seq -w: positive exponents in scientific-notation operands produce wrong padding width (leading zero not cancelled)

[Poojax21]

Summary

compute_num_digits in src/uu/seq/src/numberparse.rs counts integral digits of an operand, then adds a positive decimal exponent to that count. When the mantissa has a leading 0 (the .X / -0.X special case), shifting the decimal point past that zero annihilates it, but the code still keeps the counted digit and adds the exponent on top. The inflated integral-digit count makes seq -w pad output one character wider than the widest rendered value.

Static-analysis finding based on reading main; not executed here.

Location

• File: src/uu/seq/src/numberparse.rs
• Function: compute_num_digits
• Relevant code:

let (mut int_digits, mut frac_digits) = match parts[0].find('.') {
    Some(i) => {
        // Cover special case .X and -.X where we behave as if there was a leading 0:
        // 0.X, -0.X.
        let int_digits = match i {
            0 => 1,
            1 if parts[0].starts_with('-') => 2,
            _ => i,
        };
        (int_digits, parts[0].len() - i - 1)
    }
    None => (parts[0].len(), 0),
};
...
if exp > 0 {
    int_digits += exp.try_into().unwrap_or(0);
}

Problem

For the operand "0.5e1" (value 5):

1. find('.') -> i = 1; the special case yields int_digits = 1 (the counted leading 0) and frac_digits = 1.
2. exp = 1 > 0 -> int_digits += 1 -> 2.
3. But 0.5e1 == 5, whose rendered form "5" has exactly 1 integral digit: shifting the point past the leading zero removes that zero (05e1 would be 5, not 05). The correct count is therefore max(1, i + exp - 1)-style logic that first cancels the pad digit; instead both are summed.
4. Downstream, seq -w 0.5e1 0.5e1 derives padding = max(integral digits) = 2 and prints 05.

The file's own tests encode the same over-count (e.g. -0.1e2 expected to contribute width 4, while -10 renders as 3 characters).

Trigger / Reproduction

Based on source reading:

$ seq -w 0.5e1 0.5e1
05          # actual
5           # expected

Same class: .5e2, -0.1e2, any operand whose mantissa starts with 0/. and whose exponent is at least the number of fractional digits.

Expected Behavior

The equal-width padding should equal the width of the widest actually-rendered value among all sequence elements - here 5, so no padding.

Actual Behavior

Padding is computed from an inflated digit count that double-counts the special-cased leading zero once the exponent moves past it, producing spurious leading zeros / overwide fields.

Impact

Wrong textual output for documented input syntax (scientific notation is accepted by seq's parser). Cosmetic rather than data-destructive, but deterministic and visible in scripts relying on fixed-width columns.

Suggested Direction

When i corresponds to the leading-zero special case, subtract the consumed pad digit from int_digits before applying a positive exponent (or compute the integer-part width from the normalized value mantissa * 10^exp directly).

Evidence

• Lines 60-66 explicitly count the synthesized leading zero as an integral digit.
• Lines 77-78 add the exponent unconditionally for exp > 0; there is no interaction between the two adjustments anywhere in the function.

[Devel08]

GNU seq outputs the same, maybe you'd want to report it there?

$ seq -w 0.5e1 0.5e1
05
$ seq --version
seq (GNU coreutils) 9.11
Copyright (C) 2026 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <https://gnu.org/licenses/gpl.html>.
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.

Written by Ulrich Drepper.