SYSDEBUG and the compile listing do not have the same information for APA to report the consistant CPU time

IBM Enterprise COBOL for z/OS is a continuous delivery offering, and aims to satisfy the needs of a rapidly evolving market segment. Product strategy evolves and requirements are continuously evaluated against that strategy. This RFE has recently been reappraised against the wider product strategy and does not fall into IBMs delivery plans for the next 12 month. Whilst this RFE might be valid and we might look to deliver longer term, this RFE is being rejected at this time. The requirement will be kept in our internal RFE backlog and might be reassessed in the future.

Due to processing by IBM, this request was reassigned to have the following updated attributes:
Brand - Servers and Systems Software
Product family - Programming Languages
Product - COBOL Compilers

For recording keeping, the previous attributes were:
Brand - Rational
Product family - Design & development
Product - COBOL Compilers

We would like to provide a solution to the missing END-PERFORM information part of this RFE

PMR79826:
Problem Details
.
Product or Service: Application Performance Analyzer z/OS Base 11.1.0
Component ID: 5697Q0300
.
Operating System: z/OS
.
Problem title
COBOL SYSDEBUG Mapping
.
Problem description
Using SYSDEBUG in mapping a COBOL program for CPU time per instruction
results in listing the CPU time at the end of a code black rather than
the specific instruction. Using the COBOL compile listing produces
expected CPU time per instruction information.

I will upload the COBOL source, SYSDEBUG file, Compile listing, and the
applicable sample dataset.
.


Call requeue from BNKNET,03D Greg Banknet Martin 3/11/13 6:34 AM

In the procedure map in the source listing, there is an entry for
the END-PERFORM, line # 31, at offset 378. I then traced loading of
the SYSDEBUG file and found that the PMAP entries returned from
SYSDEBUG do not include this offset and statement number.
APA assigns CPU active samples to the offset that is greater than or
equal to the sampled offset. This causes a number of samples to be
assigned to the last verb with an offset, rather than the END-PERFORM
where a number of instructions are generated.
This PMR should go to COBOL support.
Contact made Jayson Liu 3/11/13 10:00 AM
Call requeue from ADS,114 Jayson Liu 3/11/13 10:01 AM
Jayson Liu 3/11/13 10:01 AM
Action Taken: contacted COBOL L2 Dan
Action Plan: pass call to COBOL compiler support

Call Generate on COBP,114 (Secondary) Carolyn B Mazurek 3/12/13 12:42 PM
PMR field(s) altered Carolyn B Mazurek 3/12/13 12:44 PM
Call requeue from COBP,114 Carolyn B Mazurek 3/12/13 12:44 PM
Carolyn B Mazurek 3/12/13 12:44 PM
Action Taken: Returning primary record to Jayson at ADS,114
Action Plan: COBOL Level 2 to review secondary copy at COBP,114.
Contact made (Secondary) Carolyn B Mazurek 3/12/13 12:46 PM
Jayson Liu 3/12/13 1:02 PM
the following files are Saved on mvs1.
LANG.P79826.B122.C000.APASAMP
LANG.P79826.B122.C000.COBCOMP
LANG.P79826.B122.C000.COBSRC
LANG.P79826.B122.C000.SYSDEBUG

Action Taken: I reviewed this problem and discusssed it with Dolly H
in COBOL Development.
I created two additional files on STLMVS1:
LANG.P79826.B122.C000.APASAMP.RECEIVED
which is a RECEIVE of the file
INDSN('LANG.P79826.B122.C000.APASAMP')
and LANG.P79826.B122.C000.IBMSRCE
which is the customer source plus all the compiler
options from the listing.
Action Plan: Dolly H will review and advise in the next couple
of days (March 12-15).

Action Taken: This is from Carolyn M in COBOL Support.
As mentioned in my update yesterday, I asked
D H, who works with COBOL Runtime and Debug Tool,
to review first. However, after discussing with
T R in COBOL Compiler Development, it appears to
involve the compiler SYSDEBUG design. Therefore,
it would not be a defect.
***
The program listing we were sent was compiled
with the NOLIST option, so we cannot see the
generated code.
A listing with the LIST option and a description
of exactly what generated assembler instructions
are involved might be helpful in describing this
issue.
Action Plan: Return to Jayson for next step.
**********
Hello Jayson,
Please see my update above.
You may contact T R if you want to discuss this with him.
Regards,
Carolyn
Call requeue from COBP,114 (Secondary) Carolyn B Mazurek 3/13/13 4:14 PM
Jayson Liu 3/13/13 8:09 PM
"The problem APA development trying to point out is that the verb map in
listing has the END-PERFORM listed with a statement number and offset,
but SYSDEBUG data ignores it. "
Action Taken: forwarded APA developer's note to COBOL developer Tom Ross


