There is a significant performance regression on the wire3d_animation example.

test_backend_ps

test_savefig_to_stringio[ps with distiller=xpdf-landscape-letter]

xpdf output differences?

test_ft2font

test_fallback_missing

Font fallback is not implemented.

test_image

test_figimage[pdf], test_figimage0[pdf], test_figimage1[pdf], test_interp_nearest_vs_none[pdf,svg], test_rasterize_dpi[pdf,svg]

Invalid dpi manipulations in vector output.

test_jpeg_alpha

Not respecting savefig.facecolor.

We appear to be leaking memory per memleak.py.

The main list of xfailed tests is in run-mpl-test-suite.py. Other "interesting" tests are listed below.

Matplotlib's hatching is inconsistent across backends (#10034).

test_artist::test_clipping

Matplotlib's SVG backend does not implement Gouraud shading.

test_axes::test_pcolormesh[svg]

Matplotlib's partially transparent markers with edges are inconsistent across backends (#10035).

test_axes::test_rgba_markers[pdf]

Matplotlib does not write SVGs with image-rendering: pixelated (#10112).

test_backend_svg::test_noscale[svg]
test_image::test_rotate_image[svg]
test_tightlayout::test_tight_layout5[svg]

Matplotlib's software alpha compositor is incorrect (#8847).

test_image::test_image_composite_alpha[pdf,svg]

Matplotlib's draw_path_collection has inconsistent semantics across backends (#12021). (No test.)

Matplotlib's SVG backend writes fills as hex (#bf8040) whereas cairo writes them using rgb (rgb(75%,50%,25%)). The latter seems rendered less precisely when combined with transparency.

test_patches::test_clip_to_bbox[svg]
test_skew::test_skew_rectangle[svg]

#9963 (behavior with fontsize < 1pt)