Consider the attached testcase. In Gecko, the two percentages on each line are the same. In stylo, the following lines have them different: 15% 15.000001%; 27% 27.000002%; 30% 30.000002%; 53% 52.999996%; 54% 54.000004%; 59% 58.999996%; 60% 60.000004%; This was caught by some mochitests. Specifically, this causes dom/base/test/test_bug338679.html to fail. Good thing it happened to use "15" as the value... If I had to guess, servo is taking floats, converting to double, then serializing the double.

Actually, looks like Gecko just asks the float-to-string machinery to produce 6 decimal digits of precision, rounding the last one; this is the general behavior of nsTSubstring_CharT::AppendFloat (the version that takes 32-bit floats). The errors above are all in the 8th digit, so they get rounded away.

Filed https://github.com/servo/servo/issues/17205

A potential fix at https://github.com/servo/rust-cssparser/pull/173

This seems to affect several tests, so bump to P2.

Pushed by xquan@mozilla.com: https://hg.mozilla.org/integration/autoland/rev/a203d42085ea Bump version of cssparser.

Pushed by xquan@mozilla.com: https://hg.mozilla.org/integration/autoland/rev/04080bc695e3 followup - Bump version of cssparser again to 0.19.4.

Pushed by xquan@mozilla.com: https://hg.mozilla.org/integration/autoland/rev/e16dba457260 followup 2 - Downgrade cssparser to 0.19.2.

Pushed by xquan@mozilla.com: https://hg.mozilla.org/integration/autoland/rev/6b0ed8c31532 followup 3 - Update several web-platform test expectation.

https://hg.mozilla.org/mozilla-central/rev/a203d42085ea https://hg.mozilla.org/mozilla-central/rev/04080bc695e3 https://hg.mozilla.org/mozilla-central/rev/e16dba457260 https://hg.mozilla.org/mozilla-central/rev/6b0ed8c31532