I have a large epochs file, that gets split into two files when I call epochs.save()
When I load it again using mne.read_epochs(), the events corresponding to the second file are incorrect, offset by [last epoch from the first file + tmax + 10s].

The bug comes from _concatenate_epochs.

This breaks the relationship between the raw file and the events list, which (aside from causing downstream problems with my code) can lead to an error in epochs.get_data if the data were loaded with preload=False.