[Chromecast] Dump the runtime deps for all test targets.

chromecast/build/tests/cast_test.gni

Index: chromecast/build/tests/cast_test.gni
diff --git a/chromecast/build/tests/cast_test.gni b/chromecast/build/tests/cast_test.gni
index 6a98bbe4ddb5a7ef0f5d606b4dfe6b5dcaf430e5..dea21632224044a955bd0e18755040357bc5943d 100644
--- a/chromecast/build/tests/cast_test.gni
+++ b/chromecast/build/tests/cast_test.gni
@@ -201,6 +201,20 @@ template("cast_test_group") {
     _test_names += [ get_label_info(_test, "name") ]
   }
 
+  # Create a dummy target so that we can get the runtime deps for each test
+  # and output them into a file. The runtime deps include all of the data files,
+  # data directories, and shared libraries that a test needs in order to run.
+  foreach(_test, invoker.tests) {
+    _test_name = get_label_info(_test, "name")
+    group(_test_name + "_cast_runtime_deps") {

[alokp, 2017/01/17 21:30:25]

"_cast" prefix seems unnecessary

[slan, 2017/01/18 17:49:21]

Actually, I think having "cast" in the name is useful, so that somebody upstream
creating net_unittests_runtime_deps doesn't accidentally collide with our
target. This is just an internal detail, after all.

[alokp, 2017/01/18 23:35:36]

Don't we already have test_name in the group target-name, which should be unique
to start with?

[slan, 2017/01/19 00:15:03]

Yes, but since we build tests from other parts of Chrome, those names could
create a chance of collision. For example, we include in our test list
//net:net_unittests. _test_name is "net_unittests", and the name of this target
then becomes "net_unittests_cast_runtime_deps". If we were to remove "cast" from
the suffix, this target would be named "net_unittests_runtime_deps", which has a
much higher chance of future collision with something in //net. 

...but as I type this, I realize that that isn't really a problem: it would
simply necessitate typing the fully-qualified path on the command-line (which we
will never do). So, I don't really care whether or not we leave "cast" in or
out. It is an internal detail, so I don't think it matters all that much.

+      testonly = true
+      data_deps = [
+        _test,
+      ]
+      write_runtime_deps = "$root_out_dir/${_test_name}.runtime_deps"

[alokp, 2017/01/17 21:30:25]

I would suggest using a proper file type (txt, json, etc) instead of a
runtime_deps. You should probably use "_runtime_deps" suffix in the filename.

[slan, 2017/01/18 17:49:21]

I assume this convention was chosen to match the exsiting GN default convention
(gn help --runtime-deps-list-file). But I agree, I think that I prefer normal
extensions for files that we're generating.

I am more concerned that this clutters the root_out_dir substantially. Could we
put these next to the other files generated by this target?

write_runtime_deps = 
    "$_shared_dir/runtime_deps/${_test_name}.txt

You could also then update the python script to compile all of these files into
a single key-value file (like the run list), leaving you with exactly three
output files from cast_test_lists:
 - build list 
 - run list
 - runtime_deps

That would obscure some of the internal details of this from the test infra
scripts, meaning that we won't have to keep anything from this file in sync with
that code (only the paths specified in chromecast/BUILD.gn). Thoughts?

Thoughts?

+    }
+  }
+
   # This action generates a list of target names to build and run. It will be
   # depended upon by the "pack_build" action of the cast_test_group_list
   # instance which depends on this cast_test_group.