Consider stripping out only IMAGES as option about nbstripout

[kynan]

Interesting, that sounds like a useful option. How do you see the "UI" of this option?

Suppose this could even be easily generalized to support any data type to either keep or strip out.

[rsvp]

As an example, how about along these lines?

$ nbstripout --only=images [notebook(s)]
#  where images is just an alias for *.png, *.jpg, etc.

$ nbstripout --only='tmp-*.png' [notebook(s)]
#  _wildcard support_, say, for inconsequential images labeled as such by user.

[jpeacock29]

I thought about this as I was implementing --keep-count and --keep-output and considered an options structure:

--keep-count, -c
--keep-output, -o
    --keep-text-output, -t
    --keep-image-output, -i
--keep-metadata, -m

Thus nbstripout -o would be equivalent to nbstripout -ti. Stripping only images would use nbstripout -ctm. For more fine-grained control at the cell level, keep_output-type metadata would suffice. That said, this doesn't scale well as maybe there are more output types we'd consider in the future. (Widgets?)

[rsvp]

@jpeacock29 Re: more output types. Consider a flag like --regex, -r

where a user can implement their custom stripout by giving a regular expression.
This is what I did with sed -- then understood that an edited notebook
must leave in a trusted state. Going from regex to images would
then be a one-liner.

[kynan]

Sounds quite sensible! Let's start with images since this seems to be the most important case.

One of you happy to have a go at this @rsvp @jpeacock29 ?

[jpeacock29]

What would the regex flag be matched against? Every key in the ipynb?

[rsvp]

@jpeacock29 here's a mock example for PNG images: $ nbstripout --regex='png": "i'

That should get rid of the super long lines encoding PNG images for now.
I question the permanence of that regular expression since it is subject
to the formatting whims upstream. Historically, both "image/png":
and "png": have been used as keys, so my example would fortunately
work for both cases.

But --regex would be a handy tool in any case.

[kynan]

Feel free to have a go at this, I'm not sure when I might find time to work on it myself.

[kynan]

One of you interested in working on this @rsvp @jpeacock29 ?

[kynan]

Are you still interested in this @rsvp @jpeacock29 ?

[rsvp]

hi @kynan my spare cycles are going to refactoring https://git.io/fecon235
so realistically maybe later this year.

Interestingly, one of the reasons leading to the spin-off of the source code
to another repository https://git.io/fecon236 was to leave behind all
the archival bulky images preserved in the .git for notebooks.

So this issue is still pertinent.

[kynan]

@rsvp do you still have this use case?

[kynan]

There's an in flight pull request (#135) that's somewhat related: only strip outputs that are larger than a certain size. Would that fit the bill?

[kynan]

Given #135 has been released in nbstripout 0.5.0 and is arguably even more flexible than what's requested here I'll close this as fixed.