209289 2020-03-19 09:59:40 -0700 [JSC][BigEndians] JSTests/stress/dataview-jit-set-nan.js subtest misbehaves 2020-03-19 12:02:40 -0700 1 1 1 Unclassified WebKit JavaScriptCore WebKit Nightly Build Unspecified Unspecified NEW P2 Normal --- 1 tpopela webkit-unassigned mcatanzaro oldest_to_newest 1631699 0 tpopela 2020-03-19 09:59:40 -0700 The whole background story could be found in bug 194007. While working on patch I found one case where the suggested solution doesn't work: "use strict"; function assert(b) { if (!b) throw new Error; } function getIsLittleEndian() { let ab = new ArrayBuffer(2); let ta = new Int16Array(ab); ta[0] = 0x0102; let dv = new DataView(ab); return dv.getInt16(0, true) === 0x0102; } let isLittleEndian = getIsLittleEndian(); function adjustForEndianessFloat32(value) { if (isLittleEndian) return value; let ab = new ArrayBuffer(4); let ta = new Float32Array(ab); ta[0] = value; let dv = new DataView(ab); return dv.getFloat32(0, true); } function test() { function storeLittleEndian(dv, index, value) { dv.setFloat32(index, value, true); } noInline(storeLittleEndian); function store(dv, index, value, littleEndian) { dv.setFloat32(index, value, littleEndian); } noInline(store); let buffer = new ArrayBuffer(4); let arr = new Float32Array(buffer); let bits = new Uint32Array(buffer); let dv = new DataView(buffer); for (let i = 0; i < 10000; ++i) { storeLittleEndian(dv, 0, adjustForEndianessFloat32(12912.124123215122)); assert(arr[0] === 12912.1240234375); assert(bits[0] === 0x4649c07f); } } test(); On big endian machine the arr[0] contains 6.905458702346266e-41 and bits[0] 0xc07f. If I replace the storeLittleEndian() call with: store(dv, 0, 12912.124123215122, isLittleEndian); then it does work, but it's strange that the storeLittleEndian() doesn't. Is the 12912.124123215122 number somehow special? If I observe what's in arr[0], when I change that number then it's: input to adjustForEndianessFloat32() : arr[0] value 12912 : 12912 12912.1 : 12912.099609375 12912.12 : 12912.1201171875 12912.124 : 6.905458702346266e-41 Does anyone have an idea?