Re: Support bytesize comparison in sort

[Mart Somermaa]

Paul Eggert wrote:
> Also, I'm still leery of assuming properly scaled input.
> Is it that much harder to implement it correctly without that assumption?
>   
Sorry, missed that bit in my previous mail... yes, as without that
assumption we have to use multiplication for normalization -- but then
we face the usual problems of value overflow and imprecise
floating-point comparison. That may lead to straightforward incorrect
behaviour.

But I really don't see what's wrong with that assumption. It holds for
other coreutils and that's what matters most. A clearly documented
limitation is not a bug, but a feature :) .