Feature #3346: __DIR__ revisted

• Status: Closed
• Assignee: yhara (Yutaka HARA)
• Target version: 2.0.0

[trans (Thomas Sawyer)]

=begin
I'd like to know why DIR was rejected? I use File.dirname(FILE) all the time, and I frequently see others do so as well. #relative_require is helpful but it covers only one specific use case --and probably not the most common one. I am often using File.dirname(FILE) in build scripts, when I am loading examples for tests, and when I load output templates or other pluggable modules that reside relative to my code. For something so common, having to clutter my code with a 22 character sequence, when an perfectly obvious 7 character sequence would do semms very uncharacteristic of Ruby, which is usually quite concise. Indeed, it is not uncommon to see code that defines a constant such as DIR = File.dirname(FILE) when it will be used more than once because it quickly becomes an eye-sore. For these reasons I hope you will reconsider the earlier rejection.
=end

[naruse (Yui NARUSE)]

=begin
I like DIR.
=end

[shyouhei (Shyouhei Urabe)]

=begin
Which is "the earlier rejection"? I see nothing (though this request is not accepted yet).
=end

[usa (Usaku NAKAMURA)]

=begin
Hello,

In message "[ruby-core:30436] [Feature #3346] DIR revisted"
on May.26,2010 20:59:48, [e-mail] wrote:

I like DIR.

I like Kernel#dir.

Regards,

U.Nakamura [e-mail]

=end

[rkh (Konstantin Haase)]

=begin

I like Kernel#dir.

But then there should be Kernel#file.

Konstantin
=end

[Eregon (Benoit Daloze)]

=begin
Hi,
On 26 May 2010 14:17, U.Nakamura [e-mail] wrote:

Hello,

In message "[ruby-core:30436] [Feature #3346] DIR revisted"
   on May.26,2010 20:59:48, [e-mail] wrote:

I like DIR.

I like Kernel#dir.

Regards,

U.Nakamura [e-mail]

I prefer DIR.

( as a keyword,
because it's clearly the same family with FILE,
and it's a constant (so in uppercase) )

My first use would be things like Dir[File.join(DIR,'.rb')].each
{ |f| require f }
That's only because I am sometimes too lazy to "require_relative" all
the files under the same dir.
But maybe that's not the best way to do it, tough.
Currently, there is also another possibility, which is kind of a hack:
Dir[File.expand_path("../.rb",FILE)].each { |f| require f }

Just some ideas in the air for this case:

• require_relatives
• require_relative '*.rb'
• require_relative '*' (then assume we want only .rb files)
• require_relative (then the meaning would be "require the relative
  stuff, what is relative to this")
  The 2nd and 3rd one obviously more flexible as they accept a glob
  matching pattern.

Anyway, there are still many cases in which DIR would be useful,
even if that's kind of low-level stuff.

Regards,
B.D.

=end

[rogerdpack (Roger Pack)]

=begin

Which is "the earlier rejection"? I see nothing (though this request is not accepted yet).

#1961 wasn't rejected, just not accepted/committed yet.

For me, the "ideal" would be a compiler time constant FILE, but if it's a method instead, then name it file just to not confuse people.

-rp
=end

[matz (Yukihiro Matsumoto)]

=begin
Hi,

In message "Re: [ruby-core:30434] [Feature #3346] DIR revisted"
on Wed, 26 May 2010 20:45:55 +0900, Thomas Sawyer [e-mail] writes:

|I'd like to know why DIR was rejected?

I haven't rejected. I agreed with the basic idea, and proposed to add
Kernel#dir; a method not to add a new keyword, a lowercase name to
clarify it's implemented by a method. But some didn't agree with me,
so I haven't added.

						matz.

=end

[zenspider (Ryan Davis)]

=begin

On May 26, 2010, at 09:11 , Roger Pack wrote:

#1961 wasn't rejected, just not accepted/committed yet.

wow... it is just like Seattle politics... if we just keep voting OVER and OVER on the same issue, eventually it'll get accepted!

You should run for Seattle City Council.

Don't forget DIR_DIR and DIR_DIR_DIR... I'd use them lots.

=end

[trans (Thomas Sawyer)]

=begin
Good to hear. dir works for me if that is deemed better, though I concur that it inclines me to expect FILE to be file too. Also, I do not recall who suggested it, but an optional join parameter was once suggested too, e.g.

dir('foo')

Which would be equivalent to:

File.expand_path(File.join(File.dirname(FILE),'foo'))

Such a parameter would seem more appropriate to a lowercase method name. Also perhaps it would quell the "DIR_DIR" gripes, which I don't earnestly get, but if they should be taken seriously...

dir('..')

