Renaming of dir using `move()` returns wrong path

[grr]

If you do $dir_source->move($dir_target) to rename a directory, according to the docs it should return the new path ($dir_target), but instead it returns $dir_target->child($dir_source->basename). It seems from the implementationm and is sort of implied by the docs, that the source should only be a file, but if the source is a directory, the rename does succeed.

[ap]

It seems from the implementation and is sort of implied by the docs, that the source should only be a file

The documentation explicitly talks about a case where the source is a file (“If the destination already exists and is a directory, and the source is not a directory”), which implies to me that the source may just as well not be a file.

The implementation however simply ignores that case. Intentional or an oversight? Could be either.

Probably the clearest indication is that the tests only test moving files. The patch (72d498d) was clearly written without any thought given to moving directories.

Maybe @atoomic just assumed that moving directories would be no different than moving files.

I’ve done a quick test for how File::Copy::move behaves when the source is a directory and discovered some behavior I would not have expected. Assuming a directory tree ./a/b and another tree ./c/d for each case,

1. move('c','a/b/x') succeeds and produces ./a/b/x/d – as expected; the trivial case
2. move('c','a/b/x/y') fails (“No such file or directory”) – as expected
3. move('c','a') fails (“Is a directory”) – as expected
4. move('c','a/b') succeeds and produces ./a/b/d – definitely not what I expected

There is however a certain logic to it:

1. The target directory basename is never inferred from the basename of the source directory, it is always the verbatim target directory basename.
2. The target directory is allowed to already exist – but only as long as it is empty, as shown by cases 3 and 4.

Aspect 1 is in sharp contrast to the rules that apply to files and aspect 2 seems maybe not entirely crazy but certainly questionable. And the File::Copy docs are basically silent about all of this.

The question is how much of that behavior Path::Tiny wants to own as its own documented behavior. If any.