On Tue, 23 Apr 2024 19:20:22 GMT, John Hendrikx <jhendr...@openjdk.org> wrote:
>> tests/system/src/test/java/test/com/sun/javafx/text/TextLayoutTest.java line >> 453: >> >>> 451: TextLayoutTest::assumeMac, >>> 452: "The quick brown लोमड़ी jumps over the lazy कुत्ता", >>> 453: 160.0f, List.of(155.3047f, 91.04719f) >> >> Question: these numbers are compared exactly (Float::compare), can we expect >> them to be exactly the same on all possible platforms/versions? > > Exact comparisons on float numbers should not be a problem, as long as you > don't assume that with a different order of operations you will get the same > result. For example, `y + x - y` may not exactly equal `x + y - y` or `x`. > Since we don't have different code paths for different platforms, and since > all code paths are deterministic(*), it should be fine to do an exact > comparison. > > If this ever does fail, I think then is a good time to do a more thorough > investigation. So far these tests have run dozens of time on my machine, and > on several of the machines of people reviewing without problems. > > (*) In tests, the most often encountered source of non-determinism in Java is > when a `Set` or `Map` is involved which contains objects with the default > hash code which can change each time a new JVM is created; usually I change > such collections to linked versions, or do an explicit sort somewhere to > ensure it is deterministic again. makes sense, thank you. ------------- PR Review Comment: https://git.openjdk.org/jfx/pull/1236#discussion_r1576788278
