Update unit test file format.

mojo/tools/test_runner.py

Index: mojo/tools/test_runner.py
diff --git a/mojo/tools/test_runner.py b/mojo/tools/test_runner.py
index 27fe1fa8b20a28619e91f29d580934a7cd0390cf..269fe36ad98601c26a2dfe9f1995f155686b17d2 100755
--- a/mojo/tools/test_runner.py
+++ b/mojo/tools/test_runner.py
@@ -6,6 +6,7 @@
 """A "smart" test runner for gtest unit tests (that caches successes)."""
 
 import argparse
+import json
 import logging
 import os
 import platform
@@ -42,8 +43,9 @@ def main():
 
   _logging.debug("Test list file: %s", args.gtest_list_file)
   with open(args.gtest_list_file, 'rb') as f:
-    gtest_list = [y for y in [x.strip() for x in f.readlines()] \
-                      if y and y[0] != '#']
+    stripped_file = ''.join(
+        [l for l in f.readlines() if not l.lstrip().startswith('#')])

[viettrungluu, 2014/12/03 16:15:37]

I see.

Rather than doing something weird, why don't you just say that the file should
be a valid Python dictionary and eval() it (like gyp)?

(Then you'll get proper comments, including same-line comments, and trailing
commas in dictionaries/lists.)

[qsr, 2014/12/03 16:38:32]

Done.

+    gtest_list = json.loads(stripped_file)
   _logging.debug("Test list: %s" % gtest_list)
 
   print "Running tests in directory: %s" % args.root_dir
@@ -78,13 +80,16 @@ def main():
   # TODO(vtl): We may not close this file on failure.
   successes_cache_file = open(args.successes_cache_filename, 'ab') \
       if args.successes_cache_filename else None
-  for gtest in gtest_list:
-    if gtest[0] == '*':
-      gtest = gtest[1:]
+  for gtest_dict in gtest_list:
+    if gtest_dict.get("disabled"):
+      continue
+    if args.android and not gtest_dict.get("run_on_android"):

[viettrungluu, 2014/12/03 16:15:37]

Having arbitrary/hacky entries like run_on_android kind of speaks to why having
a non-programmatic test list format is doomed to failure. (It's not a successful
format if you have to keep on modifying the consumer(s) of the list to support
various conditions.)

Soon-ish, I want to extract the higher-level list out of get_test_list.py into
something that's exec'ed in Python, which will provide a cleaner programmatic
model and separation between (programmatic) "data" and code.

For now, I can live with this sort of thing (except we shouldn't bother with
pretending that it's JSON), but it's not really that much (if any) better than
just embedding the test list directly in this file....

+      continue
+
+    gtest = gtest_dict["test"]
+    cacheable = gtest_dict.get("cacheable", True)
+    if not cacheable:
       _logging.debug("%s is marked as non-cacheable" % gtest)
-      cacheable = False
-    else:
-      cacheable = True
 
     gtest_file = gtest
     if platform.system() == 'Windows':