Xref: utzoo comp.bugs.4bsd:683 comp.unix.questions:5286
Path: utzoo!utgpu!water!watmath!clyde!rutgers!gatech!ncsuvx!lll-winken!lll-crg.llnl.gov!casey
From: casey@lll-crg.llnl.gov (Casey Leedom)
Newsgroups: comp.bugs.4bsd,comp.unix.questions
Subject: Re: Dbm library
Keywords: ndbm library
Message-ID: <3046@lll-winken.llnl.gov>
Date: 28 Jan 88 23:54:08 GMT
References: <580@xyzzy.UUCP>
Sender: usenet@lll-winken.llnl.gov
Reply-To: casey@lll-crg.llnl.gov.UUCP (Casey Leedom)
Organization: Lawrence Livermore National Laboratory
Lines: 40

In article <580@xyzzy.UUCP> meissner@.UUCP (Michael Meissner) writes:
> I remember quite awhile ago, there was a discussion of problems in the
> dbm library, particularly when it has to split an index node.

  The problem is on line 499 of the 4.3BSD source for
/usr/src/lib/libc/gen/ndbm.c.  There is a comparison that involves a
sizeof which converts an expression yielding a possibly negative value
into an unsigned expression which doesn't compare right.  As it turns
out, sizeof yields an int in the VAX compiler instead of an unsigned, so
the comparison works.

  There's some question as to what the official result type of sizeof
should be since the early K&R simply said sizeof(E) was semantically an
integer constant (K&R; C Reference Manual; section 7.2; page 188).
However, later versions of K&R specified that it was an unsigned integer
constant (sorry, I don't have a copy of such a later version to
reference).

  The fix:

*** /usr/src/lib/libc/gen/ndbm.c	Sun Mar  9 19:51:28 1986
--- /tmp/xxx	Thu Jan 28 15:52:11 1988
***************
*** 496,502 ****
  	if (i2 > 0)
  		i1 = sp[i2];
  	i1 -= item.dsize + item1.dsize;
! 	if (i1 <= (i2+3) * sizeof(short))
  		return (0);
  	sp[0] += 2;
  	sp[++i2] = i1 + item1.dsize;
--- 496,502 ----
  	if (i2 > 0)
  		i1 = sp[i2];
  	i1 -= item.dsize + item1.dsize;
! 	if (i1 <= (i2+3) * (int)sizeof(short))
  		return (0);
  	sp[0] += 2;
  	sp[++i2] = i1 + item1.dsize;

Casey