Re: chdir(..) dance for 5.10 on osx?

On Jul 9, 2008, at 00:29, Eric Wilhelm wrote:

> # from David E. Wheeler
> # on Thursday 19 June 2008 12:25:
>
>> For my install of File::Path on Perl 5.10, it
>> won't actually delete a directory if your current cwd is unknown.  
>> So I
>>   added appropriate calls to `chdir` when `Cwd::getcwd` returns
>> false. And now all tests pass for me.
>
> I'm not seeing how this is required (since -d $tmp should be false if
> you're in the directory) and what's with the '..'?  Is that supposed  
> to
> be $cwd?

It's not because cwd returns `undef`.

>   END {
>     if(-d $tmp) {
> -      File::Path::rmtree($tmp) or warn "cannot clean dir '$tmp'";
> +      # Go back to where you came from!
> +      chdir '..' or die "Couldn't chdir to $cwd";
> +      File::Path::rmtree($tmp) or diag "cannot clean dir '$tmp': $@";
>     }
>
> Again.  In what way is '..' correct?  DistGen requires you to chdir  
> back
> yourself before calling remove().  Were there tests failing on the
> original tree?

Yes, there were. Otherwise I wouldn't have done this. I changed the  
code to go back to the original directory where such was tracked, but  
in some places it was not.

> sub remove {
>   my $self = shift;
>   croak("invalid usage -- remove()") if(@_);
> +  unless (Cwd::getcwd) {
> +    # How'd we get here?
> +    chdir '..' or die "Couldn't chdir to ..";
> +  }
>
> I would rather resolve whatever is happening without chdir('..') -- if
> only to avoid writing an enormous comment to explain why something
> which looks so wrong exists.

I can give you access to my box to play with it yourself, if you don't  
have one handy.

Best,

David