[antonsoo]

Extracting a trajectory from a partially filled preallocated file includes unwritten frames. Time-range searches also use the padded timestamps, which need not be sorted. For example:

import tempfile
from pathlib import Path
from pde import FileStorage, ScalarField, UnitGrid

with tempfile.TemporaryDirectory() as directory:
    storage = FileStorage(Path(directory) / "run.hdf5", max_length=8)
    field = ScalarField(UnitGrid([4]))
    storage.start_writing(field)
    for time in (1, 2, 3):
        field.data[:] = time
        storage.append(field, time)
    storage.end_writing()
    selected = storage.extract_time_range((1, 2))
    assert list(selected.times) == [1, 2]
    storage.close()

Limit time searches and component extraction to len(storage). The HDF datasets keep their existing lazy access and allocated capacity.

The regression covers partial, full and extensible files, compressed and uncompressed storage, time bounds, and scalar/vector component copies. Native stopped diffusion and wave trajectories also agree with independently copied callback snapshots; existing correct results and solver states stay unchanged.

File-storage tests, the storage/tracker/diffusion/wave checks, and project hooks pass. Whole-project Mypy reports the same diagnostics as unchanged upstream.

Rebased after #922 merged, retaining its complete storage-finalization regressions. The updated branch passes 166 storage, tracker, diffusion and wave tests (26 optional skips) and all configured hooks. All 576 extraction records and all 100 close/resume records remain exactly equal to the previously validated composition, including the independent callback/Fourier references and solver controls. Mypy still reports the same 169 baseline diagnostics.

[david-zwicker]

LGTM