Thanks for the issue.

I believe that we have another issue opened for that but I can't find it.
It ties into #10128 and #10255 and IIRC when we did go back and forth between changing the default and not and in both cases we got complaints so this is delicate.

In particular IIRC, the background color and default size if we don't change them are problematic.
But I believe its a good time to revisit these default.

Please remove this behavior of overriding the settings in matplotlibrc. I understand that the (ugly) matplotlib defaults cause problems out of the box. However, I think the current behavior is extremely confusing for new users trying to change these defaults. For users who do not understand that %matplotlib inline is actually using an element of IPython (not matplotlib, not Jupyter), the path to solving this problem can be long. It took me quite a while the first time.

These default settings also seem to produce figures that are too small for modern monitors. A dpi of 72 times a 4.0" height produces a figure less than 300 pixels high, which is a quarter the height of a 1920 x 1200 monitor, but figures rendered with these defaults always seem significantly smaller than this.

I remember that in past years a line like
c.InlineBackend.rc = {'dpi': 72, etc.}appeared by default in one of the IPython config files. However, the current files created byipython profile create` make no mention of the inline backend. Searching the IPython help for "InlineBackend" (no space) returns no results. The top results of googling "c.InlineBackend.rc" are threads of people discussing this issue, e.g. here.

Possible compromise: overwrite rc settings only if the current matplotlibrc file is not in the config directory, using mpl.matplotlib_fname().startswith(mpl.get_configdir()) or similar. This would avoid problems for totally new users but would allow people to use matplotlibrc without having to learn about IPython config and insert a mysterious line into a file that does not exist in a fresh install.

Thanks for your time.

I completely agree that we should now revert this behavior in InlineBackend, esp. now that %matplotlib inline isn't explicitly required anymore in most cases. This results in super confusing behavior, as we only do the override when the first plot call is made. This leads to this type of situation:

So if a user wants to set some rcParams at the top of their file, they basically get overridden on first plot call. They have to go back and rerun that first cell again for their changes to take, for no comprehensible reason. I just spent a good 1/2 hour debugging this just now, even though I know I was one of the people who wrote that code years ago! :)

I think we should fall back to being as hands-off as possible now, and leave this configuration to mpl/the users. We made those choices for good reason early on when the notebook was very new (I actually think we started that with the Qt console), but now I think it causes more confusion than good. I think we should revert our overrides, and let folks make their fine-tuning as they wish, using standard mechanisms.

My vote would be to drop all our overrides here and leave things clean (note that code now lives in the separate matplotlib_inline package).

I ran some quick tests manually restoring the values we override, and things seem to work fine.

That's reasonable, we could provide a matplotlib notebook "style" for backward compat.

👍 I have a few min in between meetings, let me whip up a PR real quick :)

Oops!!! Sorry - I was being lazy and used the in-browser code editor (.) and I didn't realized it pushed straight to master!! 😱

I quickly force-pushed to revert, and will make a PR next. Serves me well! 🤦

OK, now it's there, properly made. I have never touched that repo, so LMK if you'd like a bit of doc update too, happy to do it. gotta run now...

It's been a few months since this issue was fixed. If anyone here is also a maintainer on matplotlib-inline, could you publish an updated version?

@davidnero - see here where we're discussing indeed making a release. I'm volunteering for it b/c I need this before I start teaching!

Should happen soon.

I'm going to close this as the real issue is in the matplotlib-inline repo, fixed and only pending release of a new version.