Add hidpi support

[christoph-heinrich]

In a mixed dpi setup, setting the resolution in the config is not sufficient for good results.

[hooke007]

It would break the output in my device.

[christoph-heinrich]

How come things keep breaking for you when it comes to hidpi stuff x.x works without a problem for me with any scaling value, be it integer or fractional.

[hooke007]

I quickly check the possible values. If scale =1.1xxxx , it will be broken. Too weird.

edit: 2.1xxxx 3.1xxxx meet with the same issue

[christoph-heinrich]

Huh, 1.09 - 1.16 also breaks for me... that is so weird.

Edit: Now I don't know what kind of sorcery is going on here, but but that makes absolutely no sense at all...
If i set scale to 1.0 and have a max width of 220 in my config, resulting in a thumbnail size of 220x121, it works.
HOWEVER if I set scale to 1.1, and have a max width of 200 in my config, resulting in a thumbnail size of 220x121, it doesn't work.
I've tried it multiple times already, it's repeatable... no idea how that's even possible.

[hooke007]

Perhaps the filter dislike fractional numbers.

effective_w = math.floor(options.max_width * scale)

effective_h = math.floor(options.max_height * scale)

[christoph-heinrich]

With the rounding it works. No idea why my previous example didn't work, as 200*1.1 should still be a clean 220, and the output also didn't show any comma digits, but oh well. Thanks for finding that.

Edit: I've been trying out a few things, inserting rounding in different places, and it makes less and less sense. Things that should be identical just don't work, sometimes the player even crashes. Maybe I'm missing something because I'm awfully tired. But what's important is that the commit above works.

[po5]

If your main mpv instance is crashing, or more accurately freezing, it's caused by an overlay-add stride parameter that doesn't match the thumbnail's size.
This had taken me forever to figure out and get reliable.

[christoph-heinrich]

I already suspected that it must be something like that, but I had rounding in draw() to make sure it's exact integers and I printed the scale parameter that's passed to the new mpv instance, which also contained the numbers as integers. So any place where it actually matters the numbers were the exact same integers and yet it only worked when it gets rounded directly in the calculation. Thus my confusion.

But the commit above works reliably afaict, and that's what matters.