Add --dry-run option to simplify CI system integrations

[davido]

Fixes: #105.

Test Plan:

$ git show --name-only HEAD | grep java$ | xargs -r gjf --dry-run
Need Formatting:
[gerrit-common/src/main/java/com/google/gerrit/common/IoUtil.java,
gerrit-server/src/main/java/com/google/gerrit/server/account/Emails.java]

[jbduncan]

@davido I believe this is a duplicate of #106.

[davido]

@jbduncan Oh, indeed, I missed it. And thanks for letting me know.

[jbduncan]

You're very welcome! :)

[sormuras]

And still waiting for an reply from the Google guys. The jitpack solution is a good work-around, though. :)

[dpursehouse]

@cushon any chance to get this or #106 merged? Thanks!

[davido]

Interestingly, the buildifier tool, that is another Google's open source tool for formatting of Blaze and Bazel files, offers the --mode=check|diff|fix option: [1]. Unfortunately, the return code doesn't reflect the outcome of the check: [2].

• [1] https://github.com/bazelbuild/buildtools/blob/master/buildifier/buildifier.go#L44
• [2] Return value in check mode is always zero bazelbuild/buildtools#110

[davido]

supposed to be used from the CI

[davido]

Actually, different check modes are needed: fix|check|diff. This is what Googel's buildifier is offering, so that a CI could run -mode=check first, to see if there were any format issues, and re-run it again with diff mode, to be more verbose on what is actually wrong. That way, tools, that support online editing, like Gerrit Code Review, could address the offending format issues and re-upload new patch set for verification. And all this in your favorite browser.

Example for this approach could be seen in Tensorflow's ci_sanity.sh script.

https://github.com/tensorflow/tensorflow/blob/ce717be4a2966a268e9a4831c39254b611f1c9e7/tensorflow/tools/ci_build/ci_sanity.sh#L264-L294

[cushon]

#106 has been merged.