If You Wrote This Code, You Suck

I don't know who wrote the following code...somebody at Sun, but it's cost me my entire night:

#if !defined(O_DSYNC) || !defined(O_SYNC)
#define O_SYNC  (0x0800)
#define O_DSYNC (0x2000)
#endif

There I was building out my hadoop cluster. Everything was going pretty well on my laptop, so I decided to move it over to my FreeBSD box to get a real instance running. I configured it all up, formatted the namenode, and then tried to start it. Then it told me this:

07/04/10 00:17:23 ERROR dfs.NameNode:
java.io.FileNotFoundException:
/data/hadoop/data/dfs/name/current/VERSION (File exists)
  at java.io.RandomAccessFile.open(Native Method)
  at java.io.RandomAccessFile.<init>(RandomAccessFile.java:212)
  at org.apache.hadoop.dfs.Storage$StorageDirectory.read(Storage.java:144)
  [...]

So, it's pretty obvious what's happening there. The native method is opening my file with O_EXCL. But why? Must be a bug in Hadoop, right? RandomAccessFile has no option that's anything like O_EXCL, so it has to be a bug in the JVM.

I started digging through my source and it seemed pretty straightforward (although excessively limiting). The only thing it could be was something overriding O_SYNC to have the same value as FreeBSD's O_EXCL.

And there it was: j2se/src/share/native/java/io/io_util.h. Luckily, someone else had already found this problem between the time I installed this JVM and ran into it myself, so I'm doing a build.

I'll be cursing the author of the above for several hours while this JVM build completes. Curse you, anonymous Sun programmer. You must use your laziness for good, not for evil.