[jacobolus]

Try encoding it the obvious way as text instead and view your SVG file on different renderers, and you will often find that they messed up the fonts/text layout.

This is a defensive practice to ensure that your plot will look the way you intended.

[kragen]

yes, i understand why it's done, and i agree that svg doesn't offer a better alternative. that's because the only fonts svg guarantees the existence of are Serif, Sans-Serif, Monospace, Cursive, and Fantasy, none of which have guaranteed metrics. the reason i was surprised and appalled is that both of adobe's previous standards in the line from which svg descends (postscript and pdf) avoid that problem by guaranteeing the availability of certain core fonts with (in practice, though not in theory) well-defined standard metrics, and it makes svg a much worse standard for small images that include text

with pdf.use14corefonts and ps.useafm turned on in matplotlib, my plot from http://canonical.org/~kragen/sw/dev3/plotrc.py is 8k as eps, 9k as pdf, 21k as png http://canonical.org/~kragen/sw/dev3/rc.png, and 24k as svg

(with optipng the png shrinks to 16k but you could definitely optimize the other formats too; matplotlib fills all of them with ridiculous amounts of bloat. but the embedded font is most of it, and it's only necessary in svg)

like, you really have to fuck your vector file format up for it to make a simple line plot format larger in it than as an antialiased png

or for it to make it three times larger than a fairly shitty eps or pdf

i like svg a lot (it's replaced postscript for me as the language i use for easy 2-d vector graphics) but it has some really serious deficiencies. this one is news to me; the other one i've run into is that there's no defined real-world scale, so i can't send an svg to a laser-cutting shop and ask them to cut it out at 1× or 2× scale. i have to use pdf or eps or dxf for that