Re: [Qemu-devel] [PATCH 1/7] iotests: Allow out-of-tree run

[Max Reitz <mreitz@redhat.com>]

On 16.05.2014 17:09, Eric Blake wrote:
> On 05/16/2014 08:43 AM, Max Reitz wrote:
>
>>>> +    source_iotests="$(cd "$(dirname "$(readlink "$0")")"; pwd)"
>>> This is potentially dangerous. If readlink or dirname fails, you can
>>> invoke cd "" (which on bash is stupidly a no-op instead of an error),
>>> and end up calling pwd in the wrong directory.  But in the common case
>>> it works, so I'm not sure it's worth bending over backwards to make it
>>> more robust.
>> I guess using something like
>>
>> source_iotests="$(dirname "$(readlink "$0")")"; if [ -z "$source_iotests" ]; then; /* abort */; fi;
>> source_iotests="$(cd "$dirname"; pwd)"
>>
>> should work better, then?
> Or even safer with
>
> source_iotests=$(cd "$dirname" && pwd) || /* abort */
>
> in the second step, to ensure both the cd and pwd commands succeeded.
> (By the way, assignment context does not require "" when passing a
> single shell word, such as command substitution, so foo=$(...) and
> foo="$(...)" are identical)

The more you know... (and of course I meant "$source_iotests" in the 
second step)

>>> Is [ ! '(' -o ')' ] true or false? Depends on whether it was parsed as {
>>> ! '(' } -o ')' (false -o true => true) or as ! { '(' -o ')' } (! (true
>>> -o true) => false)
>>>
>>> But this is bash, so you could do:
>>>
>>> if [[ $arch && -x $build_root/$arch-softmmu/qemu-system-$arch ]]
>>>
>>> for less typing, and no risk of [] ambiguity.
>> If you're telling me I'm free to use bashisms, I'll believe you. :-)
> Well, the script IS being run by /bin/bash, and you already ARE using
> bashisms elsewhere.

I feared so. *g*

Max