Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/18/84; site burl.UUCP
Path: utzoo!watmath!clyde!burl!rcj
From: rcj@burl.UUCP (R. Curtis Jackson)
Newsgroups: net.lang.c,net.bugs.usg,net.unix-wizards
Subject: Unix SVR2 on Vax: bug in cpp causes confused sdb
Message-ID: <676@burl.UUCP>
Date: Wed, 1-May-85 11:26:52 EDT
Article-I.D.: burl.676
Posted: Wed May  1 11:26:52 1985
Date-Received: Thu, 2-May-85 01:35:30 EDT
Distribution: net
Organization: AT&T Technologies, Burlington NC
Lines: 36
Xref: watmath net.lang.c:5137 net.bugs.usg:223 net.unix-wizards:13010

I was debugging a program with sdb and was totally bewildered until
I noticed that, on a breakpoint before C instruction N, sdb was actually
executing instruction N and breakpointing before N-1 even though it
told me it was breakpointing before instruction N.  This can really
warp your mind when debugging tight code.  I called the Unix Hotline
and received a VERY fast and accurate reply (my thanks to Alan Berkman
at the Hotline).  Here it is:

cpp(1) has a bug in it which causes it to output bogus line numbers and
therefore confuses sdb about what is where in the source.  If you define
a multiline macro like this:

# define KILLIT\
{\
	register int	c;\
\
	while(((c=input()) != NL) && (c != 0))\
	{\
		/* do nothing!! */\
	}\
	unput(NL);\
}

all hell breaks loose (i.e., the bug appears).  If you put a space
after KILLIT and before the backslash in the first line, however:

# define KILLIT \

then things work fine.

Please no flames on my coding style, I'll ignore them anyway,
-- 

The MAD Programmer -- 919-228-3313 (Cornet 291)
alias: Curtis Jackson	...![ ihnp4 ulysses cbosgd mgnetp ]!burl!rcj
			...![ ihnp4 cbosgd akgua masscomp ]!clyde!rcj