[phobos] RFC: std.path

[Lars Tandle Kyllingstad]

On Sun, 2011-06-12 at 15:35 -0300, Jose Armando Garcia wrote:
> On Sun, Jun 12, 2011 at 3:26 PM, Jose Armando Garcia <jsancio at gmail.com> wrote:
> > Em 12/06/2011, às 14:00, Lars Tandle Kyllingstad <lars at kyllingen.net>
> > escreveu:
> >
> >> On Sun, 2011-06-12 at 13:36 -0300, Jose Armando Garcia wrote:
> >>>
> >>> On Sun, Jun 12, 2011 at 12:59 PM, Lars Tandle Kyllingstad
> >>> <lars at kyllingen.net> wrote:
> >>>>
> >>>> On Sun, 2011-06-12 at 12:41 -0300, Jose Armando Garcia wrote:
> >>>>>
> >>>>> On Sun, Jun 12, 2011 at 12:15 PM, Lars Tandle Kyllingstad
> >>>>> <lars at kyllingen.net> wrote:
> >>>>>>
> >>>>>> On Sun, 2011-06-12 at 11:39 -0300, Jose Armando Garcia wrote:
> >>>>>>>
> >>>>>>> On Wed, Jun 8, 2011 at 4:29 PM, Lars Tandle Kyllingstad
> >>>>>>
> >>>>>> These functions are from the old std.path, and I haven't made any
> >>>>>> changes to them in my version.
> >>>>>>
> >>>>>> - toAbsolute()
> >>>>>> - toCanonical()
> >>>>>>
> >>>>>
> >>>>> In the comments where you say that it doesn't perform any IO you
> >>>>> should add these functions.
> >>>>
> >>>> Does getcwd() perform any IO on Windows?  AFAIK, on POSIX it just
> >>>> queries /proc/self/cwd, which is a virtual file.
> >>>>
> >>>
> >>> The way I look at IO is anything that is external to the process.
> >>> Another way to thinking about it is that
> >>> toAbsolute()'s and toCanonical()'s result is dependent on state
> >>> outside of the process. While the rest of the templates/functions
> >>> aren't.
> >>
> >> The way I've interpreted the "no IO" principle of std.path is "no
> >> disk/network IO", since those would come with an enormous performance
> >> penalty as compared to in-memory operations.  But maybe you are right.
> >>
> >>
> >>>>> Speaking of which can we add a template
> >>>>> called normalize (maybe you can come up with a better name) that does
> >>>>> what canonical does but doesn't make it absolute. E.g.:
> >>>>>
> >>>>> version(windows) assert(normilize("dir/file") == "dir\\file");
> >>>>> version(windows) assert(normilize("dir/./file") == "dir\\file");
> >>>>> //etc
> >>>>
> >>>> That sounds like a good idea.  Then I guess normalize("../foo") should
> >>>> just return "..\\foo", i.e. leave the ".." unresolved?
> >>>>
> >>>
> >>> It is hard to resolve '..' without looking at the file system when
> >>> considering soft/sym link due to multiple parents. if 'somedir' is a
> >>> simlink "somedir/../" != ".".
> >>
> >> That is a matter of choice, I guess.  In both bash and zsh, if I type
> >>
> >>  cd some_dir/some_symlink/..
> >>
> >> I end up in some_dir, regardless of where some_symlink is pointing.
> >> That is how toCanonical() does things as well, and how I think
> >> normalize() should work if I end up adding that.
> >
> > But most program dont behave this way. For example ls, less and vim don't do
> > that. I am okay with resolving symlinks but just take note.
> 
> Err. I am okay with resolving "..".

I have thought some more about this, and I think I will simply remove
toCanonical() and replace it with normalize().  After all,

    auto canonicalPath = normalize(toAbsolute(path));

What do you think?

-Lars