1929955 - Rounding issue fixup

Status: RESOLVED FIXED

(Core :: CSS Parsing and Computation, defect)

Description

[Fatih Kilic [:fkilic]]

Fixup of https://bugzilla.mozilla.org/show_bug.cgi?id=1873382#c23. it looks like mAppUnitsPerDevPixel uses int instead of float and that causes different dPR between RFP and non-RFP.

Comment 1

[Fatih Kilic [:fkilic]]

Attachment: Bug 1929955: Correct dPR calculation. r?tjr

This patch still doesnt match values 100%, and to be honest, I have no idea why. I tried every possibility but still couldn't make it match 100%. I suspected this line was causing a minor rounding issue, but it wasn't the case. I also tried removing and adding float(...)s to the calculation but that didn't help either. So now, there's a very small rounding issue, but matchMedia seem to match without an issue because it is very very small difference.

Comment 2

[Fatih Kilic [:fkilic]]

Copying and pasting the comment on phab:

For context, the difference is

• 110% zoom, in rfp: 2.222222328186035, in non-rfp 2.
• 120% zoom, in rfp: 2.4000000953674316, in non-rfp 2.4

I think I tried everything and yet still get this rounding issue.

The code I wrote is also 100% same with the actual calculation. So I'm not sure what is going on.

Comment 3

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/d75e6033017d
Correct dPR calculation. r=tjr

Comment 4

[agoloman]

https://hg.mozilla.org/mozilla-central/rev/d75e6033017d

Comment 5

[Thorin [:thorin]]

Attachment: all default zoom levels.jpg

FWIW ... very close on my device with a real DPR of 1 (I'll check other real various DPRs at some stage)

in the examples

• 90% = 1.81818 (spoofed)
• 110% = 2.22222 (spoofed)

modify toolkit.zoomManager.zoomValues and add 1.8 and 2.2 zoom levels, test without RFP

MediaQueryList { media: "(resolution: 1.71429dppx)", matches: true, onchange: null }
MediaQueryList { media: "(resolution: 1.81818dppx)", matches: true, onchange: null }
MediaQueryList { media: "(resolution: 2dppx)", matches: true, onchange: null }
MediaQueryList { media: "(resolution: 2.22222dppx)", matches: true, onchange: null }

the values are the same, so IDK what gives :)

Comment 6

[Fatih Kilic [:fkilic]]

I'm not sure what you mean (literally, my English fails me very often). Are you saying it is still not 100% correct? or are you verifying that it works?

I get the following values on a naturally 2 DPR device
non-RFP

• 30%: 0.6
• 50%: 1
• 67%: 1.3333333333333333
• 80%: 1.5789473684210527
• 90%: 1.8181818181818181
• 100%: 2
  RFP:
• 30%: 0.6
• 50%: 1
• 67%: 1.3333333333333333
• 80%: 1.5789473684210527
• 90%: 1.8181818181818181
• 100%: 2

With layout.css.devPixelsPerPx set to 1, I get
non-RFP:

• 30%: 0.3
• 50%: 0.5
• 67%: 0.6666666666666666
• 80%: 0.8
• 90%: 0.8955223880597015
• 100%: 2
  RFP:
• 30%: 0.6
• 50%: 1
• 67%: 1.3333333333333333
• 80%: 1.5789473684210527
• 90%: 1.8181818181818181
• 100%: 2

The numbers I get seem to match with spoofed and naturally 2 dPR. I used window.devicePixelRatio though, is it somehow different for media queries?

Comment 7

[Thorin [:thorin]]

edit: noticed something, will post image soon

---

We're comparing values per test, not across two different configs, although that helps and is something I plan to do - i.e that we're consistent across each dpr value regardless of how it was achieved

What I did was on the resize event of TZP (locally, not live) I added a console log for window.matchMedia('(resolution: ' + window.devicePixelRatio + 'dppx)') and then just caught everything in the console as I zoomed from 30 to 500. The image shows at least two that do not match, and this is device with DPR == integer (so we would expect subpixel entropy to not occur, so to speak)

Something is going on - but IDK what. There's also -moz-device-pixel-ratio, -webkit-min-device-pixel-ratio (which rounds down to 0.25, 0.5, 1, 2 or 4 ), dpcm and dpi.

Comment 8

[Thorin [:thorin]]

Attachment: hmmm.png

false alarm

device is DPR 1, RFP is off ... zoom level 110% (nightly) and there's a diff

• 1.0909091406250007 (dppx) vs 1.0909090909090908 (dpr)
• but that's because I'm using a binary search between resolution and max-resolution to return the min value

Comment 9

[Fatih Kilic [:fkilic]]

hmm it is still interesting that media query doesn't always match. I'm also not sure. I suspect float conversions but I'm also not sure how to solve it.

Comment 10

[Thorin [:thorin]]

continuing over at https://gitlab.torproject.org/tpo/applications/tor-browser/-/issues/43239