On Mon, 01 Nov 2021 21:04:54 -0500, Scott Cheloha wrote:
> Yes it would be simpler. However I didn't want to start changing the
> input -- which we currently don't do -- without discussing it.
>
> The standard says we should "write one copy of each input line on the
> output." So, if we are being strict, we don't add a newline that isn't
> there, because that isn't what we read. Any other interpretation
> requires handwaving about what an "input line" even is.
The System V version of uniq actually ignores the last line if it
doesn't end in a newline. For example:
$ printf "bar\nfoo\nfool" | uniq
bar
foo
AIX, Solaris, and HP-UX still exhibit this behavior. What happens
is that the gline() function (which reads the line) returns non-zero
when it hits EOF, discarding any input in that line. Interestingly,
it does realloc the line buffer as needed to handle long lines.
So really, this is a corner case where you can't count on consistent
behavior among implementations and we just need to do what we think
is best. If you prefer we retain the existing behavior wrt a final
line without a newline that is OK with me.
- todd
