From: Brian Ford <brixen@...>
Date: 2009-02-15T05:41:22+09:00
Subject: [ruby-core:22108] Re: Dir.glob and duplicates?

On Feb 14, 10:47�am, Charles Oliver Nutter <charles.nut...@sun.com>
wrote:
> Nobuyoshi Nakada wrote:
> > Hi,
>
> > At Sat, 14 Feb 2009 16:35:30 +0900,
> > Charles Oliver Nutter wrote in [ruby-core:22065]:
> >> The spec MRI produces dupes for is:
>
> >> Dir.glob recursively matches directories with '**/<characters>'
> >> at core/dir/shared/glob.rb:205
>
> > The pattern at line 204 is '**/*fil?{,.}*', which is equivalent
> > to to Dir.glob('**/*fil?*')+Dir.glob('**/*fil?.*'), so
> > "subdir_two/nondotfile.ext" matches against both. �That is,
> > brace expansions needs attention to use.
>
> I think the presence of duplicates is a glob implementation detail that
> should not be specified. Surely different implementations are going to
> produce difference results. I don't think we should specify when dupes
> are expected to occur, and glob's specification should either be "may
> produce dupes" or "will not produce dupes".

Saying that there are duplicate entries because it emulates shell
behaviors that are only present on _some_ platforms that Ruby runs on
is flimsy justification at best. For everyone's sanity, I would
strongly favor the "no dupes" over "may be dupes" or "undefined --
implementation specific".

I'd really like to see a use case for dupes in anything that processes
a list of filenames or the files themselves. Just to repeat, Ruby
should not be forcing emulation of a partial set of platforms unless
there is a compelling reason to do so.

Cheers,
Brian

>
> - Charlie