removing sgskip

[choldgraf]

fingers crossed that the tests magically pass...

[anntzer]

looks like you need to remove the coding utf8 line (but that may be considered as a s-g bug too)

(I guess you may want to see whether other files were sgskipped for the same reason)

[choldgraf]

hmm - let's see if removing that line solves the problem and I can look into whether this is an issue elsewhere too

[anntzer]

The other files that could be affected are userdemo/pgf_*.py, but they probably need to be modified a little bit: they use a noninteractive backend, even though they do ultimately generate a png file so at least in theory s-g may be able to handle them... I guess a workaround is to indeed write their output, but then add a line that reads the png and displays it using an interactive backend.

EDIT: they actually need the coding set (for py2) as they do contain unicode... but that line needs to be moved before the docstring

[choldgraf]

how could I test this out on my computer? last night I tried building sphinx gallery w/ both python 2 and 3 on my machine and it worked OK, so it made things a bit hard to debug. Did you figure out how to make this error on your computer?

[anntzer]

test what? right now python2 examples/userdemo/pgf_fonts_sgskip.py just errors out, but works with py3

[choldgraf]

huh - it was working for me on python2 that's what was confusing me

[choldgraf]

oops sorry, not pgf_fonts_skip, I was talking about demo_text......lemme try this one

[choldgraf]

ok I reproduced the error on my end too. So I have two questions about these examples:

1. If we move the encoding to the top line, it works, but won't that make SG error similarly to the original example in this PR?
2. I tried adding a call to mpl.use but got errors since pyplot had already been imported with a backend. Is there a way to do switch the backend on the fly without creating an error?

[anntzer]

1. I think this is really a bug in s-g, they need something like add strip_encoding_cookie transformer ipython/ipython#3662.
2. I thought you could switch between noninteractive backends, or from a noninteractive to an interactive backend? If not I dunno.

[choldgraf]

@anntzer so it'd be fixed if s-g stripped the encoding line before trying to do anything? I can open an issue though it's unclear whether I've got my head wrapped around this problem totally...encoding w strings always confuses me

[anntzer]

I think so.