That is totally true indeed. ...but you couldn't you use a little "hack"
to the "hadoop" script and the "org.apache.hadoop.fs.Path" class?
(Sorry, this is just sketchy but gives the idea...)
...in "hadoop" you would add:
cygwin=false
case "`uname`" in
CYGWIN*) cygwin=true;;
esac
...
if $cygwin; then
CYGWIN_HOME=`cygpath -w /`
fi
...
HADOOP_OPTS="$HADOOP_OPTS -Dcygwin.home=$CYGWIN_HOME"
and in the "Path" class do the following (simplified pseudo code):
static final String CYGWIN_HOME = System.getProperty("cygwin.home");
...
public Path(String pathString) {
...
if(isAsbolutePath(pathString) && CYGWIN_HOME != null &&
!CYGWIN_HOME.isEmtpy() ) {
pathString = normalizePath(CYGWIN_HOME) + pathString;
}
...
}
Cheers,
Holger
Doug Cutting wrote:
Holger Stenzhorn wrote:
I am using Hadoop under Cygwin with the default settings.
So hence the "hadoop.tmp.dir" is set to "/tmp/hadoop-${user.name}"
via the "hadoop-default.xml".
Now when I start using Hadoop it creates a directory
"c:\tmp\hadoop-holste" (as "holste" is my user name obviously).
But shouldn't Hadoop rather create the directory "hadoop-holste" in
the Cygwin "tmp" directory?
The Cygwin /tmp directory, as with the cygwin /usr, and /etc, are only
visible to Cygwin applications, like the shell and other C
applications compiled against the Cygwin library. Java is not a
Cygwin application, so paths in Java are always relative to the
Windows filesystem, not to the synthetic paths that Cygwin uses.
Doug
