Resolve symlink issue.

[intelzhongjie]

If there is certain symlink in the container image like this:

/A --> /X/Y/Z

AND if the "/X/Y/Z" doesn't exists on the host OS, then the current
invocation to os.Stat will give errors saying that the "Could not
obtain size for /X/Y/Z". This is because that it will follow the
link to check the /X/Y/Z on the host. That is improper.

The proper behaviour should be:
From the perspective of the symlink inside the container image, the
valid target file it pointing to is always some "existing file" in
the container image.
From the perspective of the host OS, it should just respect the
"closure" of the file system inside the container image.

To resolve it, instead of the os.Stat, we can use os.Lstat which
will not follow the link to check the target file.

Signed-off-by: zhongjie [email protected]

[intelzhongjie]

REPRO STEP:
test_symlink.zip
The attached zip package include the test data to repro the issue and verify the fix. To use it, just unzip it and cd to the sub folder to run the ./build.sh in each sub-folder to build the Docker images for compare.
After that, just run the command like this:

container-diff diff daemon://scratch:base0.0 daemon://scratch:base0.1 --type=file --json --no-cache
ERRO[0000] Could not obtain size for /tmp/scratch@sha256:5b479b1952dd61130f21d5d74dacb20b4c0dd9b04f8674c130507ed98fe4762a487663054/vendor: stat /tmp/scratch@sha256:5b479b1952dd61130f21d5d74dacb20b4c0dd9b04f8674c130507ed98fe4762a487663054/vendor: no such file or directory

[intelzhongjie]

The build error in https://travis-ci.org/github/GoogleContainerTools/container-diff/builds/762932232 are some conflicts between the expectation and actual result on certain symlinks (e.g. /usr/bin/cc, /usr/lib/cpp, /usr/lib/gcc/x86_64-redhat-linux/4.8.2/32/libasan.a, etc.) .
And that's just the new behavior of using Lstat instead of Stat. If you think OK, then I can update the tests/file_diff_expected.json accordingly.

What's your opinion?
@nkubala @tstromberg

Best regards!

[intelzhongjie]

Updated the tests/file_diff_expected.json accordingly.

[intelzhongjie]

@nkubala @tstromberg ping?

[nkubala]

LGTM, thanks for the fix @intelzhongjie!

[intelzhongjie]

Thanks for your review @nkubala !
I'm very happy that I can give the contribution to your helpful tool!