[Hadoop-common-dev] [jira] Created: (HADOOP-3839) hadoop should handle no cygwin on windows more gracefully

hadoop should handle no cygwin on windows more gracefully
---------------------------------------------------------

Key: HADOOP-3839
URL: https://issues.apache.org/jira/browse/HADOOP-3839
Project: Hadoop Core
Issue Type: Improvement
Environment: Windows XP, cygwin not installed
Reporter: Steve Loughran
Priority: Minor


There have been a couple of postings to hadoop core-user in which people can't get hdfs to come up on windows because whoami isnt on the path, which fails with an IOException error code 2.

To people not experienced in DOS error codes, this is a fairly meanless number which invariably leads to time wasted and support emails.

1. the error could be caught and handled by printing some better hints (point to a wiki page?)
2. is whoami really needed on DOS-based filesystems?

[ https://issues.apache.org/jira/browse/HADOOP-3839?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12617500#action_12617500 ]

Doug Cutting commented on HADOOP-3839:
--------------------------------------

There are periodic requests to remove Hadoop's cygwin requirements, e.g., to convert Hadoop's shell scripts to batch files. The problem is that most Hadoop developers develop only on Linux, so that, if direct support for Windows is added, it will not likely be well maintained, and will result in more bug reports, not less.
It's not just whoami. There are a number of places where Hadoop depends on having standard unix command line utilities, as well as shell scripting. It seems like perhaps 'whoami' is the just the first one that folks encounter currently.

Perhaps we should add a test somewhere that, when running on Windows, explicitly checks for cygwin and throws an exception if it's not present. The bin/hadoop script is an obvious place to put it, but that would not have caught this case, where someone is running unit tests.

[ https://issues.apache.org/jira/browse/HADOOP-3839?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12617600#action_12617600 ]

Steve Loughran commented on HADOOP-3839:
----------------------------------------

I've seen the other places where cygwin is used (e.g file permission checks), and it does make sense -at least for the servers. I'm not so convinced it needs to be mandatory client side.

A check fror cygwin would be good...one option is just to handle that failure we see in the stack trace, and encode the knowledge that if whoami fails with error code=2 on windows, somebody probably hasn't got cygwin installed, or its path set up correctly.