[Info-vax] Home-grown application process dumps

[abrsvc]

> The thing that can be a problem to believe with such is, based upon how 
> long the application has been in use, How could it be considered 
> "working"?  There can of course be seldom used parts of the code.
> 
> There is the question, what has changed recently?
> 
> Lots to look at with such problems, and, if you're like me, usually it's 
> the last thing you look at.  Maybe I should say that that it's far down 
> the list of things you look at, since when you find it, you stop 
> looking, thus the last thing you looked at.
> 
> :-)

>From experience, I can tell you that the above statements are not necessarily true.  I have recently located and fixed a few bugs in an application that is over 30 years old.  The portion that had the problem was over 20 years old and had been in production at many sites for that time.

There are a number of reasons for the above including:

1) Other bugs caused problems which prevented the "new" one from being seen.
2) Some conditions under which the bug occurred, just never were encountered.
3) Got lucky.

In one particular case, the problem was the overwriting of data.  In this case, the code was in FORTRAN and the data area was a character string.  The code wrote 20 bytes of character data into a variable that was 2 bytes long (defaulted to Integer*2 for those that care.)  As long as the data that succeeded the names character string was not referenced after being overwritten, no problem was ever found.  I found it because of an accvio that only occurred occassionally.  I turned on bounds checking and ran into that and many other cases of similar overwriting.  Simple fix was to make the data area the correct size in some cases.  In one other case, the variable was defined but with a type which caused it to not truely be defined.  A fix to the variable name type resolved that one.  That particular one was a bug from day one 30 years ago!!

While the real "bug" here may be difficult to locate, it is likely to be a real simple error as well.

Dan