John,

Haven't got time to test this, but does the problem go away if you sandwich all your IML commands inside start and finish statements and then add a command to run the module?

Ian.

Ian Wakeling Qi Statistics.

> > Date: Tue, 15 Mar 2005 09:06:51 -0500 > From: John Hixon <john.hixon@KODAK.COM> > Subject: How to detect an Error in an IML subroutine, then exit IML gracefully? > > Hi Folks, > > I used to have time to monitor the SAS-L list regularly, but... > I realize I am missing out on the opportunity to grow > as a SAS pgmer by not finding the time to scan the > list, but... > > Right now, I have a problem that may have a simple > solution, and I'm not seeing it? I'm hoping maybe > some clever IML programmer has a work-around > for me. > > SAS V9.12, Windows > > The following code snippet is from within a bigger macro. > The problem arises because sometimes the CALL LTS > subroutine fails, and I can not find a diagnostic to allow > me to gracefully exit PROC IML. I work around this > problem by detecting that the SC&i dataset does not > exist after the Proc IML exits (with errors) and the macro > runs fine, but...ONLY IN INTERACTIVE MODE. > > I have dozens of SAS Batch-mode tasks that run > overnight. It was a real eye-opener to me how > "picky" Batch mode execution is compared to > interactive mode. The code below works fine > in Interactive mode, but NOT IN BATCH MODE. > > The key issue is how can I tell that the CALL LTS > subroutine failed and quit the IML Procedure? > > I've tried pre-defining the vectors that > CALL LTS creates when it executes, hoping that > I could detect that they were "still the same" as > before the CALL LTS, but...they do not exist > after CALL LTS exits with errors. > > The code below should allow you to reproduce > the problem. Note that in INTERACTIVE mode > I can handle the errors from Proc IML, but, > in BATCH mode, the macro fails after the > 1st time that the CALL LTS subroutine throws > an error. > > *Make a sample dataset that has 3 params. > Params 2 and 3 are OK, but param#1 will > cause the Call LTS subroutine to fail; > data raw; > do nparam=1 to 3; > do seq=1 to 100; > if nparam=1 then do; > Response=0; > if mod(seq,10)=0 then Response=1; > output; > end; > else do; > Response=rannor(12345); > output; > end; > end; > end; > run; > > * Here is a dummy macro just to illustrate > the problem. In interactive mode I can > trap the fact that Call LTS failed by > looking for the existence of the output > dataset that is created when the Call LTS > executes successfully. If the dataset does > not exist, then I know Proc IML exited with > and error.. > BUT...in Batch mode, once the error occurs > one time in Proc IML...all is lost for > subsequent iterations through the macro.; > > %macro ShowProblem(start=1,stop=3); > %do i=&start %to &stop; > data _one; > set raw(where=(nparam=&i)); > run; > %let rc=%sysfunc(exist(work.SC&i)); > %if &rc=1 %then %do; > proc datasets; > delete SC&i; > run; > %end; > proc iml; > use _one; > read all var ("Response") into b; > read all var ("nparam") into nparam; > read all var ("seq") into Seq; > n=nrow(b); > optn = j(8,1,.); > CALL LTS(sc,coef,wgt,optn,b); > wgt2=wgt`; > Quantile=J(n,1,sc[1]); > outlabel="nParam"||"Seq"||"OneZero"||"RdivS"||"Quantile"; > out=nparam||Seq||wgt2[,1:2]||Quantile; > create outlier&i from out[colname=outlabel]; > append from out; > create SC&i from SC; > append from SC; > quit; > %let rc=%sysfunc(exist(work.SC&i)); > %if &rc=1 %then %do; > *stuff for CALL LTS executed...; > %put >>>> In Interactive mode I can catch the errors; > %put >>>> in Call LTS and execute code to deal with it; > title "Iteration &i, Fine and Dandy in Interactive Mode"; > title2 c=blue "CALL LTS executed successfully for Iteration &i"; > proc print data=_one (obs=1); > run; > %end; > %else %do; > *stuff for CALL LTS failed....; > %put >>>> In Interactive mode I can catch the errors; > %put >>>> in Call LTS and execute code to deal with it; > title "Uh Oh, CALL LTS failed, so deal with it..."; > title "Iteration &i, Fine and Dandy in Interactive Mode"; > title2 c=blue "CALL LTS failed Iteration &i, (but I can detect it)"; > proc print data=_one (obs=1); > run; > %end; > %end; > %mend; > > > *If you execute this code in INTERACTIVE mode, > everything is fine (even with the errors in the log > from the failure of CALL LTS). > > But, if you make a small batch file to execute > this same code......the first time the error > occurs in CALL LTS...you are lost. All further > iterations fail as well (even if the data is > 'OK' for CALL LTS.) > > So...how can I exit gracefully from Proc IML > when CALL LTS fails?; > > %ShowProblem(start=1,stop=3); > > > Note, to set up the batch file, make a text file with (typically) these > statements: > > "C:\Program Files\SAS\SAS 9.1\sas.exe" -SYSIN c:\temp\testcode.sas > > and name the file xxxx.bat. Then double click on it to run. > > Does anyone have a cure for this? I'll bet there is a very simple > way to detect an error from CALL LTS within Proc IML and then > exit the Proc IML gracefully, but I just haven't found it. > > Note that the errors in BATCH execution have messages like this: > > > "NOTE: SAS set option OBS=0 and will continue to check statements. > This may cause NOTE: No observations in data set." > > [This is a killer error, since from this point on, EVERY dataset > has 0 observations ! Yikes!] > > In iteration 2 of BATCH mode, I get this error as well: > > "NOTE: IML Ready > WARNING: IML is now in syntax-check-only mode due to errors in previous > steps. No statements > will be executed. > NOTE: Exiting IML." > > [Obviously not a good thing!] > > TIA. > > Best regards, > > John Hixon > Eastman Kodak Company > Rochester, NY USA > 585-477-1984 >