[libc][Obvious] Fix a mismatch signature of HighPrecisionDecimal::should_round_up.

[lntue]

Summary

Its input should be int32_t instead of uint32_t.

[michaelrj]

This is incorrect, should_round_up is called on decimal_point which is specifically signed because it represents the index of where the decimal point should be, relative to the digits. Specifically a negative decimal point represents a number that is less than 1. There is an if statement on line 115 to handle if roundToDigit is negative. While the code will still work if this change is made (any reasonable negative number will create an unsigned number greater than this->num_digits) it would make the code less correct.

[sivachandra]

In D118791#3295095, @michaelrj wrote:

This is incorrect, should_round_up is called on decimal_point which is specifically signed because it represents the index of where the decimal point should be, relative to the digits. Specifically a negative decimal point represents a number that is less than 1. There is an if statement on line 115 to handle if roundToDigit is negative. While the code will still work if this change is made (any reasonable negative number will create an unsigned number greater than this->num_digits) it would make the code less correct.

Isn't your reason exactly why this change is required? For, if roundToDigit was unsigned, roundToDigit < 0 will never be true?

[michaelrj]

My reason is that roundToDigit can be negative, and changing it to be unsigned would break that. The only line where should_round_up is called is line 367, and it's passed decimal_point, which is a signed integer which may sometimes be negative.

[sivachandra]

In D118791#3295156, @michaelrj wrote:

My reason is that roundToDigit can be negative, and changing it to be unsigned would break that. The only line where should_round_up is called is line 367, and it's passed decimal_point, which is a signed integer which may sometimes be negative.

As I read it, this change is making roudnToDigit to be of a signed type instead of an unsigned type. It is switching the type of roundToDigit from uint32_t (unsigned 32-bit integer) to int32_t (signed 32-bit integer).

[michaelrj]

oops, you're right, I misread the patch. Sorry about that, LGTM.

libc/src/__support/high_precision_decimal.h

class HighPrecisionDecimal {
   static constexpr uint32_t MAX_NUM_DIGITS = 800;
 
   uint32_t num_digits = 0;
   int32_t decimal_point = 0;
   bool truncated = false;
   uint8_t digits[MAX_NUM_DIGITS];
 
 private:
-  bool should_round_up(uint32_t roundToDigit) {
+  bool should_round_up(int32_t roundToDigit) {
     if (roundToDigit < 0 || roundToDigit >= this->num_digits) {
       return false;
     }
 
     // If we're right in the middle and there are no extra digits
     if (this->digits[roundToDigit] == 5 &&
         roundToDigit + 1 == this->num_digits) {