Add support for iterating EXtra-data SourceData objects

[philsmt]

While extra_data.SourceData does not have native support for iterating yet, I figured it's actually fairly easy to do on the pasha side. This has the upsides that it will continue to work for older versions even once added there.

@takluyver While not really part of the public API, would you consider the initializer of DataCollection private?

[codecov]

Codecov Report

Merging #14 (c4cec68) into master (b9e7564) will decrease coverage by 2.26%.
The diff coverage is 0.00%.

❗ Current head c4cec68 differs from pull request most recent head 769f086. Consider uploading reports for the commit 769f086 to get more accurate results

@@            Coverage Diff             @@
##           master      #14      +/-   ##
==========================================
- Coverage   91.34%   89.07%   -2.27%     
==========================================
  Files           5        5              
  Lines         358      366       +8     
==========================================
- Hits          327      326       -1     
- Misses         31       40       +9     

Impacted Files	Coverage Δ
pasha/functor.py	68.04% <0.00%> (-7.24%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update b9e7564...769f086. Read the comment docs.

[takluyver]

While not really part of the public API, would you consider the initializer of DataCollection private?

🤔 I wouldn't encourage people to depend on it, especially passing lots of parameters. I try to avoid breaking even private things if practical, but e.g. I changed the meaning of the second parameter in 1.10 (European-XFEL/EXtra-data#274) in a way that would probably break something like this. And I can imagine that the 'origins' idea we talked about earlier today might involve further changes.

[philsmt]

e.g. I changed the meaning of the second parameter in 1.10 [...] in a way that would probably break something like this.

Indeed, I thought I remember the initializer differently 😉

Ultimately I would like to move the ExtraDataFunctor into EXtra-data proper, so it can freely use internal APIs. I just happened to want to iterate over sources several times in the past weeks. So maybe that's the best course of action?

[takluyver]

Can we design it so that pasha can recognise EXtra-data objects without either having to import the other? E.g. IPython looks for a _repr_html_() method (among others) to display objects in the notebook - we could look for something like a _pasha_functor_() method.

[philsmt]

Of course, that was the intent. pasha supports this via a _pasha_functor_ method that you actually inspired 😉

[takluyver]

I see, so it needs to return something with .split() and .iterate() methods. That should be doable.

If .split() is expected to return a list of slices for plain integer indexing, I wonder if we can simplify this a bit, so the returned object can define __len__() rather than .split(), and pasha will call gen_split_slices() using the len() it gets.