Revert to passing matrices to/from C++ via uintptr_t's.

CHANGELOG.md

@@ -1,5 +1,11 @@
 # Changelog
 
+## Version 0.3.1
+
+- Cast to/from `uintptr_t` so that downstream packages aren't forced to rely on **pybind11** converters.
+- Added a `mattress.h` to ensure developers use the correct types during casting.
+- Shift all responsibility for GC protection to C++ via the new `mattress::BoundMatrix` class.
+
 ## Version 0.3.0
 
 - Switch to **pybind11** for the Python/C++ interface, with CMake for the build system.

MANIFEST.in

@@ -0,0 +1 @@
+recursive-include src/mattress/include *

README.md

@@ -32,22 +32,26 @@ pip install mattress
 The aim is to allow package C++ code to accept [all types of matrix representations](#supported-matrices) without requiring re-compilation of the associated code.
 To achive this:
 
-1. Add `assorthead.includes()` to the compiler's include path. 
+1. Add `mattress.includes()` and `assorthead.includes()` to the compiler's include path. 
 This can be done through `include_dirs=` of the `Extension()` definition in `setup.py`
 or by adding a `target_include_directories()` in CMake, depending on the build system.
 2. Call `mattress.initialize()` on a Python matrix object to wrap it in a **tatami**-compatible C++ representation. 
 This returns an `InitializedMatrix` with a `ptr` property that contains a pointer to the C++ matrix.
-3. Pass `ptr` to [**pybind11**](https://pybind11.readthedocs.io)-wrapped C++ code as a [shared pointer to a `tatami::Matrix`](lib/src/def.h),
+3. Pass `ptr` to C++ code as a `uintptr_t` referencing a `tatami::Matrix`,
 which can be interrogated as described in the [**tatami** documentation](https://github.com/tatami-inc/tatami).
 
 So, for example, the C++ code in our downstream package might look like the code below:
 
 ```cpp
-int do_something(const std::shared_ptr<tatami::Matrix<double, uint32_t> >& mat) {
+#include "mattress.h"
+
+int do_something(uintptr_t ptr) {
+    const auto& mat_ptr = mattress::cast(ptr)->ptr;
     // Do something with the tatami interface.
     return 1;
 }
 
+// Assuming we're using pybind11, but any framework that can accept a uintptr_t is fine.
 PYBIND11_MODULE(lib_downstream, m) {
     m.def("do_something", &do_something);
 }
@@ -64,7 +68,7 @@ def do_something(x):
     return lib.do_something(tmat.ptr)
 ```
 
-See [`lib/src/def.h`](lib/src/def.h) for the exact definitions of the interface types used by **mattress**.
+Check out [the included header](src/mattress/include/mattress.h) for more definitions.
 
 ## Supported matrices
 
@@ -165,10 +169,10 @@ init2 = initialize(wrapped)
 This is more efficient as it re-uses the `InitializedMatrix` already generated from `x`.
 It is also more convenient as we don't have to carry around `x` to generate `init2`.
 
-## Extending `initialize()`
+## Extending to custom matrices
 
 Developers can extend **mattress** to custom matrix classes by registering new methods with the `initialize()` generic.
-This should return a `InitializedMatrix` object containing a shared pointer to a `tatami::Matrix<double, uint32_t>` instance (see [`lib/src/def.h`](lib/src/def.h) for types).
+This should return a `InitializedMatrix` object containing a `uintptr_t` cast from a pointer to a `tatami::Matrix` (see [the included header](src/mattress/include/mattress.h)).
 Once this is done, all calls to `initialize()` will be able to handle matrices of the newly registered types.
 
 ```python
@@ -178,9 +182,9 @@ import mattress
 @mattress.initialize.register
 def _initialize_my_custom_matrix(x: MyCustomMatrix):
     data = x.some_internal_data
-    return mattress.InitializedMatrix(lib.initialize_custom(data), objects=[data])
+    return mattress.InitializedMatrix(lib.initialize_custom(data))
 ```
 
 If the initialized `tatami::Matrix` contains references to Python-managed data, e.g., in NumPy arrays,
 we must ensure that the data is not garbage-collected during the lifetime of the `tatami::Matrix`.
-This is achieved by storing a reference to the data in the `objects=` argument of the `InitializedMatrix`.
+This is achieved by storing a reference to the data in the `original` member of the `mattress::BoundMatrix`.

lib/CMakeLists.txt

@@ -24,6 +24,7 @@ pybind11_add_module(mattress
 )
 
 target_include_directories(mattress PRIVATE "${ASSORTHEAD_INCLUDE_DIR}")
+target_include_directories(mattress PRIVATE "../src/mattress/include")
 
 set_property(TARGET mattress PROPERTY CXX_STANDARD 17)
 

lib/src/compressed_sparse_matrix.cpp

@@ -1,4 +1,4 @@
-#include "def.h"
+#include "mattress.h"
 #include "utils.h"
 
