COGONI Guillaume <cogoni.guillaume@xxxxxxxxx> writes:
> @@ -390,7 +390,7 @@ test_expect_success SYMLINKS 'stash file to symlink' '
> rm file &&
> ln -s file2 file &&
> git stash save "file to symlink" &&
> - test -f file &&
> + test_path_is_file file &&
This is not wrong per-se, and I know I shouldn't demand too much
from a practice patch like this, but for a real patch, I hope
contributors carefully check if the original is doing the right
thing.
What does the code want to do?
- The starting state, HEAD, has a 'file' that is a regular file.
- We remove and replace 'file' with a symbolic link.
- We stash.
So the expectation here is at this point, 'file' is a regular file
and not a symbolic link. Some anticipated errors are that "stash
save" fails to turn 'file' back to a regular file include leaving it
as a symbolic link and successfully remove the symblic link version
but somehow failing to recreate a regular file.
Is "test -f file", which was used by the original, the right way to
detect these possible errors?
Whey file2 is a regular file that exists and file is a symbolic link
points at it, i.e. if "stash save" fails to operate, "test -f file" would
still say "Yes, it is a file".
$ >regular-file
$ rm -f missing-file
$ ln -s regular-file link-to-file
$ ln -s missing-file link-to-missing
$ test -f regular-file; echo $?
0
$ test -f link-to-file; echo $?
0
$ test -f link-to-missing; echo $?
1
$ test ! -h regular-file && test -f regular-file; echo $?
0
$ test ! -h link-to-file && test -f link-to-file; echo $?
1
As "test_path_is_file" is merely a wrapper around "test -f", this
patch may not make it any worse, but I am skeptical if this is a
good idea, given that possible follow-on project may be one or more
of these:
* verify that all existing users of test_path_is_file want to
reject a symlink to file, and add 'test ! -h "$1" &&' to the
implementation of the test helper in t/test-lib-functions.sh
(we may want to do the same for test_path_is_dir).
* introduce test_path_is_symlink and use it appropriately. This
will be a more verbose version of "test -h".
* introduce test_path_is_file_not_symlink and use it here.
If the proposed log message leaves a note on the issue, e.g.
There are dubious uses of "test -f" in the original that should
be differentiating a regular file and a symbolic link to an
existing regular file, but this mechanical conversion patch does
not fix them.
it would be nicer.
Thanks.
