarr-python (requires uv)
+# Zarr readback validation via zarr-python and TensorStore (requires uv)
 add_test_exe(test_zarr_readback test_zarr_readback.c stream_cpu test_zarr_helpers writer test_platform chucky_log)
 target_compile_definitions(
     test_zarr_readback
     PRIVATE SOURCE_DIR="${CMAKE_SOURCE_DIR}"
 )
 set_tests_properties(test-test_zarr_readback PROPERTIES SKIP_RETURN_CODE 77)
 
+# Full NGFF pyramid readback, sharing the fixture and codec matrix across backends.
+add_test_exe(test_ngff_readback_cpu test_ngff_readback.c
+    stream_cpu test_zarr_helpers writer test_platform chucky_log
+)
+target_compile_definitions(
+    test_ngff_readback_cpu
+    PRIVATE SOURCE_DIR="${CMAKE_SOURCE_DIR}"
+)
+set_tests_properties(test-test_ngff_readback_cpu PROPERTIES TIMEOUT 180)
+
 # Cross-validate GPU vs CPU (byte-exact shard comparison with CODEC_NONE)
 add_gpu_test_exe(test_cross_validate test_cross_validate.c LINKS CUDA::cuda_driver CUDA::cudart_static stream stream_cpu writer test_data test_shard_verify)
 
@@ -201,7 +207,7 @@ target_link_libraries(test_data PUBLIC stream_config writer chucky_log)
 # Tests with zstd
 add_gpu_test_exe(test_stream          test_stream.c          LINKS CUDA::cuda_driver CUDA::cudart_static stream Zstd::Zstd index_ops_util test_shard_sink test_gpu_helpers test_shard_verify test_voxel_encode)
 add_gpu_test_exe(test_compress        test_compress.c        LINKS CUDA::cuda_driver CUDA::cudart_static stream Zstd::Zstd Lz4::Lz4)
-if(CHUCKY_ENABLE_GPU AND HAVE_BLOSC)
+if(CHUCKY_ENABLE_GPU)
     add_gpu_test_exe(test_compress_blosc_gpu test_compress_blosc_gpu.c
         LINKS CUDA::cuda_driver CUDA::cudart_static stream Blosc::Blosc
     )
@@ -217,6 +223,15 @@ if(CHUCKY_ENABLE_GPU AND HAVE_BLOSC)
         test-test_zarr_readback_gpu
         PROPERTIES SKIP_RETURN_CODE 77 TIMEOUT 120
     )
+    add_gpu_test_exe(test_ngff_readback_gpu test_ngff_readback.c
+        LINKS CUDA::cuda_driver CUDA::cudart_static stream test_zarr_helpers
+              writer test_platform chucky_log
+    )
+    target_compile_definitions(
+        test_ngff_readback_gpu
+        PRIVATE TEST_NGFF_READBACK_GPU SOURCE_DIR="${CMAKE_SOURCE_DIR}"
+    )
+    set_tests_properties(test-test_ngff_readback_gpu PROPERTIES TIMEOUT 180)
 endif()
 
 add_test_exe(test_json_writer test_json_writer.c zarr_metadata ngff_metadata attr_set json_validate strbuf chucky_log)

diff --git a/tests/test_compress_cpu.c b/tests/test_compress_cpu.c
@@ -3,9 +3,7 @@
 #include "threadpool/threadpool.h"
 #include "util/prelude.h"
 
-#ifdef HAVE_BLOSC
 #include <blosc.h>
-#endif
 #include <lz4.h>
 #include <stdlib.h>
 #include <string.h>
@@ -259,7 +257,6 @@ test_nthreads_1(void)
   return 1;
 }
 
-#ifdef HAVE_BLOSC
 static int
 test_codec_blosc(enum compression_codec id, const char* name)
 {
@@ -421,7 +418,6 @@ test_blosc_explicit_blocks(void)
   free(decoded);
   return 1;
 }
-#endif // HAVE_BLOSC
 
 int
 main(int ac, char* av[])
@@ -441,11 +437,9 @@ main(int ac, char* av[])
   rc |= test_codec_lz4();
   rc |= test_codec_zstd();
   rc |= test_nthreads_1();
-#ifdef HAVE_BLOSC
   rc |= test_codec_blosc(CODEC_BLOSC_LZ4, "blosc_lz4");
   rc |= test_codec_blosc(CODEC_BLOSC_ZSTD, "blosc_zstd");
   rc |= test_blosc_explicit_blocks();
-#endif
 
   threadpool_free(g_pool);
   threadpool_free(g_pool1);