2.0.0rc2: Regression in MPI_File_get_size() after two call to MPI_File_preallocate()

[dalcinl]

from mpi4py import MPI
comm = MPI.COMM_SELF
mode = MPI.MODE_RDWR | MPI.MODE_CREATE
info = MPI.INFO_NULL
fh = MPI.File.Open(comm, "/tmp/datafile", mode, info)
fh.Set_size(0)
assert fh.Get_size() == 0
fh.Preallocate(1)
assert fh.Get_size() == 1
fh.Preallocate(100)
assert fh.Get_size() == 100

$ python test-ompi-4.py 
Traceback (most recent call last):
  File "test-ompi-4.py", line 11, in <module>
    assert fh.Get_size() == 100
AssertionError

[edgargabriel]

will provide a fix for this, I think it is fairly straight forward.

[edgargabriel]

just as an FYI, the solution is already available in my private repo, just waiting to finish the get_view issue to file the joint pull request.

[jsquyres]

Sweet. I added the tests so that we can make sure these don't regress again.

[edgargabriel]

Cool. The get_byte_offset will take me a bit longer however.

[dalcinl]

@edgargabriel @jsquyres There are still issues with MPI_File_preallocate(). The following extended example calls preallocate with a size less than the current file size. According to the MPI standard (page 498, lines 23-25), the file size should not change

from mpi4py import MPI
comm = MPI.COMM_SELF
mode = MPI.MODE_RDWR | MPI.MODE_CREATE
info = MPI.INFO_NULL
fh = MPI.File.Open(comm, "/tmp/datafile", mode, info)
fh.Preallocate(0)
assert fh.Get_size() == 0
fh.Preallocate(1)
assert fh.Get_size() == 1
fh.Preallocate(100)
assert fh.Get_size() == 100
fh.Preallocate(10)
assert fh.Get_size() == 100
fh.Preallocate(200)
assert fh.Get_size() == 200

$ mpiexec -n 1 python test-ompi-issue-1633.py 
Traceback (most recent call last):
  File "test-ompi-issue-1633.py", line 13, in <module>
    assert fh.Get_size() == 100
AssertionError

[edgargabriel]

I'll have a look

[jsquyres]

@edgargabriel How's it going on this one -- has it been resolved?