Overflow error when writing anonymized FIF with BIDS guideline

[raphbrd]

Description of the problem

Here is an issue I encountered while anonymizing preprocessed data. BIDS specification recommends choosing a measurement date before 1925 when anonymizing data. However, the FIF format stores only 32-bit timestamps. It means that one cannot go back more than ~68 years.
It is not an issue for info['meas_date'] (apparently not written as a timestamp?), but it is for all timestamps in info['proc_history'].

Possible solution

We could set the timestamps for all block_id in the info['proc_history'] to DATE_NONE by default. But I am not sure if one ever needs this value. I did not find any place in the MNE codebase where the secs and usecs fields of block_id are used.

For the Maxwell filter, mne.preprocessing.maxwell._update_sss_info already sets this field to DATE_NONE. So there is no point in shifting the proc_history again with an identical number of days as the file_id and the meas_date (DATE_NOTE is set to 1970-1-1).

The current function _check_dates checks for overflow, but only on the meas_id and file_id, not on proc_history (which is incorrect most of the time).

So, maybe mne._fiff.meas_info.anonymize_info should set all proc_history timestamps to DATE_NONE ? This would avoid any overflow when writing the FIF file. What do you think?

Steps to reproduce

import os
import mne

sample_data_folder = mne.datasets.sample.data_path(download=False)

fine_cal_file = os.path.join(sample_data_folder, "SSS", "sss_cal_mgh.dat")
crosstalk_file = os.path.join(sample_data_folder, "SSS", "ct_sparse_mgh.fif")

sample_data_raw_file = os.path.join(
    sample_data_folder, "MEG", "sample", "sample_audvis_raw.fif"
)
raw = mne.io.read_raw_fif(sample_data_raw_file, verbose=False)
raw.crop(tmax=60)

# add stuff to processing history
raw_sss = mne.preprocessing.maxwell_filter(
    raw, cross_talk=crosstalk_file, calibration=fine_cal_file, verbose=True
)
# copying the info object to ensure it is writable 
# error occurs for daysback > 24820 or so
raw_sss.info = mne.io.anonymize_info(raw_sss.info.copy(), daysback=36000)
raw_sss.save("tmp_raw.fif", overwrite=True)

Link to data

No response

Expected results

Readable .fif file.

Actual results

Traceback (most recent call last):
  File "/Users/raphaelbordas/neurospin_local/2026_MNE_sprint/mne-python/mne/io/base.py", line 3022, in write
    _write_raw_metadata(
  File "/Users/raphaelbordas/neurospin_local/2026_MNE_sprint/mne-python/mne/io/base.py", line 3215, in _write_raw_metadata
    write_meas_info(fid, info, data_type=data_type, reset_range=reset_range)
  File "/Users/raphaelbordas/neurospin_local/2026_MNE_sprint/mne-python/mne/_fiff/meas_info.py", line 3252, in write_meas_info
    _write_proc_history(fid, info)
  File "/Users/raphaelbordas/neurospin_local/2026_MNE_sprint/mne-python/mne/_fiff/proc_history.py", line 142, in _write_proc_history
    writer(fid, id_, record[key])
  File "/Users/raphaelbordas/neurospin_local/2026_MNE_sprint/mne-python/mne/_fiff/write.py", line 260, in write_id
    arr = np.array(
          ^^^^^^^^^
OverflowError: Python integer -3110397853 out of bounds for int32

Additional information

MNE: 1.12.1 (latest release)

Could be related to #8128 (in that case, the file_id timestamp does not match the meas_date). For some reason, the overflow error does not occur for the timestamp fields of file_id.