download deletes folder when using a folder name in parameter output

[franzfurbass]

The function download assumes that the output parameter (of type string) is a file. When given a folder name as output parameter the whole folder is deleted which is unexpected. e.g. download(url_to_file, ".") removes the current directory. Compared to copy the expected behavior is to put the file into the directory like in cp path_to_file/file.txt ..

The output parameter should be checked with isdir and in that case the a filename derived from the folder name should be used instead.

[StefanKarpinski]

I agree that downloading to . should definitely not delete the current directory. However, I'm not convinced that it's good for non-interactive programming usage that the behavior of download(url, "path") depends on the state of the file system. I think it should either save the content of url to "path" or fail. It might make sense to add a force::Bool=false option that controls whether the output path is overwritten or not. It might also make sense to have an into::Bool=false option that causes the download to be saved into the target rather than at the target. The error message when someone tries download(url, "dir") and it fails could suggest using into=true to save into the directory or force=true to replace the directory.

[StefanKarpinski]

Technically adding a force option defaulting to false is breaking, but I think we could do it in a minor release. It seems unlikely that people are relying on this deleting directories.

[ararslan]

That would break my workflow of using the shorter and clearer download(url, "dir") in place of rm("dir"; recursive=true).

[StefanKarpinski]

:trollface: