Date: Tue, 27 Jul 2004 12:36:48 -0400 From: Sergey Babkin <babkin@bellatlantic.net> To: Giorgos Keramidas <keramida@ceid.upatras.gr> Cc: freebsd-arch@freebsd.org Subject: Re: some PRs Message-ID: <410684A0.5C65ED42@bellatlantic.net> References: <20040718184008.GC57678@darkness.comp.waw.pl> <20040719075952.GG57678@darkness.comp.waw.pl> <20040719.081356.51946167.imp@bsdimp.com> <200407191855.19885.max@love2party.net> <4105987E.5FC50517@bellatlantic.net> <20040727032253.GA24778@gothmog.gr> <20040727032917.GA24942@gothmog.gr>
next in thread | previous in thread | raw e-mail | index | archive | help
Giorgos Keramidas wrote:
>
> On 2004-07-26 19:49, Sergey Babkin <babkin@bellatlantic.net> wrote:
> > Max Laier wrote:
> > > The question to me is, do we really want to support (read fertilize)
> > > such a stupid thing? Given the chance that once we do support it
> > > people will use it. In my opinion it is bad to integrate something
> > > into base that we agree is nothing one should ever have created (at
> > > least that's my reading of the thread so far). I see no user-pessure
> > > for this.
> >
> > I'm about a week behind :-) but here are my 2 cents: it's a VERY
> > useful device for testing. Not checking the error code of write(),
> > printf() and such is a typical bug, so making it easy to detect by
> > switching the output to /dev/full (or creating a symlink to it) is a
> > very good idea. Like this:
> >
> > yourprogram >/dev/full \
> > && echo "The program does not check for success of write()"
>
> If a program doesn't check the return code of write() but merrily goes
> on doing other stuff or even terminates with a zero return value, how
> will the redirection affect its operation? I think it won't, as shown
> in the test below (run on a Linux machine):
If you run a test in which you know the program must fail (such
as writing to /dev/full) yet it does not, this means that there
is abug in the program.
> : $ ls -ld /dev/full
> : crw-rw-rw- 1 root root 1, 7 Jun 14 00:24 /dev/full
> : $ cat -n lala.c
> : 1 #include <sys/types.h>
> : 2 #include <string.h>
> : 3 #include <unistd.h>
> : 4
> : 5 int
> : 6 main(void)
> : 7 {
> : 8 char buf[] = "hello world\n";
> : 9 size_t len;
> : 10
> : 11 len = strlen(buf);
> : 12 write(1, buf, len);
> : 13 return 0;
> : 14 }
> : $ cc -O -W -Wall -o lala lala.c
> : $ ./lala
> : hello world
> : $ ./lala >/dev/full
> : $ echo $?
> : 0
> : $
>
> The fact that /dev/full was used as the output device didn't reveal the
> potential write() problem.
That's _exactly my point: if the program writes to /dev/full
and yet does produce an error exit code or an error message,
there is a bug in the program.
> I must have misunderstood something. How do you mean that we could use
> /dev/full for testing?
Well, as described above: for each file that your program can produce,
try to substitute it with /dev/full and watch the prgoram fail.
If it does not fail, there is a bug. That's much easier than producing
an actual full filesystem.
-SB
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?410684A0.5C65ED42>