plans for 0.0.6 release

[Martin Pool]

On  3 Jun 2005, Rusty Russell <rusty at rustcorp.com.au> wrote:
> On Wed, 2005-06-01 at 10:39 -0500, Nicholas Nethercote wrote:
> > - I think the "to" syntax is horrid -- ambiguous and thus error-prone both
> >    to implement and use, and inconsistent with every other command line
> >    option in the universe.  "4 to 5" might look nice and Python-ish, but
> >    Python-style syntax doesn't mix well with shell syntax.
> > 
> > - I think using '-' as a range operator is bad.  '-' is already used
> >    to introduce options (ie. -r), and also in some revision names, so why
> >    overload it with another meaning?  It just makes command lines harder to
> >    parse (for humans and machines).  I mentally parse "-r4-5" as "-r4 -5".

A worse case is '-r-4' vs '-r -4'.

> We went down the ":" path in netfilter for port ranges, and I've
> regretted it ever since.

It seems it's particularly confusing there because colon is already
mentally associated with HOST:PORT syntax.  Do you think that's the main effect?

> It is the most common mistake I make in
> writing rules.

What happens?  You write 20-21 when you should say 20:21?

> There is no parsing ambiguity when combined with the "<namespace>:"
> syntax, which I think everyone quite likes.
> 
> > - I quite like ':' as a range operator, it works fine in Subversion.  But
> >    if ':' has another use in revision names, then avoiding overloading
> >    it is a good idea.
> 
> : has merit because it is used in CVS and svn.  CVS also uses :: to
> indicate a collapsed range in some cases.  Two -r options are used in
> RCS.  

Two -r options would nicely avoid the debate, and they've been
proposed in some patches.  I think the main drawback is they can't
distinguish "20-" and "-20" and "20".

> > - This leaves '..' which seems best:  it's intuitive, easy to read, easy
> >    to implement and doesn't have any prior meaning, AFAIK.  And I don't
> >    recall anyone else complaining about it.
> 
> Unfortunately, this is neither intuitive nor established.  My preference
> would be to have a canonical form of "-r A-B" or "-r A -r B" but also
> allow svn-style : ranges for simple numeric ranges (which otherwise
> would be a clear syntax error and just irritate svn users).

On the whole I think '-r A-B' is right, perhaps with ':' supported
where it doesn't clash with the prefix form for dates and so on.

-- 
Martin