Duh. Thanks a lot!

Why is static analysis not finding this? It should.

I have not seen any compiler warnings about this either.

Maybe because lev is outside the loop?

Also, as I read it, lev is never set back to 0. So because this variable is outside the loop, won't this apply to all following lines? Has ASL0 ever worked?

lev can be set to any int value by the sscanf() on line 609:

607:		if((asg=strtok(NULL," \t\n"))) {                           // ASG / ASL
608:			if((asl=strtok(NULL," \t\n")) &&
609:			  (sscanf(asl,"%d",&lev)!=1)) lev=1;
610:		} else {
611:			asg=(char*)default_group;
612:			lev=1;
613:		}

The inner if () only runs the lev=1 if the asl=strtok() fails or the sscanf() doesn't convert exactly one decimal integer.

Given that the default in the code above is ASL 1, I suspect this PR should be changed to set lev=1 instead of lev=0.

Dirk, can you look at the pvlist file where you were seeing this issue and confirm that the first PV name pattern in it gives an ASG name but not an ASL value? That's the only way I can see lev not being initialized by the existing code, although it is documentated that the ASL is optional in the pvlist's ALIAS and ALLOW lines.

None of my pvlist files use ASL. I usually use two ASGs, DEFAULT and WRITE, so my pvlist files have lines like
Pattern ALLOW
Pattern ALLOW WRITE

Good, all our pvlist patterns that name an ASG also give an ASL, which explains why we haven't seen this problem here. We do make extensive use of ASL (or more accurately our pvlist files have patterns with both settings), but I don't know how important the choice is to our protection design.

I'm not certain that I understand how ASL is used in the gateway, but we do have ASGs with separate rules for level 0 and level 1 (not sure if higher level numers are allowed here or not, they aren't by the IOC) so this lets us set different access rules for different users on the same group or records.

docs/GATEWAY.pvlist also says the default is 1:

#        <asl>             = Access Security Level (0 or 1)                  [1]

I will change it.

Initializing to 1 is a improvement. I think it should be done inside the loop though.

	while(fgets(inbuf,sizeof(inbuf),fd)) {
		++line;
		pattern=rname=hname=NULL;
                lev = 1; /* <-- add */

imo. a better long term strategy would be to move these variable definitions into the loop. I think this would more clearly represent the intent to readers, and to static analysis tools.