int () truncates its result to 32 bits

从 builtin.js 的第 291 行开始，其中 int() 当前使用 JavaScript 的按位转换，并检查 development/automated_tests/transcrypt/div_issues/init.py 中的 Issue 911 testlet。让大整数检查通过，同时保留负浮点数向零截断的行为，然后使用指定的编译器选项在 Node 下运行该 testlet。

int (x) compiles to float (x) | 0. JavaScript's | converts its left operand with ToInt32, so the result is always a signed 32 bit integer. Arguments outside -2**31 ... 2**31 - 1 come back wrong, multiples of 2**32 come back as 0.

Reproducer

values = [8589934588, 9007199254740988, 2147483648, 4294967296]
for v in values:
    print (v, '->', int (v))

CPython:

8589934588 -> 8589934588
9007199254740988 -> 9007199254740988
2147483648 -> 2147483648
4294967296 -> 4294967296

Transcrypt:

8589934588 -> -4
9007199254740988 -> -4
2147483648 -> -2147483648
4294967296 -> 0

Run with 3.7.16 on Python 3.7, compiled with -b -n -e 6 and executed under
node. Line 291 of __builtin__.js is identical on current master.

The failure is quiet. A memory size in kB, a file offset or a timestamp in microseconds passes 2**31 during normal operation and then yields a small or negative number instead of an error.

Testlet

For development/automated_tests/transcrypt/div_issues/__init__.py:

    autoTester.check ('Issue 911')  # int () truncated to 32 bits
    autoTester.check (int (2147483648))
    autoTester.check (int (4294967296))
    autoTester.check (int (8589934588))
    autoTester.check (int (9007199254740988))
    autoTester.check (int (-5.7), int (5.7))

Possible fix

export function int (any) {
    var f = float (any);
    return f === Infinity || f === -Infinity || isNaN (f) ? 0 : Math.trunc (f);
};

This makes int () exact over the whole range where a JavaScript number is exact, up to 2**53 - 1, and leaves truncation toward zero for negative floats unchanged. With the patch applied the reproducer above matches CPython.

Math.trunc is ES6. The default target is esv 6 and javascriptVersion gates nothing in the code generator, so no version switch is needed. int has no callers inside the runtime.

Note on NaN and Infinity

NaN | 0 and Infinity | 0 currently yield 0, where CPython raises ValueError and OverflowError. The version above keeps the 0 to stay minimal. Raising instead would match CPython but changes behaviour beyond the range fix, so it seems worth deciding separately.