Mailing-List: contact dev-help@ant.apache.org; run by ezmlm
Precedence: bulk
Reply-To: "Ant Developers List" <dev@ant.apache.org>
Message-ID: <4048524F.5010906@iseran.com>
Date: Fri, 05 Mar 2004 10:11:27 +0000
From: Steve Loughran <steve_l@iseran.com>
User-Agent: Mozilla Thunderbird 0.5 (X11/20040208)
MIME-Version: 1.0
To: Ant Developers List <dev@ant.apache.org>
Subject: Re: Sshexec fail on starting a service (websphere)
References: <200403042230.i24MUV4Z027613@barrierb241.nike.com>
In-Reply-To: <200403042230.i24MUV4Z027613@barrierb241.nike.com>
Content-Type: text/plain; charset=us-ascii; format=flowed
Content-Transfer-Encoding: 7bit

Anderson, Rob (Global Trade) wrote:
> The sshexec task, jsch and ssh cannot be expected to anylize the output from a command that is run. This is why there is an exit status. The example of echo is a good one. Echo will succeed and report exit status 0 to the process that calls it. If you wish to anylize the output of echo, you need to do this in the parent process.


While I agree with everything you say, the limited scope of exit codes 
really irritates me.

all we have to return errors and success is a number, a number that can 
also be used in different ways by java.exe, and which behaves 
differently on different boxes.

This code snippet from the LGPL smartfrog tasks (smartfrog.org) show the 
problem. I am forking java, but java.exe can return things if it feels 
like it, so the program we fork knows to return -1 for an error -the 
switch statement should say 0==good, -1 error, else: trouble.

Only java1.4.2 on hpux seems to return 255 instead of -1 on exec. I can 
see a bit of platform specific details creeping in there, and it makes 
me unhappy.

         switch(err) {
             case 0:
                 return;
                 //-1 is an expected error, but
                 //for some reason smartfrog on HP-UX returns something 
else.
                 //so we catch 255 as well.
             case -1:
             case 255:
                 throw new BuildException(failureText);
             default:
                 throw new BuildException(errorText+" - error code "+err);
         }


Can we ever move to a world where executed programs throw a full 
exception up to the caller, rather than just return a status code?

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@ant.apache.org
For additional commands, e-mail: dev-help@ant.apache.org