Re: Bug#62339: rc bug

Re: Bug#62339: rc bug

[Byron Rakitzis]

Thanks for tracking this down.

I'm the original miscreant as far as the documentation style of rc is
concerned.

Much of this code was written roughly 10 years ago, I'd certainly write
it differently now.

The metacharacter array is properly an array of booleans. Hence the
explicit checks against 0 and 1, which should be cleaned up somehow
to reflect their function. A macro "ismeta(p, i)", perhaps. I don't
know where the tests against '\001' snuck in. I think it's a mistake.

> Now in lmatch() when we go through looking for '*'s, we have this
> loop:
>
> 	for (i = 0; p->w[i] != '\0'; i++)
> 		if (p->w[i] != '*' || p->m[i] != 1) {
>
> Which assumes that m is as big as w (i.e, not NULL). I came up with
> this replacement, which works for all the test cases I can think of:
>
>         if (s == NULL) {
>                 if (p == NULL) /* null matches null */
>                         return TRUE; 
>                 for (; p != NULL; p = p->n) /* one or more stars match null */
>                         if (p->w && strspn(p->w, "*") == strlen(p->w) &&
>                             p->m && strspn(p->m, "\001") == strlen(p->m))
>                                 return TRUE;
>                 return FALSE;
>         }
>
> I was also able to get rid of 'okay', which i found hard to follow.
> The comment about the null string is gone too, because IMO the null
> string fails for the same reason 'foo' fails. There's really no need
> for an explicit check.

This is good.

I'm picking nits here, but since I obsessed over this code a while ago,
I might suggest two small changes to your fix:

>                         if (p->w && strspn(p->w, "*") == strlen(p->w) &&
>                             p->m && strspn(p->m, "\001") == strlen(p->m))

to:

	if (p->m != NULL &&
		strspn(p->w, "*") == strlen(p->w) &&
		strlen(p->m) == strlen(p->w))

Firstly, p->w can not be NULL here; there can be no null words in
a word list, only empty strings which are represented as "".

As long as you are getting sneaky about using the string library to find
zeroes in a character array, you may also rephrase the metacharacter
test as two calls to strlen.

Byron.

Re: Bug#62339: rc bug

[Tim Goodwin]

A big "thank you!" to Byron and Decklin for getting to the bottom of
this bug.  I've incorporated their fix, plus a couple of other recent
fixes into a new snapshot, which you can get from here.

    http://www.star.le.ac.uk/~tjg/rc/snap/rc-1.6s20000419.tar.gz

This has only been very lightly tested; there might well be some
booboos.  After the Easter break, I'll scrutinize the changes again, and
wrap them into a new beta release.

Changes since the last beta are summarized below.

Tim.


2000-04-19

  Bug: isatty() tests in input.c are relevant to any fd, not just 0.
  Now `. -i /dev/tty' works right.

  Bug: fn sigexit wasn't always cleared in child shells.

  Bug: `~ () '*'' dumped core.

Re: Bug#62339: rc bug

[Decklin Foster]

[-- Attachment #1: Type: text/plain, Size: 1882 bytes --]

Tim Goodwin writes:

> Thanks for the report.  I can reproduce this.
> 
> It's not been reported before, so I don't know what's going on, but I'll
> let you know as soon as I have a fix.

I did send something to to mailing list on Saturday, but it hasn't
gotten through yet. I seem to have figured out the use of m in a List
structure: if there are no metacharacters, NULL, and if there are, we
put a \0 where there's a non-metacharacter in the corresponding part
of n, and a \001 where there is.

Now in lmatch() when we go through looking for '*'s, we have this
loop:

	for (i = 0; p->w[i] != '\0'; i++)
		if (p->w[i] != '*' || p->m[i] != 1) {

Which assumes that m is as big as w (i.e, not NULL). I came up with
this replacement, which works for all the test cases I can think of:

        if (s == NULL) {
                if (p == NULL) /* null matches null */
                        return TRUE; 
                for (; p != NULL; p = p->n) /* one or more stars match null */
                        if (p->w && strspn(p->w, "*") == strlen(p->w) &&
                            p->m && strspn(p->m, "\001") == strlen(p->m))
                                return TRUE;
                return FALSE;
        }

I was also able to get rid of 'okay', which i found hard to follow.
The comment about the null string is gone too, because IMO the null
string fails for the same reason 'foo' fails. There's really no need
for an explicit check.

A patch is attached. Out of curiosity, why was the decision made to
use '\000' and '\001' instead of, say, '0' and '1'? For example in the
old code, we have "p->m[i] != 1", and it wouldn't be that hard to
write "p->m[i] != '1'" instead. I'm sure there's a reason, but I can't
guess it.

-- 
Written with 100% free software. Please support the following websites:
www.debian.org www.noamazon.com www.gnu.org www.opendvd.org lpf.ai.mit.edu

[-- Attachment #2: rc.patch --]
[-- Type: text/plain, Size: 837 bytes --]

--- glob.c.old	Wed Oct 21 07:01:46 1998
+++ glob.c	Mon Apr 17 15:55:30 2000
@@ -34,23 +34,13 @@
 
 extern bool lmatch(List *s, List *p) {
 	List *q;
-	int i;
-	bool okay;
 	if (s == NULL) {
 		if (p == NULL) /* null matches null */
 			return TRUE;
-		for (; p != NULL; p = p->n) { /* one or more stars match null */
-			if (*p->w != '\0') { /* the null string is a special case; it does *not* match () */
-				okay = TRUE;
-				for (i = 0; p->w[i] != '\0'; i++)
-					if (p->w[i] != '*' || p->m[i] != 1) {
-						okay = FALSE;
-						break;
-					}
-				if (okay)
-					return TRUE;
-			}
-		}
+		for (; p != NULL; p = p->n) /* one or more stars match null */
+			if (p->w && strspn(p->w, "*") == strlen(p->w) &&
+			    p->m && strspn(p->m, "\001") == strlen(p->m))
+				return TRUE;
 		return FALSE;
 	}
 	for (; s != NULL; s = s->n)