[bug?] test adf -ge 0

[bug?] test adf -ge 0

[James William Pye]

[-- Attachment #1: Type: text/plain, Size: 584 bytes --]

Using 4.2.0, why does this return successful?

Both sh(freebsd) and bash echo an error message and return 2.


> flaw@void:~ % echo $ZSH_VERSION
> 4.2.0
> flaw@void:~ % test sdf -ge 0
> flaw@void:~ % echo $?
> 0
> flaw@void:~ % /bin/sh
> flaw@void $ test sdf -ge 0
> test: sdf: bad number
> flaw@void $ echo $?
> 2
> flaw@void:~ % bash
> bash-2.05b$ test sdf -ge 0
> bash: test: sdf: integer expression expected
> bash-2.05b$ echo $?
> 2

I know zsh is not bash or sh, but this is a fairly serious inconsistency, IMO.

-- 
Regards,
        James William Pye

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 479 bytes --]

Re: [bug?] test adf -ge 0

[Dan Nelson]

In the last episode (Aug 31), James William Pye said:
> Using 4.2.0, why does this return successful?

Probably because atoi("sdf") returns 0, so the test succeeds.
 
> Both sh(freebsd) and bash echo an error message and return 2.

/bin/sh on Tru64 5.1, AIX 5.2, and Solaris 9 all succeed.

> I know zsh is not bash or sh, but this is a fairly serious
> inconsistency, IMO.

It's probably in the range of "undefined behaviour".  I can't find any
manpages or standards docs that say that test must verify that numeric
arguments are really numbers.

-- 
	Dan Nelson
	dnelson@allantgroup.com

Re: [bug?] test adf -ge 0

[DervishD]

    Hi Dan & James :)

 * Dan Nelson <dnelson@allantgroup.com> dixit:
> In the last episode (Aug 31), James William Pye said:
> > Both sh(freebsd) and bash echo an error message and return 2.
> /bin/sh on Tru64 5.1, AIX 5.2, and Solaris 9 all succeed.

    But /bin/test in GNU coreutils fails too...
 
> > I know zsh is not bash or sh, but this is a fairly serious
> > inconsistency, IMO.
> It's probably in the range of "undefined behaviour".  I can't find any
> manpages or standards docs that say that test must verify that numeric
> arguments are really numbers.

    You're true, at least SUSv3 doesn't specify anything. I suppose
that's because atoi() behaviour...

    Raúl Núñez de Arenas Coronado

-- 
Linux Registered User 88736
http://www.pleyades.net & http://raul.pleyades.net/

Re: [bug?] test adf -ge 0

[Bart Schaefer]

On Tue, 31 Aug 2004, James William Pye wrote:

> Using 4.2.0, why does this return successful?
> 
> Both sh(freebsd) and bash echo an error message and return 2.
> 
> > flaw@void:~ % echo $ZSH_VERSION
> > 4.2.0
> > flaw@void:~ % test sdf -ge 0
> > flaw@void:~ % echo $?
> > 0

It's because of this:

zsh% sdf=-1
zsh% test sdf -ge 0
zsh% echo $?
1
zsh% sdf=1
zsh% test sdf -gt 0
zsh% echo $?
0

That is, zsh's builtin test is interpreting "sdf" as a variable name and 
using the corresponding value, just as (( sdf > 0 )) would do.  The value 
of an unset variable in math context is 0.

To forestall the inevitable question about what happens when the value of
the variable is not a number:

zsh% sdf=auq
zsh% auq=42
zsh% echo $(( sdf ))
42

If you wonder why _that_ works, consider this:

zsh% sdf='[#13]qua'
zsh% qua='6 * 9'
zsh% echo $(( sdf ))
13#42

Re: [bug?] test adf -ge 0

[James William Pye]

[-- Attachment #1: Type: text/plain, Size: 962 bytes --]

On Tue, 2004-08-31 at 12:11, Dan Nelson wrote:
> Probably because atoi("sdf") returns 0, so the test succeeds.
Indeed, without errno being set too, as you are probably aware.

FreeBSD(likely, NetBSD as well) uses strtol, which is how they catch the
error.

> /bin/sh on Tru64 5.1, AIX 5.2, and Solaris 9 all succeed.
Wow, just when I thought I was writing portable scripts! /;

> It's probably in the range of "undefined behaviour".  I can't find any
> manpages or standards docs that say that test must verify that numeric
> arguments are really numbers.
I looked at what OpenGroup.org had, but nothing regarding the subject
seemed to be specified.

http://www.opengroup.org/onlinepubs/009695399/utilities/test.html

This doesn't match up to the version FreeBSD's test is said to comply
to, so I don't know for sure if FreeBSD's test is complying specified
behaviour or just historic on BSDs. =\

-- 
Regards,
        James William Pye

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 479 bytes --]