 #include "pybind11/pybind11.h"
@@ -9,7 +9,14 @@
 #include <cstdint>
 
 template<typename Data_, typename Index_>
-MatrixPointer initialize_compressed_sparse_matrix_raw(MatrixIndex nr, MatrixValue nc, const pybind11::array& data, const pybind11::array& index, const pybind11::array& indptr, bool byrow) {
+uintptr_t initialize_compressed_sparse_matrix_raw(
+    mattress::MatrixIndex nr,
+    mattress::MatrixValue nc,
+    const pybind11::array& data,
+    const pybind11::array& index,
+    const pybind11::array& indptr,
+    bool byrow)
+{
     size_t expected = (byrow ? nr : nc);
     if (indptr.size() != expected + 1) {
         throw std::runtime_error("unexpected length for the 'indptr' array");
@@ -27,12 +34,28 @@ MatrixPointer initialize_compressed_sparse_matrix_raw(MatrixIndex nr, MatrixValu
     }
     tatami::ArrayView<Index_> iview(check_contiguous_numpy_array<Index_>(index), nz);
 
-    typedef tatami::CompressedSparseMatrix<MatrixValue, MatrixIndex, decltype(dview), decltype(iview), decltype(pview)> Spmat;
-    return MatrixPointer(new Spmat(nr, nc, std::move(dview), std::move(iview), std::move(pview), byrow));
+    auto tmp = std::make_unique<mattress::BoundMatrix>();
+    typedef tatami::CompressedSparseMatrix<mattress::MatrixValue, mattress::MatrixIndex, decltype(dview), decltype(iview), decltype(pview)> Spmat;
+    tmp->ptr.reset(new Spmat(nr, nc, std::move(dview), std::move(iview), std::move(pview), byrow));
+
+    pybind11::tuple objects(3);
+    objects[0] = data;
+    objects[1] = index;
+    objects[2] = indptr;
+    tmp->original = std::move(objects);
+
+    return mattress::cast(tmp.release());
 }
 
 template<typename Data_>
-MatrixPointer initialize_compressed_sparse_matrix_itype(MatrixIndex nr, MatrixValue nc, const pybind11::array& data, const pybind11::array& index, const pybind11::array& indptr, bool byrow) {
+uintptr_t initialize_compressed_sparse_matrix_itype(
+    mattress::MatrixIndex nr,
+    mattress::MatrixValue nc,
+    const pybind11::array& data,
+    const pybind11::array& index,
+    const pybind11::array& indptr,
+    bool byrow)
+{
     auto dtype = index.dtype();
 
     if (dtype.is(pybind11::dtype::of<int64_t>())) {
@@ -54,10 +77,17 @@ MatrixPointer initialize_compressed_sparse_matrix_itype(MatrixIndex nr, MatrixVa
     }
 
     throw std::runtime_error("unrecognized index type '" + std::string(dtype.kind(), 1) + std::to_string(dtype.itemsize()) + "' for compressed sparse matrix initialization");
-    return MatrixPointer();
+    return 0;
 }
 
-MatrixPointer initialize_compressed_sparse_matrix(MatrixIndex nr, MatrixValue nc, const pybind11::array& data, const pybind11::array& index, const pybind11::array& indptr, bool byrow) {
+uintptr_t initialize_compressed_sparse_matrix(
+    mattress::MatrixIndex nr,
+    mattress::MatrixValue nc,
+    const pybind11::array& data,
+    const pybind11::array& index,
+    const pybind11::array& indptr,
+    bool byrow)
+{
     auto dtype = data.dtype();
 
     if (dtype.is(pybind11::dtype::of<double>())) {
@@ -83,7 +113,7 @@ MatrixPointer initialize_compressed_sparse_matrix(MatrixIndex nr, MatrixValue nc
     }
 
     throw std::runtime_error("unrecognized data type '" + std::string(dtype.kind(), 1) + std::to_string(dtype.itemsize()) + "' for compressed sparse matrix initialization");
-    return MatrixPointer();
+    return 0;
 }
 
 void init_compressed_sparse_matrix(pybind11::module& m) {

lib/src/delayed_binary_isometric_operation.cpp

@@ -1,4 +1,4 @@
-#include "def.h"
+#include "mattress.h"
 
 #include "pybind11/pybind11.h"
 #include "pybind11/numpy.h"
@@ -8,45 +8,60 @@
 #include <string>
 #include <cstdint>
 
