Properly keep compatibility with openRead() during a write.

Bug: 157092639
Test: atest AppIdleHistoryTests#testFilesCreation
Change-Id: I7d6808139aa41b25715293207b8ad42f9f42ef65
This commit is contained in:
Hai Zhang
2020-05-20 12:07:11 -07:00
parent 25adf3c471
commit 30afab690e

View File

@@ -227,8 +227,18 @@ public class AtomicFile {
}
}
// Don't delete mNewName here - it was okay to call openRead() between startWrite() and
// finishWrite(), and we have to keep supporting it.
// It was okay to call openRead() between startWrite() and finishWrite() for the first time
// (because there is no backup file), where openRead() would open the file being written,
// which makes no sense, but finishWrite() would still persist the write properly. For all
// subsequent writes, if openRead() was called in between, it would see a backup file and
// delete the file being written, the same behavior as our new implementation. So we only
// need a special case for the first write, and don't delete the new file in this case so
// that finishWrite() can still work.
if (mNewName.exists() && mBaseName.exists()) {
if (!mNewName.delete()) {
Log.e(LOG_TAG, "Failed to delete outdated new file " + mNewName);
}
}
return new FileInputStream(mBaseName);
}