[wdspec] browsingContext.print: fix rounding error in page.py test

[thiagowfx]

pytest uses:

def cm_to_px(cm): return round(cm * 96 / 2.54)

js uses:

const viewport = page.getViewport({ scale: 96. / 72. });
...
canvas.height = viewport.height;
canvas.width = viewport.width;

This produces a rounding error, even though the dimension is correct:

>       assert cm_to_px(expected_dimensions["height"]) == height
E       assert 454 == 453
E         +454
E         -453

The inconsistency of rounding in both ends becomes clear when we eliminate "round" in the pytest side:

>       assert cm_to_px(expected_dimensions["height"]) == height
E       assert 453.54330708661416 == 453
E         +453.54330708661416
E         -453

There are multiple ways to fix this issue.

Option #1: Use "floor" instead of "round" in pytest. This approach fixes some tests, but others still fail. I guess in some cases "round" is correct.

Option #2: Use a range in the assertion comparison, allowing a difference of up to +-1.0. This is what this PR does.

The comparison is performed in
assert_pdf_dimensions.

The problematic part is .96 / .72 which evaluates to 4/3 = 1.333333....

[thiagowfx]

Source.

[thiagowfx]

I confirm this fixes all failing tests on our end:

[thiagowfx]

@foolip rounding is hard, eh? ;)

[thiagowfx]

Here's another way that works @whimboo:

assert height in [floor(cm_to_px(expected_dimensions["height"])), round(cm_to_px(expected_dimensions["height"]))]
assert width in [floor(cm_to_px(expected_dimensions["width"])), round(cm_to_px(expected_dimensions["width"]))]

In case you don't like the approach to compare to a +-1.0 range.

[thiagowfx]

I've discussed this topic with @jgraham and we think it will be fine to keep what you have right now. It shouldn't matter too much if there is a different of 2px or 1px by using floor.

@jgraham could you merge this?

Looking into the logs, it seems to me the failures are unrelated (but please double check).

[whimboo]

Thanks!