=end

[mame (Yusuke Endoh)]

• Description updated (diff)
• Status changed from Open to Assigned
• Assignee set to matz (Yukihiro Matsumoto)

[matz (Yukihiro Matsumoto)]

UNIX style directory hierarchy prepare a directory to put everything relate to the app/lib. I hesitate to add a method/directive that encourage file system convention apparently against UNIX style.

[trans (Thomas Sawyer)]

Ruby does not provide way to get root of app/lib, and current working directory is not always root of applicable app/lib, so that's not always helpful way.

Sometimes practical choice outweighs theoretical ideal. File.dirname(__FILE__) is exhaustively used idiom.

Consider another use case:

myapp/
lib/
myapp/
template.rb
templates/
foo.erb
bar.erb

In template.rb there will be something like:

def read_template(name)
file = File.dirname(FILE) + "/templates/#{name}.rb"
raise "no such template -- #{name}" unless File.exist?(file)
File.read(file)
end

[yhara (Yutaka HARA)]

• File 3346.pdf 3346.pdf added

Adding presentation slide for the feature request meeting ([ruby-dev:45708])

[mame (Yusuke Endoh)]

Yhara-san, your slide is received. Thank you!

But, matz has already accepted this proposal basically.
The remaining issue is its API (including name). Right?
If so, I'm afraid if just listing candidates will make no progress.
It may be good to choose one target yourself and make it authorized by matz.

Anyway, your slide will be presented at the meeting.
Matz may choose one in his preference. Thanks again!

--
Yusuke Endoh [e-mail]

[naruse (Yui NARUSE)]

I was against the name DIR for years,
but now I'm for dir if file and line method is also added.

[rogerdpack (Roger Pack)]

I'm for dir or DIR since I use File.dirname(FILE) a lot. Barely prefer DIR since FILE already exists so it keeps parity with it.

[mame (Yusuke Endoh)]

• Assignee changed from matz (Yukihiro Matsumoto) to nobu (Nobuyoshi Nakada)

Thomas Sawyer and Yutaka Hara,

Matz determined Kernel#dir (as a method).

Yhara-san, do you have a patch?

--
Yusuke Endoh [e-mail]

[matz (Yukihiro Matsumoto)]

• Status changed from Assigned to Closed

[shyouhei (Shyouhei Urabe)]

• Category set to core
• Status changed from Closed to Assigned
• Assignee changed from nobu (Nobuyoshi Nakada) to yhara (Yutaka HARA)
• Target version set to 2.0.0

This one is alive.

[Eregon (Benoit Daloze)]

mame (Yusuke Endoh) wrote:

Thomas Sawyer and Yutaka Hara,

Matz determined Kernel#dir (as a method).

Yhara-san, do you have a patch?

Will dir always be an absolute path, or just File.dirname(FILE)?

File.dirname(FILE) is relative (often ".") when FILE is $0.
It would mean the path needs to be expanded if it is relative (using File.expand_path), to keep the right location if the current working directory changes.

[rogerdpack (Roger Pack)]

definitely recommend full path to avoid possible confusion, like we used to have with 1.8.x not using the full path for $: but that's just my opinion...

[yhara (Yutaka HARA)]

mame (Yusuke Endoh) wrote:

Thomas Sawyer and Yutaka Hara,

Matz determined Kernel#dir (as a method).

Yhara-san, do you have a patch?

Sorry, I forgot to reply. I don't have a patch right now.

[authorNari (Narihiro Nakamura)]

• File __dir__.patch __dir__.patch added

Hi.

yhara (Yutaka HARA) wrote:

mame (Yusuke Endoh) wrote:

Thomas Sawyer and Yutaka Hara,

Matz determined Kernel#dir (as a method).

Yhara-san, do you have a patch?

Sorry, I forgot to reply. I don't have a patch right now.

I've created a patch for Kernel#dir. How about it?

[authorNari (Narihiro Nakamura)]

• Status changed from Assigned to Closed
• % Done changed from 0 to 100

This issue was solved with changeset r37432.
Thomas, thank you for reporting this issue.
Your contribution to Ruby is greatly appreciated.
May Ruby be with you.

---

• eval.c (f_current_dirname): add the new method for Kernel.
  This method almotst same as File.dirname(FILE). One
  different behavior is it returns nil when FILE returns nil.
  [Feature #3346]

• NEWS: ditto

• test/ruby/test_method.rb: related test.