qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-devel] [PATCH v2 4/5] block: qemu-iotests - fix image cleanup


From: Eric Blake
Subject: Re: [Qemu-devel] [PATCH v2 4/5] block: qemu-iotests - fix image cleanup when using spaced pathnames
Date: Thu, 10 Apr 2014 08:43:44 -0600
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:24.0) Gecko/20100101 Thunderbird/24.4.0

On 04/10/2014 06:53 AM, Jeff Cody wrote:

>>> +++ b/tests/qemu-iotests/common.rc
>>> @@ -178,10 +178,10 @@ _rm_test_img()
>>>      local img=$1
>>
>> Since we are quoting $img, should we quote $1 as well?
>>
> 
> I believe not, because variable assignment won't undergo all the shell
> expansions.  Notably, in variable assignment word splitting is not
> performed on the parameter expansion on the argument immediately to the
> right of the '='.  Quote removal, however, will still be performed.  So
> img=$1 and img="$1" are identical once processed.

Ooh, tricky.

You are correct that in isolation:

img=$1
img="$1"

are semantically identical, no matter what $1 contains, across ALL
shells.  However, that's not the code you wrote above.

local img=$1

is not POSIX (yet - although there has been some effort in the POSIX
working group to standardize some form of local variables while still
allowing for the fact that bash and ksh implemented scoping of local
variables differently).  But 'local' is similar to 'export'; and observe
the difference when using 'export' between dash and bash:

$ dash -c 'set "a  b"; export a=$1 b="$1"; echo "$a.$b"'
a.a  b
$ bash -c 'set "a  b"; export a=$1 b="$1"; echo "$a.$b"'
a  b.a  b

Here, the shell word a=$1 is semantically NOT a raw assignment, but
rather an argument to 'export', and arguments DO undergo word splitting
in the current wording of POSIX.  There was a recent bug report stating
that the dash behavior (which is strictly POSIX) is undesirable, and
that the bash/ksh behavior of export, while not strictly compliant with
the POSIX 2008 wording, is nicer; so the next version of POSIX will be
amended to add a definition of a 'declaration utility' which can
evaluate (some) arguments in assignment context.  'export' is one such
declaration utility, 'local' (if it gets standardized) would be another:

http://austingroupbugs.net/view.php?id=351

But even with the notion of an assignment-context argument added to a
future version of POSIX, the reality is that given the present standard,
it's safer to either use "" to ensure no word splitting:

local img="$1"

or to rewrite things across two statements to avoid relying on whether
assignment-context arguments work the way you want:

local img
img=$1

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

Attachment: signature.asc
Description: OpenPGP digital signature


reply via email to

[Prev in Thread] Current Thread [Next in Thread]