Re: [Dpcl-user] DPCL problem with line numbers for large programs

[Dave W. <dwo...@us...>]

John

Sorry, this was an oversight on our part. Unfortunately there is no 
notification when a bug report is posted and we missed this.

We will look into it.

It looks, based on the bug report as if this is related to newer levels of 
compilers. For the large C program which also fails, did this work on any 
level of the C compiler? If so, do you know at which level it started 
failing?

Also, you  mention the direct socket connection. This is in fact left over 
code from initial development work, and has not been kept up do date. The 
only supported manner to invoke the DPCL daemon at this time is by 
connecting thru inetd and running as a daemon.

Our usual practice when debugging the daemon is to put a sleep() call at 
entry to main for long enough to allow us to use dbx to attach to the 
daemon, then once the daemon is started, use dbs -a to attach to it.

Dave




John May <jo...@ll...>
Sent by: dpc...@ww...
12/06/2002 02:12 PM
 
        To:     dpc...@os...
        cc:     pa...@ll..., gy...@ll...
        Subject:        [Dpcl-user] DPCL problem with line numbers for 
large programs

 

Last month I posted a bug to the DPCL web site regarding
line number info for large programs.  (See bug 3356.)  Since
there has been no response to that posting, I thought I'd
send a message to this list.  The problem is that for large,
multifile programs, DPCL fails to get line number information,
even though the program has been compiled with -g.  I have
demonstred this with a Fortran code, and a colleague here
at LLNL has seen it with a C program.  Interestingly,
when he tried the same large program all in one file, the line
number information came out correctly.  The bug report referenced
above has a link to the program that failed.  I did this test using
the DPCL listmod test program, and I have observed the behavior
on both 3.2.5 and 3.3.2 on systems running AIX 4.3 and 5.1.

I have dug into the source code a bit, but debugging is quite
tedious when you have to run the DPCL daemon as an actual daemon.
I can see the hooks for a method to debug the daemon as a regular
process that is connected to the DPCL client through a direct
socket connection.  Debugging this way would be much easier,
but a couple of hours work trying to get that method to work
has produced no results so far -- the daemon gets messages it
doesn't understand from the client, which makes me suspect that
this code is left over from an earlier release.

I would be interested to know if others have seen this problem
and if anyone is working on a solution.  I consider this
bug to be very serious, since it prevents DCPL from working on
large (i.e., real) applications.

Thanks for any help you can offer.

John May
_______________________________________________
Dpcl-user mailing list
Dpc...@ww...
http://www-124.ibm.com/developerworks/oss/mailman/listinfo/dpcl-user