Received-SPF: neutral (google.com: 216.55.181.205 is neither permitted nor denied by domain of mnecasek@yahoo.com) client-ip=216.55.181.205;
DomainKey-Signature:a=rsa-sha1; q=dns; c=nofws;
  s=s1024; d=yahoo.com;
  h=Message-ID:X-YMail-OSG:Received:X-Mailer:References:Date:From:Subject:To:Cc:In-Reply-To:MIME-Version:Content-Type;
  b=11101480bOZp29FsUcPOXe16gtkPqa1P27/FUI8LpfQgxyx5YLmlgjRvDqrv8T6YfWNRgsaFou8YLTCu/8ACMBO0pNen5+ZpdGcUJ2alo8UCj7qrMPvA3lxaZ9LAfSBeQ6uz/DGhKaEs+0YIeWEf2UVg+lCYNjxESx+x2Xts8Rs=;
Message-ID: <514280.70313.qm@web81404.mail.mud.yahoo.com>
References: <4B95AD0B.40406@vtc.vsc.edu> <526580.37972.qm@web81402.mail.mud.yahoo.com> <4B965315.5010307@openwatcom.org> <16265.40180.qm@web81408.mail.mud.yahoo.com> <4B9703D5.5060806@openwatcom.org> <207030.16169.qm@web81401.mail.mud.yahoo.com> <4B977E6D.8060004@openwatcom.org>
Date: Wed, 10 Mar 2010 04:12:31 -0800 (PST)
From: Michal Necasek <mnecasek@yahoo.com>
Subject: Re: Clib tests.
To: "Peter C. Chapin" <peter@openwatcom.org>
Cc: Leif Ekblad <leif@rdos.net>
In-Reply-To: <4B977E6D.8060004@openwatcom.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii


      Hi Peter,

 The process test is fairly clever, that is true enough. Only one executable is needed to test the parent/child process relationship, which is good.

 But it seems to me that Leif completely misunderstood what exec() does... as you know, the standard UNIX way of spawning child processes is a fork() followed by exec(). That first creates a copy of the parent process (which becomes the child process) and then replaces the copy by a new process image (the actual child).

 The reason why the test was wrong should be pretty clear. Once you exec() another process, the caller goes away. The clib documentation is crystal clear: "If the child process is successfully loaded, it replaces the current process in memory.  No return is made to the original program."

 The test read:

    rc = execve( ProgramName, child_args, child_envp );
    VERIFY( rc == CHILD_RC );
    ...
    printf( "Tests completed (%s).\n", ProgramName );


 The 'completed' message could obviously never be printed. The execve() call starts a child process which takes a different code path; the parent process goes away and does not continue. That is clear from the fact that the test, when calling execve(), no longer prints the message at the end - just try it.

 I hope both you and Leif now understand why the test execve() was wrong and I had to undo it.


         Regards,

              Michal



----- Original Message ----
From: Peter C. Chapin <peter@openwatcom.org>
Subject: Re: Clib tests.

Here's what Leif says in the newsgroups about the exec() business.

Yes, but the proctest is kind of clever. When it does the spawn tests, it spawns itself (and waits for completion). When the program is run with certain parameters, it knows it called itself, and just tests its environment and exits with ok or failure. I can see no reason why it can't exec() itself as well. Exec will return when the program run exists for some reason.

Peter