GeezerVison

Hmm, something might still be cached somewhere as changing the hinting factor in the middle of a script causes all text to be the wrong width.

for hinting_factor in map(int, sys.argv[1:]):
    plt.rcParams['text.hinting_factor'] = hinting_factor
    fig, ax = plt.subplots()
    ax.text(0.5, 0.5, 'text')
    plt.show()

Opens figures with progressively smaller text (or wider depending on what you specify), but it's fine if you specify anything other than 8 in a separate run.

As an example of something this PR does fix, if you add a fig.savefig('test.svg') before show (to switch backends), you get broken text for anything other than 8 on master.

@QuLogic but I guess that caching was already present before, in which case this PR does not really make things worse, does it?

@anntzer True, it just makes it more difficult to write a test.

I think that should fix the caching issue, and it even fixes the SVG+Agg save from above, without the first change. I guess the first change can still go in for consistency.

But thinking about it, SVG and PDF render text externally, so it may be better to have those use hinting_factor=1 since that's the most likely method to be used by external renderers. Maybe can discuss that in a separate issue where we also might decide to turn it off entirely.

Ah, the static was tricky...
I would favor moving the discussion of entirely dropping hinting_factor (which I agree with) to a separate issue, not clear there's much gain of doing so in a piecemeal fashion.

test failures look real though

Not sure why I needed the clear and set_size as I thought the same stuff is done in the constructor, but it matches what's done in normal rendering code and it works, so...

thanks