guard non-positive requested digits in DoubleToAscii and BignumDtoa

[jdymitarai]

Problem

Following the hardening of DigitGenCounted (#313), GenerateCountedDigits (#314), and FastFixedDtoa (#315) against non-positive/negative counts, entry points and bignum utilities still had edge cases in release builds (-DNDEBUG):

1. DoubleToStringConverter::DoubleToAscii:

   • DoubleToAscii checked if (mode == PRECISION && requested_digits == 0) to return an empty representation (*length = 0, *point = 0).
   • When called with requested_digits < 0 in release builds (where DOUBLE_CONVERSION_ASSERT is compiled out), this check failed. FastDtoa returned false (guard non-positive requested_digits in DigitGenCounted #313), and execution proceeded to BignumDtoa, setting *length = 0 (guard non-positive count in GenerateCountedDigits #314) but leaving *point at an unreset, non-zero value, after doing redundant bignum work.
   • For mode == FIXED with requested_digits < 0, DoubleToAscii fell through to BignumDtoa, where BignumToFixed could emit truncated integer digits into buffer when (*decimal_point) + requested_digits > 0.

2. BignumDtoa:

   • For mode == BIGNUM_DTOA_PRECISION && requested_digits <= 0 or mode == BIGNUM_DTOA_FIXED && requested_digits < 0, BignumDtoa lacked early return guards.

3. Bignum Shift and Exponent Operations:

   • Bignum::ShiftLeft(shift_amount): if shift_amount <= 0, non-positive shifts are a no-op, but previously called BigitsShiftLeft(local_shift) which triggered undefined behavior on negative shifts.
   • Bignum::MultiplyByPowerOfTen(exponent): checked if (exponent == 0) return;. For exponent < 0, it bypassed the multiplier loops and called ShiftLeft(exponent) with a negative shift amount.
   • Bignum::AssignPowerUInt16(base, power_exponent): checked if (power_exponent == 0) return;. Negative exponents bypassed the loops and called ShiftLeft(shifts * power_exponent).

Solution

1. In DoubleToStringConverter::DoubleToAscii, guard (mode == PRECISION && requested_digits <= 0) || (mode == FIXED && requested_digits < 0) by returning early with an empty string, *length = 0, and *point = 0.
2. In BignumDtoa, return early with an empty string, *length = 0, and *decimal_point = 0 for (mode == BIGNUM_DTOA_PRECISION && requested_digits <= 0) || (mode == BIGNUM_DTOA_FIXED && requested_digits < 0).
3. In Bignum::ShiftLeft, return early if shift_amount <= 0.
4. In Bignum::MultiplyByPowerOfTen and Bignum::AssignPowerUInt16, change checks to <= 0.
5. Add regression test cases across test/cctest/test-dtoa.cc, test/cctest/test-bignum-dtoa.cc, and test/cctest/test-bignum.cc.

[floitsch]

LGTM.