RE: Woes on ANZOBJCVN...

["Elvis Budimlic" <ebudimlic@xxxxxxxxxxxxxxxxxxxxxxxxx>]

We had a similar issue.  Built-in C macro used to return sizeof(int) as
signed value.  Then we changed a target release to a newer version and lots
of things started failing in inexplicable ways.  Turns out macro now
returned unsigned value.
Fix was simple, but that macro is VERY popular in C code.  Pretty much all
our parts had to be inspected and corrected where needed.
Suffice to say, it wasn't just a recompile.

Due to this and other gotchas we've hit upon, when we cut a new version, we
test it thoroughly on ALL releases supported by our tools.
Reading Memo To Users is just a first step.  And at times we have to read
all memo to users as well, if compiling to an older version of the OS.

Elvis

Celebrating 11-Years of SQL Performance Excellence on IBM i5/OS and OS/400
www.centerfieldtechnology.com
 

-----Original Message-----
Subject: RE: Woes on ANZOBJCVN...

There really should be regression testing after a recompile because of 
this. Last year we had to recompile over 600 programs because of adding a 
new field to a file. Everything compiled correctly, but the next morning 
one of the programs began blowing up. 

It turned out that the offending program had not be recompiled in several 
versions. During that time the RPG compiler changed the way they handled 
an error when chaining with a negative relative record number. This 
uncovered an invalid operation our program, but because of the compiler 
change, it hadn't been a problem in earlier versions of the compiler. You 
can never tell when something is going to rear up and bite you!

Steve

Steven Morrison
Fidelity Express