[net.bugs.v7] bugs in look

geoff@utcsstat.UUCP (Geoffrey Collyer) (11/08/83)

The following bugs are present in look on at least V7 and 4.1BSD.

The -d (dictionary order) option does not work as advertised.  It
ignores whitespace during comparison, despite look(1), which says

     d  `Dictionary' order: only letters, digits, tabs and
	blanks participate in comparisons.

The fix is to change, in canon(),

		if (!isalnum(c))
to
		if ((!isascii(c) || !isalnum(c)) &&
		    c != ' ' && c != '\t')

There is an undocumented -t option, identical to that of sort(1).

There are lots of coding botches.  Look uses continue statements
liberally, and returns from main() without a value.  In canon(),
isalnum and isupper are used without testing isascii first.  The fix
for isalnum is given above, the other correction is to change

	if (isupper(c))
		*new += 'a' - 'A';
to
	if (isascii(c) && isupper(c))
		*new = tolower(c);

The new version also uses tolower portably rather than the previous,
less-portable addition of a magic quantity.

Once the above fixes are applied, look may be used to look up the
spelling of a word (in /usr/dict/words) much more rapidly than egrep
can or to search a phonebook.  Look performs a binary search on sorted
text files and so could be used to look for poor passwords in
passwd(1) [strong hint].

Geoff Collyer, U. of Toronto