The discussion on #5677 noted that the backport needed to await checking of the svg results. My first random check, http://matplotlib.org/devdocs/examples/lines_bars_and_markers/scatter_with_legend.html, shows a problem. Notice that the legend is plotting the symbol twice with slightly different sizes in the svg, but not in the png.
Conclusion: the necessary review of the svg output probably has not been done. I think that some comparisons of rendering time between svg and pdf is also needed. In the example above, the svg seems rather slow, much slower than the original pdf here: http://matplotlib.org/examples/lines_bars_and_markers/scatter_with_legend.html.

Overall, at this point I suspect that switching to svg is just going to make more work and degraded performance. The only benefit is sharper plots in the web page, compared to the present png.

On the axisartist page, several axis lines appear to be too thick. This may be an SVG renderer issue. For example, see SVG vs. PNG where the left line appears correct, but the right line is a bit thicker.

The pcolor images have that unfortunate edge thing that shows up in PDFs, which I don't think is a great thing to show by default.

On 2016/07/25 10:39 PM, Jens Hedegaard Nielsen wrote:

There are several questions.

• Should the change to use SVG's be backported?

I think the answer is a clear "No!" I'm inclined to recommend reverting
the change in master--maybe after taking advantage of the temporary
change to find problems in the svg backend. Overall, I don't think svg
is a very useful format. It's not as fast and reliable in web pages as
png, and there aren't many applications in which one can use it.

Implemented in #6862