sort: scientific notation with sign is incorrectly parsed in general numeric sort

Component

sort

Description

GNU sort correctly parses scientific notation like "1e-5" (which equals 0.00001) in general numeric sort mode (-g). It uses the C standard library function strtold which handles scientific notation properly.

In GNU sort, strtold is used to parse general numeric values.

static int
general_numcompare (char const *sa, char const *sb)
{
  char *ea;
  char *eb;
  long double a = strtold (sa, &ea);
  long double b = strtold (sb, &eb);
  ...
}

However, in uutils sort, there is an off-by-one error in get_leading_gen when checking for a digit after the exponent sign.

if let Some(&(_, &next_char)) = char_indices.peek() {
    if (next_char == b'+' || next_char == b'-')
        && matches!(
            char_indices.peek_nth(2), // should be peek_nth(1)
            Some((_, c)) if c.is_ascii_digit()
        )
    {
        // Consume the sign. The following digits will be consumed by the main loop.
        char_indices.next();
        had_e_notation = true;
        continue;
    }
    if next_char.is_ascii_digit() {
        had_e_notation = true;
        continue;
    }
}

For "1e-5", when processing 'e', peek() (equivalent to peek_nth(0)) returns '-'. The code should use peek_nth(1) to check the next character '5', but it incorrectly uses peek_nth(2) which returns None. As a result, the exponent is not recognized and the number is parsed incorrectly.

Test / Reproduction Steps

# GNU
$ printf '1\n1e-5' | sort -g
1e-5
1

# uutils
$ printf '1\n1e-5' | coreutils sort -g
1
1e-5

Impact

Scientific notation with signed exponents is incorrectly parsed and causes wrong sort order for numeric data.

Recommendations

Change peek_nth(2) to peek_nth(1) to correctly check the digit after the exponent sign.