Understanding how Spikeinterface handles recording times.

[JoeZiminski]

I always thought that SI was essentially time-agnostic and computed times on the fly based on sample number and sampling rate. However I had the same experience as @yger when reading this thread and was surprised time information is tracked.

I just seen this docs page which introduces the concept which is great, I had not seen it before. However it might still be useful to flesh it out a bit and add what happens in certain scenarios e.g. 1) slicing recording 2) concatenate recording 3) appending / concatenating segments.

I'd be happy to do this ASAP (just (hopefully 🤞 ) finishing off an unrelated project and will have more time for ephys). But I thought I'd get the ball rolling here in case there are other cases I had not considered.

[h-mayorquin]

This is great.
This would be useful for us in neuroconv.

If you have time to improve the docs on that I will be happy to review and discuss.

I remember that @samuelgarcia changed his views about time recently after a discussion with @cwindolf .

We should sort this out as well:
#2515

[samuelgarcia]

I changed my view about time handling in the motion.
In short we are about to create a motion object that will handle multi segment and also absolut time.

For recording, we are time agnostic by default but you can have a t_start and also a time vector.

We wanted to move this into base so that Sorting also also inherits this 3 times mode (t_start=None, t_start=something or times vector) But this is not started yet.

Time handliong is a bit hidden here : https://github.com/SpikeInterface/spikeinterface/blob/main/src/spikeinterface/core/baserecording.py#L689

[alejoe91]

Thanks @JoeZiminski ! this sounds great :)

Note that for some extractors, e.g. read_openephys you can load the syncrhonized timestamps directly (load_sync_timestamps param). It would be important to mention!