Cell name is lost in buck audit classpath --dot output

[davido]

Am I right in assuming that buck audit classpath --dot output should be able to be used as input to Buck? At least it works like this until cell based targets are used. In this case the cell root is just omitted and the target is rellocated to the cell root.

Note, that the same command without --dot option produces the correct output file name for cell based rules.

Example:

[repositories]
  jgit = lib/jgit

Rule name in jgit cell, that is located under gerrit/lib/jgit:

maven_jar(
  name = 'jgit',
  id = 'org.eclipse.jgit:org.eclipse.jgit:' + VERS,
  bin_sha1 = 'a042910b86546a9d8f8aee745e5f8e80104e386d',
  src_sha1 = '3224099036bc3396f9ad611bc22f72fa1d126685',
  license = 'jgit',
  repository = REPO,
  unsign = True,
  deps = [':ewah'],
  exclude = [
    'META-INF/eclipse.inf',
    'about.html',
    'plugin.properties',
  ],
)

And is consumed by other rules as: @jgit//:jgit. So the output of buck audit is:

  $ buck audit classpath --dot //gerrit-common:server | grep jgit
  "//:jgit" -> "//:ewah";
  [...]

I would expect that the the cell name @JGit either preserved or replaced with its current location, so that it can be feed again to Buck:

  $ buck targets --json //:jgit
  com.facebook.buck.util.HumanReadableException: No rule found when resolving target //:jgit in build file ///BUCK

But:

   $ buck targets --json //lib/jgit:jgit
Using watchman.
[+] PROCESSING BUCK FILES...0.1s
[
{
  "binaryJar" : ":jgit__download_bin",
  "buck.base_path" : "lib/jgit",
  "buck.direct_dependencies" : [ "//lib:LICENSE-jgit", "//lib/jgit:ewah", "//lib/jgit:jgit__download_bin", "//lib/jgit:jgit__download_src" ],
  "buck.type" : "prebuilt_jar",
  "deps" : [ ":ewah", "//lib:LICENSE-jgit" ],
  "gwtJar" : null,
  "javadocUrl" : null,
  "mavenCoords" : null,
  "name" : "jgit",
  "sourceJar" : ":jgit__download_src",
  "visibility" : [ "PUBLIC" ]
}
]

[mkosiba]

Thanks for the detailed report!

The output of the --dot command is intended to be used by the graphviz tool or humans, not fed back into buck. Also, "surprise!" the --dot option does something different than the --json option (look at AuditClasspathCommand, if you pass in --dot it actually inspects the TargetGraph, not the ActionGraph).

The way we handle cells in the two graphs slightly differs but that does look a bit bogus. I don't think anyone will look at this any time soon though.

[davido]

Thanks for the clarification. Have i got it right, that a diff fixing the output for the same option (--dot) would be accepted, that preserves the cell name?