Is this the right place to comment on style changes, or is there a discussion thread somewhere else?

If you have specific issues related to these changes feel free to make them here, but the ship has largely sailed on subjective decision making.

bon voyage, then

Sorry if I seemed prickly -- that wasn't my intention. The longer story is that we have already put out the request for suggested style changes (through promotion at conferences, the mailing list and a banner on the website) -- that generated many PRs from which Thomas and I selected the best options, taking into account "best for the common case" issues, and consistency etc. This PR is the culmination of that and is largely in the round of finding bugs that it reveals, fixing poor interactions between choices etc., but the big design stuff is mostly over so that we can make a 2.0 release without too much further bikeshedding. All that said, we sincerely hope that the style infrastructure will make it easier for those the disagree with the defaults to easily override them.

As I understand it, all the test images should still be using the classic style. Something appears to be wrong with the handle length in legends as they seem to have widened in several images. It is most evident in lib/matplotlib/tests/baseline_images/test_legend/framealpha.png where it reaches the edge of the legend.

Also, maybe the what's new page should mention the classic style?

I would really like to see more changes to make the legend not so spacious, the defaults are quite unuseable with more than two entries. Also i really dislike the rounded corners, which is a change i never saw discussed anywhere :(.

I would also like to see the default markersize decreased.

@Tillsten wrote:

I would really like to see more changes to make the legend not so spacious, the defaults are quite unuseable with more than two entries. Also i really dislike the rounded corners, which is a change i never saw discussed anywhere :(.
I would also like to see the default markersize decreased.

Thanks for the feedback. As things get integrated, it may be that this does or does not make sense. Unfortunately, the official window to provide style suggestions has closed, but we're obviously still open to feedback about things in the existing changes that don't interact well or make specific kinds of common plots much worse.

A comment about the inset locator demo. With the new colors, it is hard to see the inset lines. Personally, I liked the demo the way it was. The new colors there are very dark.

Also, the ticks in that demo just seems "weird" to me. There is no x tick on the left corners, but there is one on the right corner.

Is the current rendering of the docs with this PR available somewhere? Would be interesting to look at the gallary.

A comment about the inset locator demo. With the new colors, it is hard to see the inset lines. Personally, I liked the demo the way it was. The new colors there are very dark.

Makes sense. Easy enough to put it back to the old colors.

Also, the ticks in that demo just seems "weird" to me. There is no x tick on the left corners, but there is one on the right corner.

Good point. I don't think it's a bug -- it's just the new view limiting coming into play. But we could hardcode the limits or change the locator to fix the weirdness.

Is the current rendering of the docs with this PR available somewhere? Would be interesting to look at the gallary.

Not presently, but I'll put it up somewhere to make that easier to do.

There are some changes to the inset locator demo to disable the auto-ticking which are on master but not in this PR.

Thanks for all the helpful feedback, @WeatherGod -- and sorry for not pushing and sending you on some duplicate work.

That test image you uploaded still fails (~5 RMS)

That test image you uploaded still fails (~5 RMS)

Yeah -- I think #5932 fixes that correctly. I'll just wait for that to merge and rebase.

@mdboom That should be on the 2.x branch now since #5934 was merged

@mdboom: Has the proposal in #4730 (using labels 0.1, 1, and 10 in log-scaled axes) been accepted as a default change? I cannot find an implementation with rcParams, but I could whip one up.

@zblz: In answer to your question, #4730 is not accepted as a default, but will be welcomed as an option with rcParams.

Merging this as-is so we can see the built docs.

The 3.5 failure looks like travis flakyness, but I will fix in a follow on PR if that is really broken.

Now that this PR is merged, where do we give feedback? I see some bugs in the doc build already...

New issues is probably easiest.