PHP floor函数精度异常问题及浮点数截断自定义函数咨询

Hey, let's dig into why your custom floorDec function is hitting precision snags and work out solid fixes for it.

为什么会出现精度异常？ #

The core issue here is how PHP handles floating-point numbers (the float type). PHP uses the IEEE 754 binary floating-point standard, which can't represent all decimal fractions exactly. For example, your input 96684700 divided by 10000 should be 9668.47, but in binary float storage, it might end up as something like 9668.469999999999—a value just slightly smaller than the actual decimal.

When you multiply that by $c (which is 100 for $decimals=2), you get 966846.9999999999 instead of 966847. Calling floor() on that will yank it down to 966846, and dividing back by 100 gives you 9668.46—not the 9668.47 you expected. That tiny initial error from floating-point storage gets amplified through your math operations, leading to the wrong result.

优化方案 #

Let's go through a few reliable fixes, ordered by simplicity and reliability:

方案1：用整数运算完全避开浮点数陷阱（推荐） #

Since your input $value is an integer (like 96684700), we can shift all operations to integer math, which has zero precision loss. Here's how:

function floorDec($value = 0, $decimals = 2) {
    $scale = pow(10, $decimals);
    // 先把数值放大scale倍，除以10000后向下取整，再缩回去
    $adjusted = (int)floor($value * $scale / 10000);
    return $adjusted / $scale;
}

// 测试一下
$a = 96684700;
var_dump(floorDec($a, 2)); // 输出 float(9668.47)

This works because we're doing all critical calculations with integers—no floating-point errors creep in. The final division by $scale converts it back to a float, but by that point, the value is exact.

方案2：用BCMath扩展处理高精度场景 #

If you need to handle very large numbers or non-integer inputs, PHP's BCMath extension is built for arbitrary-precision decimal arithmetic. It avoids floating-point issues entirely by working with strings:

function floorDec($value = 0, $decimals = 2) {
    $scale = (string)pow(10, $decimals);
    // 先把value除以10000，保留足够多的小数位避免截断
    $scaledValue = bcdiv((string)$value, '10000', $decimals + 4);
    // 乘以scale后取整（BCMath的bcmul可以指定0小数位实现向下取整）
    $integerPart = bcmul($scaledValue, $scale, 0);
    // 再缩回到目标小数位数
    return (float)bcdiv($integerPart, $scale, $decimals);
}

// 测试
$a = 96684700;
var_dump(floorDec($a, 2)); // 输出 float(9668.47)

Just make sure you pass values as strings to BCMath functions to avoid floating-point conversion before processing.

方案3：给原函数加微小偏移量（仅作临时修复） #

If you want to tweak your original function instead of rewriting it, you can add a tiny epsilon value to counteract the floating-point error before calling floor():

function floorDec($value = 0, $decimals = 2) {
    $c = pow(10, $decimals);
    $a1 = $value / 10000;
    // 加一个极小的数（比如1e-10），确保接近整数的浮点数被正确识别
    $a2 = $a1 * $c + 1e-10;
    $a3 = floor($a2);
    $a4 = $a3 / $c;
    return $a4;
}

// 测试
$a = 96684700;
var_dump(floorDec($a, 2)); // 输出 float(9668.47)

⚠️ Note: This is a hacky fix—you have to pick an epsilon that's smaller than the smallest meaningful decimal in your use case. It can introduce errors if your values require ultra-high precision, so use it only as a quick fix.

总结 #

The best approach is 方案1 (integer arithmetic) for your current use case, since it's simple and completely eliminates floating-point precision issues. For more complex or high-precision scenarios, go with BCMath.

内容的提问来源于stack exchange，提问作者lighter