-MatrixPointer initialize_delayed_binary_isometric_operation(MatrixPointer left, MatrixPointer right, const std::string& op) {
+template<typename Operation_>
+uintptr_t convert(uintptr_t left, uintptr_t right, Operation_ op) {
+    auto lbound = mattress::cast(left);
+    auto rbound = mattress::cast(right);
+
+    auto tmp = std::make_unique<mattress::BoundMatrix>();
+    tmp->ptr = tatami::make_DelayedBinaryIsometricOperation(lbound->ptr, rbound->ptr, std::move(op));
+
+    pybind11::tuple original(2);
+    original[0] = lbound->original;
+    original[1] = rbound->original;
+    tmp->original = std::move(original);
+    return mattress::cast(tmp.release());
+}
+
+uintptr_t initialize_delayed_binary_isometric_operation(uintptr_t left, uintptr_t right, const std::string& op) {
     if (op == "add") {
-        return (tatami::make_DelayedBinaryIsometricOperation(std::move(left), std::move(right), tatami::make_DelayedBinaryIsometricAdd()));
+        return convert(left, right, tatami::make_DelayedBinaryIsometricAdd());
     } else if (op == "subtract") {
-        return (tatami::make_DelayedBinaryIsometricOperation(std::move(left), std::move(right), tatami::make_DelayedBinaryIsometricSubtract()));
+        return convert(left, right, tatami::make_DelayedBinaryIsometricSubtract());
     } else if (op == "multiply") {
-        return (tatami::make_DelayedBinaryIsometricOperation(std::move(left), std::move(right), tatami::make_DelayedBinaryIsometricMultiply()));
+        return convert(left, right, tatami::make_DelayedBinaryIsometricMultiply());
     } else if (op == "divide") {
-        return (tatami::make_DelayedBinaryIsometricOperation(std::move(left), std::move(right), tatami::make_DelayedBinaryIsometricDivide()));
+        return convert(left, right, tatami::make_DelayedBinaryIsometricDivide());
     } else if (op == "remainder") {
-        return (tatami::make_DelayedBinaryIsometricOperation(std::move(left), std::move(right), tatami::make_DelayedBinaryIsometricModulo()));
+        return convert(left, right, tatami::make_DelayedBinaryIsometricModulo());
     } else if (op == "floor_divide") {
-        return (tatami::make_DelayedBinaryIsometricOperation(std::move(left), std::move(right), tatami::make_DelayedBinaryIsometricIntegerDivide()));
+        return convert(left, right, tatami::make_DelayedBinaryIsometricIntegerDivide());
     } else if (op == "power") {
-        return (tatami::make_DelayedBinaryIsometricOperation(std::move(left), std::move(right), tatami::make_DelayedBinaryIsometricPower()));
+        return convert(left, right, tatami::make_DelayedBinaryIsometricPower());
 
     } else if (op == "equal") {
-        return (tatami::make_DelayedBinaryIsometricOperation(std::move(left), std::move(right), tatami::make_DelayedBinaryIsometricEqual()));
+        return convert(left, right, tatami::make_DelayedBinaryIsometricEqual());
     } else if (op == "not_equal") {
-        return (tatami::make_DelayedBinaryIsometricOperation(std::move(left), std::move(right), tatami::make_DelayedBinaryIsometricNotEqual()));
+        return convert(left, right, tatami::make_DelayedBinaryIsometricNotEqual());
     } else if (op == "greater") {
-        return (tatami::make_DelayedBinaryIsometricOperation(std::move(left), std::move(right), tatami::make_DelayedBinaryIsometricGreaterThan()));
+        return convert(left, right, tatami::make_DelayedBinaryIsometricGreaterThan());
     } else if (op == "greater_equal") {
-        return (tatami::make_DelayedBinaryIsometricOperation(std::move(left), std::move(right), tatami::make_DelayedBinaryIsometricGreaterThanOrEqual()));
+        return convert(left, right, tatami::make_DelayedBinaryIsometricGreaterThanOrEqual());
     } else if (op == "less") {
-        return (tatami::make_DelayedBinaryIsometricOperation(std::move(left), std::move(right), tatami::make_DelayedBinaryIsometricLessThan()));
+        return convert(left, right, tatami::make_DelayedBinaryIsometricLessThan());
     } else if (op == "less_equal") {
-        return (tatami::make_DelayedBinaryIsometricOperation(std::move(left), std::move(right), tatami::make_DelayedBinaryIsometricLessThanOrEqual()));
+        return convert(left, right, tatami::make_DelayedBinaryIsometricLessThanOrEqual());
 
     } else if (op == "logical_and") {
-        return (tatami::make_DelayedBinaryIsometricOperation(std::move(left), std::move(right), tatami::make_DelayedBinaryIsometricBooleanAnd()));
+        return convert(left, right, tatami::make_DelayedBinaryIsometricBooleanAnd());
     } else if (op == "logical_or") {
-        return (tatami::make_DelayedBinaryIsometricOperation(std::move(left), std::move(right), tatami::make_DelayedBinaryIsometricBooleanOr()));
+        return convert(left, right, tatami::make_DelayedBinaryIsometricBooleanOr());
     } else if (op == "logical_xor") {
-        return (tatami::make_DelayedBinaryIsometricOperation(std::move(left), std::move(right), tatami::make_DelayedBinaryIsometricBooleanXor()));
+        return convert(left, right, tatami::make_DelayedBinaryIsometricBooleanXor());
     }
 
     throw std::runtime_error("unknown binary isometric operation '" + op + "'");
-    return MatrixPointer();
+    return 0;
 }
 
 void init_delayed_binary_isometric_operation(pybind11::module& m) {

lib/src/delayed_bind.cpp

@@ -1,18 +1,26 @@
-#include "def.h"
+#include "mattress.h"
 
 #include "tatami/tatami.hpp"
 
 #include "pybind11/pybind11.h"
 
 #include <vector>
 
-MatrixPointer initialize_delayed_bind(pybind11::list inputs, int along) {
-    std::vector<MatrixPointer> combined;
+uintptr_t initialize_delayed_bind(const pybind11::list& inputs, int along) {
+    std::vector<std::shared_ptr<tatami::Matrix<mattress::MatrixValue, mattress::MatrixIndex> > > combined;
     combined.reserve(inputs.size());
+    pybind11::tuple originals(inputs.size());
+
     for (size_t i = 0, n = inputs.size(); i < n; ++i) {
-        combined.push_back(inputs[i].cast<MatrixPointer>());
+        auto bound = mattress::cast(inputs[i].cast<uintptr_t>());
+        combined.push_back(bound->ptr);
+        originals[i] = bound->original; 
     }
-    return tatami::make_DelayedBind(std::move(combined), along == 0);
+
+    auto tmp = std::make_unique<mattress::BoundMatrix>();
+    tmp->ptr = tatami::make_DelayedBind(std::move(combined), along == 0);
+    tmp->original = std::move(originals);
+    return mattress::cast(tmp.release());
 }
 
 void init_delayed_bind(pybind11::module& m) {

lib/src/delayed_subset.cpp

@@ -1,4 +1,4 @@
-#include "def.h"
+#include "mattress.h"
 #include "utils.h"
 
 #include "pybind11/pybind11.h"
@@ -7,9 +7,19 @@
 #include <string>
 #include <cstdint>
 
-MatrixPointer initialize_delayed_subset(MatrixPointer mat, const pybind11::array& subset, bool byrow) {
-    auto sptr = check_numpy_array<MatrixIndex>(subset);
-    return tatami::make_DelayedSubset(std::move(mat), tatami::ArrayView<MatrixIndex>(sptr, subset.size()), byrow);
+uintptr_t initialize_delayed_subset(uintptr_t ptr, const pybind11::array& subset, bool byrow) {
+    auto bound = mattress::cast(ptr);
+    auto sptr = check_numpy_array<mattress::MatrixIndex>(subset);
+
+    auto tmp = std::make_unique<mattress::BoundMatrix>();
+    tmp->ptr = tatami::make_DelayedSubset(bound->ptr, tatami::ArrayView<mattress::MatrixIndex>(sptr, subset.size()), byrow);
+
+    pybind11::tuple original(2);
+    original[0] = bound->original;
+    original[1] = subset;
+    tmp->original = std::move(original);
+
+    return mattress::cast(tmp.release());
 }
 
 void init_delayed_subset(pybind11::module& m) {

lib/src/delayed_transpose.cpp

@@ -1,13 +1,17 @@
-#include "def.h"
+#include "mattress.h"
 
 #include "pybind11/pybind11.h"
 #include "pybind11/numpy.h"
 
 #include <string>
 #include <cstdint>
 
-MatrixPointer initialize_delayed_transpose(MatrixPointer mat) {
-    return tatami::make_DelayedTranspose(std::move(mat));
+uintptr_t initialize_delayed_transpose(uintptr_t ptr) {
+    auto bound = mattress::cast(ptr);
+    auto tmp = std::make_unique<mattress::BoundMatrix>();
+    tmp->ptr = tatami::make_DelayedTranspose(bound->ptr);
+    tmp->original = bound->original;
+    return mattress::cast(tmp.release());
 }
 
 void init_delayed_transpose(pybind11::module& m) {