GN(android): Add work-around for "project database corrupted" error

build/android/gyp/javac.py

Index: build/android/gyp/javac.py
diff --git a/build/android/gyp/javac.py b/build/android/gyp/javac.py
index f196581113a5dff8bdbf3c3f3143f0dff1e83ba0..0f4305da98add73cbb10196c3a5d0852debdfeef 100755
--- a/build/android/gyp/javac.py
+++ b/build/android/gyp/javac.py
@@ -238,11 +238,18 @@ def _OnStaleMd5(changes, options, javac_cmd, java_files, classpath_inputs,
       if md5_check.PRINT_EXPLANATIONS:
         stdout_filter = None
 
-      build_utils.CheckOutput(
+      attempt_build = lambda: build_utils.CheckOutput(
           cmd,
           print_stdout=options.chromium_code,
           stdout_filter=stdout_filter,
           stderr_filter=ColorJavacOutput)
+      try:
+        attempt_build()
+      except build_utils.CalledProcessError as e:
+        # Work-around for a bug in jmake (http://crbug.com/551449).

[Yaron, 2015/11/04 16:45:13]

yowza - now that's a hack!
There's also no guarantee that the second attempt will work. At a minimum,
please add a log that we're retrying.

Can you file an upstream bug against jmake?

[agrieve, 2015/11/04 18:20:22]

It should always work the second time since there is now no longer a pdb file to
load (error happens at start-up).

Add a log message so we can at least get a feel for how often it's triggered.

Filed: https://github.com/pantsbuild/jmake/issues/22

+        if 'project database corrupted' in e.output:
+          os.unlink(pdb_path)
+          attempt_build()
 
     if options.main_class or options.manifest_entry:
       entries = []