On Fri, 27 Jan 2017 14:53:58 -0500, Rick Troth wrote:
>On 01/27/17 14:48, Charles Mills wrote:
>> I*think* that generally that message is output by the application.
>> The application calls strerror() which returns that string,
>> and then the application prints it. I think your argument
>> is with the application, not LE.
>
>Sure, except the application in question is *ours* and there's no
>strerror() or perror() or sprintf() for this condition. What we're
>dealing with here is a call to stat() or equivalent against a file that
>is not there which we will then create.
>
What's "equivalent" to stat()?
I agree with Charles. System calls rarely issue messages; that's not kernel's
job. System calls set ERRNO and the caller issues the message.
Does the problem occur with a minimal program that does only a stat() and
reports the error? If so, numerous users would be reporting it and creating
SRs.
Google (and my experience) tells me that a frequent cause of ENOENT is
nonexistence of a directory along the path.
If the file is about to be created anyway, just do it: open( file, O_CREAT ... )
If you need to know whether the file is actually being created (perhaps you
intend to write a preamble the first time), stat() leaves a timing hazard:
two jobs running at the same time might both get ENOENT from stat()
and both create the preamble. Better:
open( file, O_EXCL O_CREAT O_WRONLY ... )
if no error then
create preamble
else if EEXIST then
open( file, O_WRONLY ... )
(But you may still get the undesired message.)
-- gil
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN
