It appears that the internal resize methods for SparseMatrix spuriously throw std::bad_alloc when using a SparseMatrix whose StorageIndex is of a wider integer type than Eigen::Index. 

For example, consider the following snippet, in which we construct a SparseMatrix using 64-bit StorageIndex in a context where Eigen::Index is a 32-bit integer: 

```
#define EIGEN_DEFAULT_DENSE_INDEX_TYPE int32_t
#include <Eigen/Core>
#include <Eigen/Sparse>

int main() {
	Eigen::SparseMatrix<double, Eigen::ColMajor, int64_t> mat(5, 5);
	for (int i = 0; i < 5; i++) {
		for (int j = 0; j < 5; j++) {
			mat.coeffRef(i, j) = 1;
		}
	}
	return 0;
}
```

Compiling this with: 

```
g++ main.cpp -o main -Ieigen-3.3.7 -std=c++11
```

(assuming the Eigen sources live in `./eigen-3.3.7`) and running it with `./main` will abort with: 

```
terminate called after throwing an instance of 'std::bad_alloc'
  what():  std::bad_alloc
Aborted (core dumped)
```

The issue appears to stem from the following line in `SparseCore/CompressedStorage.h`:

```
Index realloc_size = (std::min<Index>)(NumTraits<StorageIndex>::highest(),  size + Index(reserveSizeFactor*double(size)));
```

when `StorageIndex` is wider than `Index`, this results in `realloc_size` being set to -1, which triggers the `internal::throw_std_bad_alloc()` on the next line.