*** Email: james.i.mearns@bankofamerica.com

Additional comments
Compile listing with LIST option sent to FTP:
79826.122.000.COBOL.COMPILE2.TRS
.
Call Entry for an existing problem on ADS,114 (Secondary) TIVOLI 11 3/27/13 1:35 PM
CDDR12 PMRUPDATE BDC 3/27/13 1:35 PM
Untersed data now available on MCEVS1-System :
/ecurep/pmr/7/9/79826,122,000/2013-03-27/79826.122.000.COBOL.COMPILE2.T
RS
->ONTOP.GS122.P79826.C000.COBOL.COMPILE2
Call requeue from ADS,114 Jayson Liu 3/27/13 10:42 PM
Jayson Liu 3/27/13 10:42 PM
Hi Greg,
I have
ftped
you the compile listing customer sent as
'LANG.P79826.T1'.
Thanks,
Jayson
Contact made Greg Banknet Martin 3/28/13 7:18 AM
Call requeue from BNKNET,03D Greg Banknet Martin 3/28/13 7:18 AM
Greg Banknet Martin 3/28/13 7:18 AM
I've seen the LIST option and it also generates code at the END-PERFORM
statement. The SYSDEBUG/DWARF output is inconsistent with both LIST
and OFFSET. This is definitely a COBOL issue.
Contact made Jayson Liu 3/28/13 2:57 PM
Call requeue from ADS,114 Jayson A Liu 4/4/13 4:00 PM

Hi James,
After further discussion among COBOL, Debug Tool and APA development we
think this is a design issue of COBOL compiler. There is more testing
involved between COBOL and Debug Tool. COBOL compiler will correct
the problem in the next version of PD Tools.
We apologize for the inconvenience.
Regards,
Jayson
---
Action Taken: inform customer this is a design change issue
Action Plan: problem will be fixed in the next release

Hi James,
Could we close this record since it will be fixed in the next release?
Rgds,
Jayson
--
Action Taken: request to close
Action Plan: wait for customer's response


*** Updated by: James Mearns
*** Email: james.i.mearns@bankofamerica.com
Additional comments
Would it be possible to get the APAR number for COBOL Compiler so our
Operating System Support team can track it?
.
Call Entry for an existing problem on ADS,114 TIVOLI 11 4/11/13 1:14 PM
Contact made Jayson Liu 4/11/13 3:30 PM
Call requeue from ADS,114 Jayson A Liu 4/11/13 3:31 PM
Jayson A Liu 4/11/13 3:31 PM
Hi James,
Since this will not be fixed in a APAR, instead in the next release
I can enter a RFE for you to track. Is it ok ?
Rgds,
Jayson


*** Email: james.i.mearns@bankofamerica.com
.
Additional comments
Sure Jayson, the RFE would be fine. And just so I know all the moving
parts. Is the RFE for specifically the Cobol compiler, or for the
Language Environment in general, or for z/OS? I guess I'm just looking
for the FMID the RFE will alter. Thanks.
.
Call Entry for an existing problem on ADS,114 TIVOLI 18 4/12/13 7:06 AM
Contact made Jayson Liu 4/13/13 8:04 AM
Call requeue from ADS,114 Jayson A Liu 4/13/13 8:24 AM
Jayson A Liu 4/13/13 8:24 AM
Hi James,
It will be a COBOL compiler RFE.
Rgds,
Jayson

*** Email: james.i.mearns@bankofamerica.com.
The customer has requested this ticket be closed.

Understood, you can close this if you'd like and can send that RFE#
when you have it to:
james.i.mearns@bankofamerica.com
.
Call Entry for an existing problem on ADS,114 TIVOLI 18 4/19/13 1:00 PM
Contact made Jayson Liu 4/22/13 7:45 AM
Service given: [19] We gave a fix Jayson Liu 4/22/13 10:16 AM

Action Taken:
James,
I will pass you the RFE once created.
Thanks,
Jayson
--
close SR per customer
Action Plan: email RFE once entered.
problem summary:
Action Plan: email RFE once entered.
problem summary:
Using SYSDEBUG in mapping a COBOL program for CPU time per instruction
results in listing the CPU time at the end of a code black rather than
the specific instruction. Using the COBOL compile listing produces
expected CPU time per instruction information.
solution: enter COBOL RFE
Jayson Liu 4/24/13 7:02 PM
entered and passed RFE34085 to James

Sorry about the confusion, please ignore timestamp difference part and work on the END-PERFORM problem. (see pmr79826 attached)

The title of the RFE is something about time stamp mismatch between SYSDEBUG and listing, but the content talks about END-PERFORM marker. We understand and could do something about the END-PERFORM problem, but the timestamp difference is a not understood. How does this cause a problem? What is the timestamp compared to? Are you assuming the timestamp is at the beginning of the compile or end of the compile? How can that interfere with runtime performance measurements? Checking with submittor, but it is hard to find out who that is!