Re: set -e behaviour. Seems bug. Not the subshell issue!

[Eric Blake]

On 09/11/2012 09:24 AM, Марк Коренберг wrote:
>> This behavior conforms to POSIX.
>>
>>>
>>> Is any way to force "set -e" to work even in such cases?
>>
>> No, it is already working as specified (just not the way you want).
>>

> 1. Can you give me link (or name of) posix standard where I can read about
> this?

POSIX 2008 Technical Corrigendum 1 is not quite out yet (still in
balloting, but will be finalized later this year), but it will include
this wording:

http://www.austingroupbugs.net/view.php?id=52

> 2. Is there any workaround (like shopt) that will help for such case? I
> have read all shopts, man bash and found nothing.

There's nothing to work around.  'set -e' is behaving as required by
POSIX, which just happens to be a different behavior than what you want.
 There is no option in bash to get the behavior you want, unfortunately.

> 
> If it is the POSIX standard, this mean that I can not check return value of
> custom function at all, as this implies disabling of "set -e" inside
> functions bodies.

It doesn't disable 'set -e' inside ALL function bodies, but only inside
function bodies where the function call appears inside a context where
'set -e' is ignored (such as in the condition of 'if').  It's
non-intuitive, and therefore my advice is to never rely on 'set -e'.

-- 
Eric Blake   eblake@redhat.com    +1-919-301-3266
Libvirt virtualization library http://libvirt.org

signature.asc
Description: OpenPGP digital signature