Update to the latest tatami interface. (#6)

We now handle differences in the TileDB attribute/dimension type compared to
the tatami interface types. This requires a lot of work to test that the
dimensions are computed correctly for a variety of signed/unsigned index types.

We improved the oracular caching to consolidate slabs into a single request to
the TileDB library. We switch to a variable slab cache for sparse oracular
extraction to improve cache utilitization. We also replace the SoloCore class
as we can achieve the same memory usage by reusing the other Cores with a chunk
length of 1 and a maximum number of cached slabs to 1.

We introduce a dedicated serialize() function to document and consolidate the
locks around the TileDB library. We actually need to implement a lock during
the parallelization tests because it occasionally deadlocks otherwise.

Updated the CI test workflow to use the latest versions, avoid using ctest,
and to pass along the Codecov token to report coverage.

Finally, updated the README.

.github/workflows/run-tests.yaml

@@ -21,12 +21,12 @@ jobs:
           }
 
     steps:
-    - uses: actions/checkout@v3
+    - uses: actions/checkout@v4
 
     - name: Get latest CMake
       uses: lukka/get-cmake@latest
 
-    - uses: actions/cache@v3
+    - uses: actions/cache@v4
       id: cache
       with:
         path: libtiledb
@@ -51,7 +51,15 @@ jobs:
     - name: Run the tests
       run: |
         cd build
-        ctest
+        # Avoid using ctest because it's so slow; it starts a
+        # new process for each individual test, which is crazy.
+        for exe in $(ctest --show-only=json-v1 | jq -r ".tests[] | .command | .[0]"  | sort | uniq)
+        do
+            echo "#### RUNNING ${exe} ####"
+            echo 
+            ${exe} --gtest_brief=1
+            echo
+        done 
 
     - name: Generate code coverage
       if: ${{ matrix.config.cov }}
@@ -61,6 +69,8 @@ jobs:
 
     - name: Upload to Codecov
       if: ${{ matrix.config.cov }}
-      uses: codecov/codecov-action@v3
+      uses: codecov/codecov-action@v4
       with:
         directory: build/tests/CMakeFiles/
+      env:
+        CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}

README.md

@@ -17,25 +17,18 @@ Any matrix stored as a 2-dimensional TileDB array (dense or sparse) can be repre
 ```cpp
 #include "tatami_tiledb/tatami_tiledb.hpp"
 
-tatami_tiledb::TileDbDenseMatrix<double, int> dense_mat("some_dir", "attr_name");
-tatami_tiledb::TileDbSparseMatrix<double, int> sparse_mat("some_dir", "attr_name");
-```
-
-If the dense/sparse nature is not known beforehand, we can use the `make_TileDbMatrix()` function to decide for us instead:
-
-```cpp
-auto some_mat = tatami_tiledb::make_TileDbMatrix("some_dir", "attr_name");
+tatami_tiledb::DenseMatrix<double, int> dense_mat("some_dir", "attr_name");
+tatami_tiledb::SparseMatrix<double, int> sparse_mat("some_dir", "attr_name");
 ```
 
 Advanced users can fiddle with the options, in particular the size of the tile cache.
 For example, we could force the matrix to use no more than 200 MB of memory in the cache:
 
 ```cpp
-tatami_tiledb::TileDbOptions opt;
+tatami_tiledb::DenseMatrixOptions opt;
 opt.maximum_cache_size = 200000000;
 opt.require_minimum_cache = false; // don't allow the cache to automatically expand.
-
-auto some_mat2 = tatami_tiledb::make_TileDbMatrix("some_dir", "attr_name", opt);
+tatami_tiledb::DenseMatrix<double, int> some_mat2("some_dir", "attr_name", opt);
 ```
 
 Check out the [reference documentation](https://tatami-inc.github.io/tatami_tiledb) for more details.
@@ -50,7 +43,7 @@ If you're using CMake, you just need to add something like this to your `CMakeLi
 include(FetchContent)
 
 FetchContent_Declare(
-  tatami
+  tatami_tiledb
   GIT_REPOSITORY https://github.com/tatami-inc/tatami_tiledb
   GIT_TAG master # or any version of interest
 )