>Number:         35408
>Category:       bin
>Synopsis:       bin/cat »wastes« memory when output blocksize < BUFSIZ
>Confidential:   no
>Severity:       non-critical
>Priority:       low
>Responsible:    bin-bug-people
>State:          open
>Class:          change-request
>Submitter-Id:   net
>Arrival-Date:   Thu Jan 11 21:30:00 +0000 2007
>Originator:     Andreas Wiese
>Release:        NetBSD 4.99.8
>Organization:
	BSD-Crew Dresden
>Environment:
System: NetBSD schroeder.lan.instandbesetzt.net 4.99.8 NetBSD 4.99.8 (SCHROEDER) #0: Thu Jan 11 21:37:15 CET 2007 root@schroeder.lan.instandbesetzt.net:/usr/obj/sys/arch/i386/compile/SCHROEDER i386
Architecture: i386
Machine: i386
>Description:

I was bored and read some sources.  If in raw-mode, cat(1) does check
the blocksize of it's output, and allocates a buffer via malloc(3) with
at least BUFSIZ bytes.  But for the case malloc(3) fails, we also have a
static buffer 'static char *fb_buf[BUFSIZ]'.  Why not using this one if
we otherwise would only allocate the same size (when blocksize is not
greater than BUFSIZE)?

Yes, this are only 1024 bytes but I don't see the sense in
allocating this memory twice.  And both fp_buf and buf should be on the
heap, so where's the difference (is there any)?

>How-To-Repeat:
n/a
>Fix:
diff -ur cat.o/cat.c cat/cat.c
--- cat.o/cat.c	2007-01-09 22:24:52.000000000 +0100
+++ cat/cat.c	2007-01-09 22:32:56.000000000 +0100
@@ -295,14 +295,13 @@
 
 	wfd = fileno(stdout);
 	if (buf == NULL) {
-		if (fstat(wfd, &sbuf) == 0) {
-			bsize = sbuf.st_blksize > BUFSIZ ?
-			    sbuf.st_blksize : BUFSIZ;
+		if (fstat(wfd, &sbuf) == 0 && sbuf.st_blksize > BUFSIZ) {
+			bsize = sbuf.st_blksize;
 			buf = malloc(bsize);
 		}
 		if (buf == NULL) {
-			buf = fb_buf;
 			bsize = BUFSIZ;
+			buf = fb_buf;
 		}
 	}
 	while ((nr = read(rfd, buf, bsize)) > 0)

HAND & LG -- aw
np: Kaizers Orchestra (Ompa til du dør) -- 04. 170
-- 
Nicht die Schönheit entscheidet, wen wir lieben, sondern
die Liebe entscheidet, wen wir schön finden.