424473 - Erroneous loss of precision warning on (15).toFixed(-1) (debug only)

Status: RESOLVED INCOMPLETE

(Core :: JavaScript Engine, defect)

Description

[x0]

js> (15).toFixed(-1)
WARNING: A loss of precision for double floating point is detected.
         The result of any operation on doubles can be meaningless.
         A possible cause is missing code to restore FPU state, see
         bug 360282 for details.
20

In js_dtoa(), where the warning would be canceled if d > 0.5 + eps, d is exactly 0.5 in this case, and control flows through the fast_failed label.

Compare also bug 362641.

Comment 1

[Ed Lee :Mardak]

I haven't looked carefully, but is it a < vs <= issue?

- if (1. - d < eps) {
+ if (1. - d <= eps) { /* Clear d to avoid precision warning. */

- if (d > 0.5 + eps) {
+ if (d >= 0.5 + eps) {

Comment 2

[x0]

(In reply to comment #1)
> I haven't looked carefully, but is it a < vs <= issue?

No, eps is non-null and d is exactly 0.5, so that won't change anything.

At the warning, d is 10.

Comment 3

[Mathias Bynens]

This is still happening. Example:

> (123123).toFixed(-3)
123000

Comment 4

[Mathias Bynens]

This should throw a RangeError as per the spec: http://es5.github.com/x15.7.html#x15

I’m not sure what the “(debug only)” in the title is supposed to mean, but this still happens in the latest Firefox Nightly.

Comment 5

[Tom S. (please needinfo tschuster)]

This bug had nothing to do with your error. And per spec we are allowed to have this behavior:

>An implementation is permitted to extend the behaviour of toFixed for values of fractionDigits >less than 0 or greater than 20. In this case toFixed would not necessarily throw RangeError >for such values.

But I agree, I don't see how values <0 are useful, Chrome also clips to 0 =< x =< 20.

Comment 6

[André Bargull [:anba]]

No longer reproducible because we're now always throwing a ReferenceError, therefore closing as INCOMPLETE.