Pass file Regions along with FDs to child processes on Android

base/files/memory_mapped_file.cc

Index: base/files/memory_mapped_file.cc
diff --git a/base/files/memory_mapped_file.cc b/base/files/memory_mapped_file.cc
index 891fcffc4cb434b74729657b7981a400729f2e61..1a22bb03df02ede73606bacd402ed52e953a3f16 100644
--- a/base/files/memory_mapped_file.cc
+++ b/base/files/memory_mapped_file.cc
@@ -16,10 +16,11 @@ const MemoryMappedFile::Region MemoryMappedFile::Region::kWholeFile(
 MemoryMappedFile::Region::Region(base::LinkerInitialized) : offset(0), size(0) {
 }
 
+MemoryMappedFile::Region::Region() : offset(0), size(0) {

[Lei Zhang, 2015/05/28 18:55:53]

Perhaps this one should be (-1, -1) to indicate invalidness.

[agrieve, 2015/05/28 19:13:24]

sgtm (done)

+}
+
 MemoryMappedFile::Region::Region(int64 offset, int64 size)
     : offset(offset), size(size) {
-  DCHECK_GE(offset, 0);

[Lei Zhang, 2015/05/28 18:55:53]

why remove the DCHECKs?

[agrieve, 2015/05/28 19:13:24]

There are times (like when going through jni) where it's nice to be able to
convert to two int64s and back again. These DCHECKs prevent doing so for
kWholeFile.

[Lei Zhang, 2015/05/29 21:14:09]

Assuming we stick with -1 to indicate invalidness, and you would only want to
reconstruct valid Regions from JNI, how about:

if (size == 0) {
  DCHECK_EQ(0, offset);
} else {
  DCHECK_GE(offset, 0);
  DCHECK_GT(size, 0);
}

[agrieve, 2015/06/01 14:50:05]

This wouldn't allow you to convert an invalid region (-1,-1) to size/offset and
back to a region (as is done in JNI).

How about I move this check to Initialize (which I've gone ahead and done)? This
way we still have the same check where it counts, and Region gets to behave more
like a POD.

-  DCHECK_GT(size, 0);
 }
 
 bool MemoryMappedFile::Region::operator==(