Paul Eggert wrote:
> PƔdraig Brady <address@hidden> writes:
>
>> OK, how about the attached patch?
>
> Better, but it still has problems. For example, on my platform
> (Debian stable with GCC 4.2.0, x86) the command
> "seq 0.0 0.1 0.90000000000000000000" outputs something different from
> "seq 0.0 0.1 0.9". The former stops at 0.8, the latter at 0.9.
You're stretching :)
I'll see if I can fix that,
but I do think it's much better that what we have currently.
> In general, if we want seq's numeric behavior to be independent of the
> format of the operand numbers, I think we have to take a different
> approach.
>
> Here's an idea. Use sprintf to convert the numbers to a textual
> format and back, and then use the result of _that_ comparison instead
> of the internal floating-point comparison. For speed, do this only
> for the number "one past" the number that "seq" would ordinarily print
> now. Also, don't bother to print this extra number if it's textually
> identical to the previously printed number.
>
> Whaddya think?
I'll have another look.
PƔdraig.