FIX: generating random sparse matrix with size > max(int32)

[haidark]

This fixes the following bug:

from scipy import sparse
N = 46952
x = sparse.random(N,N, density=100/N/N)

This occurs because the product of the matrix dimensions is larger than the maximum value of np.int32 and so the the randint() call fails. The fix passes the correct type to the dtype parameter. The documentation should be updated to mention that the random_state.randint function must accept a dtype argument.

[ilayn]

Thanks @haidark Can you please also add some tests such that we can detect regressions in the future?

[haidark]

Added test to generate random matrix of size 10000000000, which is greater than np.iinfo(np.int32).max = 2147483647.

[rgommers]

There is a real test failure here

[AtsushiSakai]

We can close this PR because this issue is already solved in main branch.

In [1]: import sys, scipy, numpy; print(scipy.__version__, numpy.__version__, sy
   ...: s.version_info)
1.8.1 1.22.4 sys.version_info(major=3, minor=10, micro=4, releaselevel='final', serial=0)

In [2]: from scipy import sparse

In [3]: N = 100

In [4]: %time x = sparse.random(N, N, density=100/N/N)
CPU times: user 1.03 ms, sys: 1.98 ms, total: 3.02 ms
Wall time: 7.85 ms

In [5]: N = 1000

In [6]: %time x = sparse.random(N, N, density=100/N/N)
CPU times: user 27.5 ms, sys: 2.77 ms, total: 30.3 ms
Wall time: 29.7 ms

In [7]: N = 10000

In [8]: %time x = sparse.random(N, N, density=100/N/N)
CPU times: user 6.68 s, sys: 244 ms, total: 6.92 s
Wall time: 7.2 s

In [9]: N = 46952

In [10]: %time x = sparse.random(N, N, density=100/N/N)
CPU times: user 15min 27s, sys: 54min 10s, total: 1h 9min 37s
Wall time: 1h 38min 31s