FMS2io collective netcdf reads hang when the io layout is not the same as the layout

[uramirez8707]

Describe the bug
FMS2io collective netcdf reads hang when the io layout is not the same as the layout.
This is because in the domain_read, the code calls netcdf_read_data inside a if (fileobj%is_root) block.

FMS/fms2_io/include/domain_read.inc

Lines 163 to 206 in 23df6d7

	if (fileobj%is_root) then
	if (fileobj%adjust_indices) then
	!< If the file is distributed, the file only contains the io global domain
	c(xdim_index) = 1
	c(ydim_index) = 1
	else
	!< If the file is not distributed read, the file contains the global domain,
	!! so you only need to read the global io domain
	c(xdim_index) = xgbegin
	c(ydim_index) = ygbegin
	endif
	e(xdim_index) = xgsize
	e(ydim_index) = ygsize
	!Read in the global io domain
	select type(vdata)
	type is (integer(kind=i4_kind))
	call allocate_array(buf_i4_kind, e)
	call netcdf_read_data(fileobj, variable_name, buf_i4_kind, &
	unlim_dim_level=unlim_dim_level, &
	corner=c, edge_lengths=e, broadcast=.false.)
	type is (integer(kind=i8_kind))
	call allocate_array(buf_i8_kind, e)
	call netcdf_read_data(fileobj, variable_name, buf_i8_kind, &
	unlim_dim_level=unlim_dim_level, &
	corner=c, edge_lengths=e, broadcast=.false.)
	type is (real(kind=r4_kind))
	call allocate_array(buf_r4_kind, e)
	call netcdf_read_data(fileobj, variable_name, buf_r4_kind, &
	unlim_dim_level=unlim_dim_level, &
	corner=c, edge_lengths=e, broadcast=.false.)
	type is (real(kind=r8_kind))
	call allocate_array(buf_r8_kind, e)
	call netcdf_read_data(fileobj, variable_name, buf_r8_kind, &
	unlim_dim_level=unlim_dim_level, &
	corner=c, edge_lengths=e, broadcast=.false.)
	class default
	call error("unsupported variable type: domain_read_2d: file: "//trim(fileobj%path)//" variable:"// &
	& trim(variable_name))
	end select
	endif

And in netcdf_read_data, it does

FMS/fms2_io/include/netcdf_read_data.inc

Lines 357 to 376 in 23df6d7

	if(fileobj%use_collective) then
	varid = get_variable_id(fileobj%ncid, trim(variable_name), msg=append_error_msg)
	! NetCDF does not have the ability to specify collective I/O at
	! the file basis so we must activate at the variable level
	err = nf90_var_par_access(fileobj%ncid, varid, nf90_collective)
	call check_netcdf_code(err, append_error_msg)
	select type(buf)
	type is (integer(kind=i4_kind))
	err = nf90_get_var(fileobj%ncid, varid, buf, start=c, count=e)
	type is (integer(kind=i8_kind))
	err = nf90_get_var(fileobj%ncid, varid, buf, start=c, count=e)
	type is (real(kind=r4_kind))
	err = nf90_get_var(fileobj%ncid, varid, buf, start=c, count=e)
	type is (real(kind=r8_kind))
	err = nf90_get_var(fileobj%ncid, varid, buf, start=c, count=e)
	class default
	call error("Unsupported variable type: "//trim(append_error_msg))
	end select
	call check_netcdf_code(err, append_error_msg)
	call unpack_data_2d(fileobj, varid, variable_name, buf)

but since the nf90_get_var call is only done by the root pe, the code hangs.

In the case when the io layout is the same as the layout, each rank is reading its own section of the code so all the ranks make it to the nf90_get_var call.

To Reproduce
This small test reproduces the issue. The program works when the io layout is the same as the layout, but it hangs when it is not.
https://github.com/uramirez8707/FMS/blob/pnetcdf_test/test_fms/fms2_io/test_domain_pnetcdf.F90

Expected behavior
This test case should work

System Environment
This happens in any system.

Additional context
N/A