0-byte file is the result of copying a file to itself with `DVCFileSystem.get_file` with any file larger than COPY_PBAR_MIN_SIZE

[adamliter]

Bug report

If you use DVCFileSystem's get_file method to copy a file to itself, you'll get a file with size of 0 bytes if the file size greater than COPY_PBAR_MIN_SIZE. However, if the file size is less than COPY_PBAR_MIN_SIZE, you'll get the original file back.

You end up with a 0-byte file because of this code here.

Current behavior

$ cd /tmp
$ mkdir dvc-test
$ cd dvc-test
$ pdm init --python cpython@3.12
$ git init
$ dvc init
$ git add .
$ git commit -m "initial commit"
$ truncate -s 2G model.ckpt
$ dvc add model.ckpt
$ git add .
$ git commit -m "trained model"
$ ls -lh model.ckpt
-rw-r--r--@ 1 adam.liter  wheel   2.0G Dec  9 17:21 model.ckpt

Then from Python (e.g., pdm run python):

from dvc.api import DVCFileSystem
fs = DVCFileSystem()
fs.get_file("model.ckpt", "model.ckpt")

Now go back to a shell and check the file size:

$ ls -lh model.ckpt
-rw-r--r--@ 1 adam.liter  wheel     0B Dec  9 17:25 model.ckpt

Expected behavior

The behavior of dvc_objects.fs.utils.copyfile should be the same for all files, regardless of file size. In particular, if copying a file to itself when the file size is greater than COPY_PBAR_MIN_SIZE, the result should not be a 0-byte file.

[skshetry]

However, if the file size is less than COPY_PBAR_MIN_SIZE, you'll get the original file back.

If the file size is less than COPY_PBAR_MIN_SIZE (1GB), we use shutil.copyfile, which will raise shutil.SameFileError.

Ideally, we should also raise the same error here. But I think it'll be costly as it requires two stat calls (but so does the shutil.copyfile).

The above snippet is also questionable use, I am not sure how you will have two identical files in the get_file, unless it was intentional.

DVC operations are also unlikely to run into this case. So fixing this will always going to be a lower priority.