Hi All,

The subject of automated methods to scan a log for errors/warnings/bad notes seems to come up every month (as in concurrent thread "Easy Tracking of Errors"). I thought it might be useful to have a broad conversation about methods/techniques.

e.g. What methods have you used to automate log-checking? What traps have you fallen into? (as Peter Crawford mentioned to me at SUGI, a trusted but poorly designed log-scanner is a very dangerous thing indeed). What bells-and-whistles have you added that you really like?

I'm thinking about SAS solutions, that would be automatically invoked after a program has been submitted (or as part of that program). But of course non-SAS solutions are always interesting as well. Here are some of my own quick thoughts:

It seems to me that when I review the log scanning macros I've seen (e.g Li and Troxell NESUG 2001 paper, and posted code by Ron Fehd, Meredith Clark, Joe DeShon and others, and my own initial attempts), I really like the approach taken by Li and Troxell. One of the important questions for a log-scanning utility is what to do with various NOTE: messages. Most of the code I have seens on SAS-L includes a list of NOTEs that should be included in an error report. Li and Troxell argue for the opposite approach: give the macro a list of NOTEs to exclude and report everything else by default. (Actually, better than that, their macro includes a parameter for include/exclude, and a parameter to name a dataset listing the notes, and...). This seems the safer route, especially as I frequently come across new (to me at least) NOTEs when playing with procedures I haven't used before, and of course it's not unheard of for the actual text of a a NOTE to change from version to version. In general, it seems much easier to add a new exclusion than to recognize that you have omitted an inclusion.

As for pitfalls, one that I have fallen into, and I think some others have as well, was to search for the word "ERROR:" (note the colon as part of the string), which will miss errors such as: ERROR 180-322: Statement is not valid or it is used out of proper order.

Perhaps this is too general. Here are some more specific questions.

1. How do you "prepare" the log for scanning? That is, do you use altlog or printto to redirect the log to another file? Or do you run in batch and then scan programName.log?

2. Once you have the log sitting somewhere, how do you find the errors/notes? I have seen some code that reads in an entire line, then uses index() or indexw(). Other code reads in just the first 20 characters or so of each line. Still other code creates a SAS dataset containing select lines of the log, which can then be subset via a where statment using constructs like: Line NOT LIKE "NOTE: % records were read from the infile %."

3. Once you have identified all of the problematic messages, where/how do you like to report them? I could imagine them being appended to the main output file, or as a separate file, or emailed, or...

4. How has writing/using an automated log checking program changed your coding practices? Just having played with this myself, it has made me less likely to use the word "error" in my code. Instead of coding: IF <bad> then put "ERROR: something bad", I will code something like: IF <bad> then put "ER" "ROR: something bad" or in macro, %if &bad %then %put ER%str()ROR: something bad. Also, although my super-ego knows it's good to avoid automatic conversions, I find the presence of an error report helps my ego maintain control over the laziness of my id.

Ok, enough rambling, any thoughts welcome.

Kind Regards, --Quentin