Mike Frysinger wrote:
> On Monday 29 June 2009 04:54:52 Yang Zhang wrote:
>> Hi, I like using nounset for stricter scripts, but an annoyance is that
>> anytime I use "$@" and it's empty, I get an error, when (to me,
>> cognitively) it is not "unset" (as in someone *forgot* to set it), it's
>> just an empty, which is a common case (IIRC, in bash, variables set to
>> empty arrays and unset variables are the same).
>>
>> As a result I'm forced to use "${@:-}" or something like that everywhere
>> I use "$@" (which is really everywhere). Is there any other way around
>> this? Any way to get a more selective nounset? Thanks in advance.
>
> if you search the archives, i think the previous discussion on this topic
> said
> bash-4's behavior needed to change here
It's not quite that simple. And the discussion you reference took place
on the austin-group list, not bug-bash.
First of all, the bash-3.x behavior was inconsistent. $@ did not exit
the shell, but $...@} did.
When I got bug reports about the behavior, I took a look around. Posix.2
said that set -u applied only to "variables", which have a specific
definition and do not include $@, $* or any other special parameter. This
was so far from existing practice, I figured that it was another case of
the set -e issue: Posix.2 wanted to standardize existing practice, but
messed up. So I changed bash-4.0 to match existing shell practice, and
asked for an interpretation.
The interpretation process went back and forth, and we finally decided
that set -u will apply to references to every variable, parameter, and
special parameter *except* $@ and $*. That's going to be in the next
revision of the standard.
In the meantime, I have a patch in the pipeline that will implement the
behavior described in the Posix interpretation.
Chet
--
``The lyf so short, the craft so long to lerne.'' - Chaucer
Chet Ramey, ITS, CWRU c...@case.edu http://cnswww.cns.cwru.edu/~chet/
