[Bug]: Export with link_data=False does not copy data

[rly]

What happened?

See description in NeurodataWithoutBorders/pynwb#1683

Steps to Reproduce

import h5py
import numpy as np

with h5py.File("pynwb_1683_data.h5", "w") as f:
    f.create_dataset(name="data", data=np.random.rand(5, 4))

from pynwb import NWBFile, NWBHDF5IO, validate, TimeSeries
import datetime


f = h5py.File("pynwb_1683_data.h5", "r")

nwbfile = NWBFile(
    session_description="session_description",
    identifier="identifier",
    session_start_time=datetime.datetime.now(datetime.timezone.utc),
)

ts = TimeSeries(
    name="test_ts",
    data=f["data"],
    unit="unit",
    rate=1.,
)
nwbfile.add_acquisition(ts)

filename = "pynwb_1683_test.nwb"

with NWBHDF5IO(filename, "w") as io:
    io.write(nwbfile)

with NWBHDF5IO(filename, "r") as io:
    read_nwbfile = io.read()
    print(read_nwbfile.acquisition["test_ts"].data.file)

export_filename = "pynwb_1683_export.nwb"

with NWBHDF5IO(filename, mode="r") as read_io:
    read_nwbfile = read_io.read()
    # read_nwbfile.generate_new_id()
    # nwbfile contains a TimeSeries where the TimeSeries data array is a link to an external dataset
    # in a different HDF5 file than self.read_path

    with NWBHDF5IO(export_filename, mode="w") as export_io:
        export_io.export(src_io=read_io, nwbfile=read_nwbfile, write_args={'link_data': False})

with NWBHDF5IO(export_filename, "r") as io:
    read_nwbfile = io.read()
    print(read_nwbfile.acquisition["test_ts"].data.file)

Traceback

n/a

Operating System

macOS

Python Executable

Conda

Python Version

3.10

Package Versions

dev branch of both PyNWB and HDMF

Code of Conduct

• I agree to follow this project's Code of Conduct
• Have you checked the Contributing document?
• Have you ensured this bug was not already reported?

[rly]

Note that in this example, NWBFile is unmodified between read and export, so the read builders are written to disk unchanged. Generating new object IDs will mark the NWBFile as modified, and so the "link_data": True will be applicable. This is a different use case than adding an external link between read and export.