Complete rework of our example file handling

[jpfeuffer]

I dont like this urlretrieve idea to master at all.

Data and code will inevitably diverge. And binder notebooks are duplicating files, spending ages to download + are very hard to test.

We should parameterize URLs with a sphinx extension to point to the correct branch (while someone is at it, they could add the banner message that comes with the extension)

Regarding differences between ReadTheDocs and binder (where we can just use local files): a) Write a pandoc filter that replaces urlretrieve calls to the pyopenms-extra repo to local links

or

b)

• Add helper.py in the repo
• try import helper.py at the beginning of every notebook (will fail silently for copied snippets)
• Redefine urlretrieve in helper.py:

  def urlretrieve(URL ... same params as urllib):
  if (exists(URL - everything before /src/.. )):
    return localURL
  else:
    import urllib.urlretrieve
    return urllib.urlretrieve(*)

[timosachsenberg]

Data and code will inevitably diverge. And binder notebooks are duplicating files, spending ages to download + are very hard to test.

Yes they should point to the same branch/tag to prevent them from diverging. For :latest this is not a problem right? As it points to master anyway. For notebook at a certain version tag it could fail.

Is the downloading / testing an issue right now? I think the large (slow) example uses data not contained in this repo but downloads it from buildarchive.

[jpfeuffer]

It is a problem on latest, for testing PRs.

Not a big issue. But is messy (e.g. copies data into the folder with the notebooks) and takes some extra seconds.