On 04/19/2010 03:51 PM, pacman@kosh.dhis.org wrote:
> Eric Blake writes:
>>
>> First, you have to prove that sort is not behaving according to a
>> particular standards document.  And since POSIX 1992 _did_ document
>> 'sort +1', that means quoting a relevant portion of that document along
>> with a simple test case demonstrating why you think coreutils does not
>> comply with that document.  If you can prove that, then we will gladly
>> fix the bug.
> 
> So your plan is to ignore the many easily available documents that explain
> exactly what sort +1 -2 means, in favor of a document that you don't have,
> and then blow off bug reports by sending people on a wild goose chase.

I am not blowing off this bug report - I agree with you that we need
better documentation.  However, I don't have the resources to do all the
legwork myself (in particular, access to the POSIX 1992 standard would
be a big help), so in open source fashion, I'm asking for help.  And
since it seems to be your itch, the best you can do is scratch it by
providing that help, rather than criticizing the fact that we are trying
to spread the work load.

> 
> Please just take a current BSD or Solaris man page as definitive.

It may be definitive for Solaris, but it is not definitive for POSIX,
and it does not good to copy someone else's documentation if the
implementation is subtly different.  That's why I'm insisting that we go
to the sources - somewhere where the behavior is standardized, then
contrast that against our implementation.  At that point, if the two
agree (which I hope they do), then we are done; while if the two
disagree, then it is much easier to justify a change to coreutils to
come in line with the standards.

> Compatibility with other actually-existing implementations is what matters
> here, not some former standard.

The POSIX standard was written with the goal of being compatible with
existing implementations; it did not always achieve that goal, but the
hope is that it did in the case of sort.  However, these days, it is
better to comply to POSIX than it is to be bugwards compatible to subtle
bugs in older implementations.

-- 
Eric Blake   eblake@redhat.com    +1-801-349-2682
Libvirt virtualization library http://libvirt.org