Incorrect pixel calculation on Chrome 98

[sculpt0r]

Type of report

Bug

Provide detailed reproduction steps (if any)

We have a lot of failing tests due to the fact that Chrome 98 starts produce different values in pixel calculations. The difference is 1px - but still... Below is the list if failing tests:

• http://0.0.0.0:1030/tests/core/tools (3 failing)
• http://0.0.0.0:1030/tests/plugins/pastefromlibreoffice/generated/generic (3 failing)
• http://0.0.0.0:1030/tests/plugins/pastefromlibreoffice/generated/genericwithword (3 failing)
• http://0.0.0.0:1030/tests/plugins/pastefromlibreoffice/generated/images (3 failing)
• http://0.0.0.0:1030/tests/plugins/pastefromlibreoffice/generated/imageswithword (3 failing)
• http://0.0.0.0:1030/tests/plugins/pastefromword/generated/excel (3 failing)
• http://0.0.0.0:1030/tests/plugins/pastefromword/generated/generic (17 failing)
• http://0.0.0.0:1030/tests/plugins/pastefromword/generated/list_indentation (4 failing)
• http://0.0.0.0:1030/tests/plugins/pastefromword/generated/table (2 failing)
• http://0.0.0.0:1030/tests/plugins/pastefromword/generated/tickets1 (32 failing)
• http://0.0.0.0:1030/tests/plugins/pastefromword/generated/tickets2 (16 failing)
• http://0.0.0.0:1030/tests/plugins/pastefromword/generated/tickets3 (16 failing)
• http://0.0.0.0:1030/tests/plugins/pastefromword/generated/tickets4 (36 failing)
• http://0.0.0.0:1030/tests/plugins/pastefromword/generated/vertical_margin (3 failing)
• http://0.0.0.0:1030/tests/plugins/pastetools/filter/functions (1 failing)

Expected result

What is the expected result of the above steps?

All test are passing on the Chrome 98 (same as Chrome 97)

Actual result

What is the actual result of the above steps?

Tests are failing.

Other details

We should verify is this an upstream - and what exactly?
Is that causing some serious damage to the editor? Is it crashing at some point or is useless?

• Browser: Chrome 98
• OS: macOS
• CKEditor version: 4.17.2
• Installed CKEditor plugins: (dev version)

[jacekbogdanski]

I took a quick look at this issue. From what I see indeed Chrome 98 changed the way it calculates units - I didn't dig what's actually happening on the Chrome side, but we are using DOM-based unit conversion in convertToPx function.

This upstream seems serious as it changes calculations for very low values like rounding points to 2px instead of 1px for table borders when pasting HTML from Word. Some time ago I've reported a ticket for using simple math for calculating absolute units instead of using DOM container - #5097. It should resolve the issue for absolute units, but not sure what about relative ones.

Nevertheless, the issue is caused mostly by convertToPx function and that it's based on browser calculation. Fixing this function solves the issue for PFW (I already checked it by using math for calculating points to pixels).

[Comandeer]

It seems like an upstream issue: https://bugs.chromium.org/p/chromium/issues/detail?id=1302246

However, the workaround seems to be easy: just use the value from element#getBoundingClientRect() and do the rounding ourselves.

[Comandeer]

I've changed the type of the issue to the bug as it was actually influencing conversion of Word content inside the CKEditor 4.

[CKEditorBot]

Closed in #5109