Differences from matplotlib

What is equal, and how it is checked, is on Exactness. This page is the rest.

What does not equal matplotlib

Text

Glyphs are filled outlines (go-gfx coverage), not FreeType’s hinted rasterisation with matplotlib’s text antialiasing, so the pixels of text differ, and a string’s box is its advance where matplotlib’s is FreeType’s hinted box (up to a few pixels wider on long labels). Positions are matplotlib’s (Text._get_layout: line metrics from the OS/2 table, alignment, rotation, anchoring), and like Agg the pen is put on a whole pixel. Kerning comes from DejaVu’s GPOS/kern tables through go-opentype.

Math text is TeX, not mathtext

$...$ is typeset by go-tex/math with STIX Two Math — real TeX layout (fractions, radicals, big operators, limits, matrices, \left...\right), which matplotlib’s mathtext only approximates. The exception is \mathdefault{...}, what matplotlib’s tick formatters emit: it is set in DejaVu Sans with mathtext’s own superscript rules, so tick labels match matplotlib’s look and position. See TeX labels.

Strokes

Anti-aliasing is go-gfx’s, not Agg’s; widths, caps, joins and dashes are matplotlib’s. Pixel snapping follows matplotlib’s rule (odd widths at pixel centres, even at edges) for paths made of horizontal and vertical segments. Strokes of more than 512 segments are drawn in pieces and get round ends (see performance).

Images

Images are resampled nearest-neighbour when enlarged and box-averaged when reduced; matplotlib’s default (interpolation="auto", antialiased resampling in data space) differs when an image is shown smaller than its resolution, or less than three times enlarged.

Contours

Contours are marching squares with the saddle resolved by the cell mean, not contourpy’s serial algorithm (corner_mask, chunking): levels and colours are matplotlib’s, line vertices differ slightly. Filled bands are drawn per grid cell.

Legend “best”

"best" counts the vertices, segments, boxes and scatter points under each of the nine candidate positions and takes the first with none, else the least; matplotlib’s badness also weights offsets and is computed in display space with a pad. Usually the same place; not always.

Annotate

Annotate draws arrowstyle="->" only, shrunk 2 points at both ends.

SVG

The SVG carries the same geometry but not matplotlib’s document structure, ids or metadata.

No module-level pyplot

plt.plot, plt.xlabel and the other functions acting on a hidden current Axes are not provided, by design: see the API.

Not in v0

• suptitle, and TightLayout with explicit pads or rect;
• symlog, logit and date axes;
• minor ticks on linear axes (off by default in matplotlib too) and AutoMinorLocator;
• constrained_layout;
• subplots sharing;
• hatching;
• boxplot, violin, pie, quiver, stem, step, pcolormesh, 3-D;
• colorbar extend and discrete norms;
• image interpolation other than nearest;
• Text bounding boxes;
• legend columns, titles and bbox_to_anchor;
• native and browser windows (go-widgets/window);
• PDF output.