On Mon, Nov 26, 2018 at 6:26 PM Trey Harris <t...@lopsa.org> wrote:
>
>
>
> On Mon, Nov 26, 2018 at 04:56 Fernando Santagata <nando.santag...@gmail.com> 
> wrote:
>>
>> On Sun, Nov 25, 2018 at 5:34 PM Brad Gilbert <b2gi...@gmail.com> wrote:
>>>
>>> The reason `dir('a', test => { .IO.d })` doesn't work like you expect
>>> is that it is passed strings.
>>
>>
>> Thank you!
>> I could have inferred that from the declaration of the 'test' argument here 
>> https://docs.perl6.org/routine/dir where the test is being done against '.' 
>> and '..', two strings:
>>
>>> sub    dir(Cool $path = '.', Mu :$test = none('.', '..'))
>>> method dir(IO::Path:D: Mu :$test = none('.', '..'))
>>
>>
>> but I guess it's a long stretch.
>> Maybe a note in the docs would be useful to others.
>>
>>>
>>> I would argue that inside of the `test` that `$*CWD` should be set to
>>> `a`, and/or the values should be IO objects.
>>> (They are going to become IO objects eventually anyway.)
>>
>>
>> Indeed I think that even tough one can do any kind of test on the 
>> directory's content, the most usual ones are those using the file test 
>> operators, so defaulting to IO is probably more useful than defaulting to 
>> Str.
>
>
> I'm not sure that's true. I think it's a question of following the Principle 
> of Least Surprise.

Exactly, the way it works right now is surprising (to new Perl 6 programmers).

It even took me a few seconds to realize what the problem was, and
I've been learning Perl 6 for about a decade.

> The only substantive difference between `my $i = "foobar".IO` and `my $s = 
> "foobar"` besides needing to stick the `.IO` in when you want to do IO-y 
> things with `$s` is that `$i` retains the knowledge of $*CWD at the time of 
> instantiation. But `dir` always returns paths that stringify _relative to 
> `$*CWD` at the time of the `dir` call._ Notice the difference between 
> `$i.gist` (which, if evaluated, will not consistently re-create the original 
> object) and `$i.perl` (which will):
>
>     > $f.gist
>     "foobar".IO
>     > $i.perl
>     IO::Path.new("foobar", :SPEC(IO::Spec::Unix), 
> :CWD("/home/trey/perl6/rakudo-star"))
>     > chdir "/tmp"
>     "/tmp".IO
>     > $f.gist
>     "foobar".IO
>     > $i.perl
>     IO::Path.new("foobar", :SPEC(IO::Spec::Unix), 
> :CWD("/home/trey/perl6/rakudo-star"))
>
> The string, OTOH, is also relative to the `$*CWD` at the time of the `dir()` 
> call, but if you call `.IO` on it after changing directories, the path object 
> is interpreted relative to the new directory. This is unsurprising; what 
> _would be_ surprising is if a string changed as a result of external program 
> state.

I was only talking about the object within the `:test` argument of the
`dir` call.

The resulting sequence wouldn't change at all.

> Large-scale file-munging operations have always been squarely in Perl's 
> wheelhouse, and doing fancy operations with complex subtrees (such as 
> syncing, normalizing, archiving, collating, etc.) are common.

Precisely why I suggested changing it.
The way it works now is harder to use than it could be.

> The string doesn't offer as much in the way of features, but I think it 
> behaves much more predictably when doing these sorts of operations. If you 
> `chdir $dest` after the `dir` but before doing your munging, then you must 
> remember to prepend the old working directory if you need to access the 
> original file (i.e., `$src = $*CWD;  for @files -> $f { $src.add($f) ... }`), 
> while the new files would just be `$f`. If you don't change directories, then 
> you must say `$dest.add($f)` to refer to the new files, while the old file is 
> just `$f`. But in either case, it's unsurprising and predictable.

Actually Str behaves less predictably than an IO object for the `test` argument.
If you want to compare it stringwise, do so and it works:

    my $dir = IO::Path( 'b', :CWD('a'.IO) );

    say $dir eq 'b'; # True

If you want to see if it is a directory, just ask it:

    say $dir.d; # True

`.gist` doesn't stringify an object, it returns enough information for
a human to be able to figure out what the object is.

When you do stringy operations on an IO::Path object (or any object),
it calls the `.Str` method not `.gist`.
The only things which implicitly call `.gist` is `say` and `note`.

Now what I proposed was having the CWD set to the directory being
examined inside of `test`, which is a different `CWD` than the result
of the `dir` call.

    > say dir('a'):test( {temp $*CWD = 'a'.IO; .IO.d && .IO ~~ none(<. ..>)} )
    ("a/b".IO)
    > say dir('a', :test( {temp $*CWD = 'a'.IO; .IO.d && .IO ~~
none(<. ..>)} ))».CWD
    (/home/brad)

Note that `none(<. ..>)` still worked just fine against such an
object. (the default `:test` argument)

I also didn't say they had to be IO objects, just that the CWD should
be inside of the directory of the dir being tested within the `:test`
argument.

>
> If, OTOH, these were IO objects, if you changed directories, you'd think from 
> the gist that you'd do the same. And if you don't change directories, you 
> _could_ do the same (`$dest = "/destdir".IO; $dest.add($f)` works using the 
> gist). But if you do change directories, it's now the _old file_ that's just 
> `$f`; the new file becomes `$*CWD.add($f)`. This feels quite strange and 
> surprising to me.

The result of `dir()` is already a sequence of IO::Path objects with a
predetermined CWD.

    > say $*CWD
    "/home/brad".IO
    > my $dir = dir('a').first( *.basename eq 'b' );
    > say $dir.CWD;
    /home/brad
    > indir 'a'.IO, { say $*CWD; say $dir.CWD }
    "/home/brad/a".IO
    /home/brad           # notice that it didn't change

I was only talking about changing what gets passed to the `:test`
argument of the `dir` call.

> The very slight keystroke-saving convenience of not requiring the outputs of 
> `dir` to be followed with `.IO` before doing IO operations on individual IO 
> objects in-place doesn't seem worth the strangeness of their behavior when 
> doing bulk operations on them from a distance.

That isn't what we are talking about.
We are talking about the `:test` argument being harder to write and
understand than it should be.
Right now you have to append the directory you are looking at before calling .IO
By either setting $*CWD, or creating an IO object with :CWD set; it
makes such code easier to understand.

The behaviour wouldn't change for Str based comparisons, so I don't
understand how this relates to my solution.

> (I was thinking through possible have-your-cake-and-eat-it-too solutions like 
> `.gist` and `.path` of IO objects created from strings changing their string 
> representation if and only if you changed working directories and these all 
> seem to just Waterbed Complexify the problem elsewhere. Perhaps someone can 
> come up with an everybody-wins solution—but except for moving the filetest 
> operations into `Cool` so they work like in Perl 5, a ship that's already 
> sailed, I can't think of one.)

I don't understand what you are talking about here.

Reply via email to