Writer wishlist

[tstenner]

I've started a rough prototype for writer support. Currently, it looks like this:

with pyxdf.writer.Writer('/tmp/foo.xdf.gz') as w:
    w.add_stream(streamid=1, header='<info><streamname>XYZ</streamname>…</info>')
    w.add_stream_data(streamid=1, data=np.array([[1,2],[2,3],[3,4]], ts=np.array([5.1, np.nan, 5.3]))
    w.add_stream_offsets(streamid=1, offsets=np.array([[1.0, 5.0], [1.3, 5.3]]))

Goals:

• Acceptable performance in pure Python
• Optional cython-compiled parts
• Full support for the XDF 1.0 spec

Non-goals:

• accept the reader output directly as input
• type conversion (i.e. convert data to the stream's data type; write str objects to a string stream)
• write everything in C and call a compiled library

[Argzero]

I have a rudimentary exporter nearly complete in pure python using numpy and such at the moment. Needs some adjustments for the XDF to be readable in sigviewer. I was also planning to add a few features not included in the LabRecorder / listed here. I will post here when I am done but I'm going to be delayed until after my MCAT in Mid-Jan. I'll be in touch.

After I get the python basic functionality up and running, I'll toss my code onto GitHub and post a link here. I also plan to re-implement it in C or C++ afterward.

[tstenner]

I've also started a prototype, could you upload your not-yet-ready code so I could take a look at it?

[LMBooth]

I don't suppose there's been any progress on this front from anyone? @tstenner @Argzero

[LMBooth]

For anyone looking for a python wrapper for an xdf writer, @agricolab has done a great job with pyLieSL, following the example for the LabRecorderCLI here.

[pablomainar]

Hi, any progress on this? I'm looking for a way of loading an XDF, modifying it and saving it back. I have checked pyLieSL but it doesn't seem to be able to do that (correct me if I'm wonrg).

[agricolab]

No, it can't.

[s-de-haan]

Would be great to have this feature

[cbrnr]

@rob-luke, you mentioned elsewhere that you (or someone from your group) might be interested in contributing. Is this still true? Do you want to discuss your ideas here? I don't think that anyone else currently has the resources to implement a writer.

[rob-luke]

Hi @cbrnr,

Indeed, we have written an XDF writer for our needs at our workplace. Sorry for the delay in responding to you, I have taken some time to look at both code bases and look at potential barriers to submitting the code here. I will describe the code we have and the potential barriers to merging. Then we can decide how to proceed together.

• pxydf only requires numpy. Our code requires pydantic. Would you be open to adding pydantic as a requirement? (I understand that adding requirements is a big decision, so let's decide this up front)
• Our writer is not a streaming writer as described in this issue. It simply writes an entire file in one shot. Would this be acceptable?

Thanks for considering these issues. It would be great to clarify these issues up front. The next step would be to decide on an API, then I could open a PR. Thanks 🙏

[cbrnr]

Regarding the second point, I don't think we intended to implement a streaming writer. This is already done by LSL, and the purpose of an XDF writer here would be to dump data (which is already available in some form TBD) into XDF.

Re pydantic, would it be possible to remove the type checks? Or does your writer critically rely on this functionality so it doesn't work without it?

[rob-luke]

Re pydantic, would it be possible to remove the type checks? Or does your writer critically rely on this functionality so it doesn't work without it?

Unfortunately, it critically relies on this and won't work without it. My primary concern here is that this will change the versions of python that pyxdf would work with. Plus the additional changes to maintenance etc.

[cbrnr]

This would require Python 3.7+, which is fine with me. I've never worked with pydantic, but I assume it could be replaced by data classes (and removing all type checks) if we really don't want it?

[rob-luke]

I assume it could be replaced by data classes (and removing all type checks) if we really don't want it?

Correct. Pydantic handles lots of checking of types and conversion too, to ensure you pass in the right data. So this would need to be all written manually instead.

[cbrnr]

OK, so it could be done, and the least effort solution would be to not check and convert at all, but assume that the input is in the desired format.

Let's wait what others have to say though. @cboulay @tstenner @agricolab