capture rich output as well as stdout/err in capture_output

[minrk]

closes #3742

[ivanov]

I provided this feedback verbally to Min, but it'd be nice to get the rich displays as an attribute of the capture variable in the same way we can already get stdout and stderr. Min said he could make them available.

%%capture the_output
display(...)
...

and then the same way you can already do the_output.stdout, the_output.stderr
you'll be able to do

the_output.display[0]

and get a handle on that image, or whatever it was.

Pinging @qili1g, who was interested in this issue.

[qili1g]

I want to stay updated

[qili1g]

Thank you for update the code for the function "grab rich output", we test it, it works

[qili1g]

we need to maybe understand how far out or a release this would be

[ivanov]

it will be merged inside the code soon after we release 1.0, and will make the next stable release around December

[Carreau]

can we merge now ?

[ivanov]

let's not merge this without an update to the examples (see how I did the double negative there? I'm getting better not getting worse at this.

i'll cook something up and send Min a PR tonight

[ivanov]

I sent @minrk a PR that documents this new features, and adds a --no-display flag to %%capture (to go along with the --no-stdout and --no-stderr flags which were already there. Also added docstrings to the properties and the CapturedIO object. minrk#7

[minrk]

Thanks, merged your PR and rebased. I will try to add a few tests, so that this can be merged.

[minrk]

added some tests

[ivanov]

looks good - I was going to comment about how we decided a while back (scipy 2011?) that we shouldn't use generators in tests because nose would stop iterating on first failure, but by the looks of it, this issue has gone away (or never existed for the particular way in which you're using yield and nose). @fperez might remember better, maybe this only applied to the parametric tests?

[minrk]

Yeah - I chatted with @fperez about that a week or two ago. I can remove the generators, but I like the visual feedback they provide about how much is actually tested, rather than how many functions I happen to have written.

[minrk]

@ivanov want to write an IPEP about removing test generators :)