On 05/27/2016 09:22 AM, Ján Tomko wrote:
On Fri, May 27, 2016 at 08:24:36AM -0600, Eric Blake wrote:
> On 05/27/2016 04:33 AM, Michal Privoznik wrote:
>
>>>
>>> Pushed under gnulib maintenance rule, since RC1 still hasn't been
tagged.
>>>
>>
>> Unfortunately, this breaks build on my system:
I have bisected it to:
54615b95ff238e235e806855efc46a9abad09f2e is the first bad commit
commit 54615b95ff238e235e806855efc46a9abad09f2e
Author: Paul Eggert <eggert(a)cs.ucla.edu>
Date: Sat Feb 6 18:11:48 2016 -0800
misc: port better to gcc -fsanitize=address
Thanks; I independently came to the same conclusion, and have pushed the
new gnulib submodule.
char *name2 = realpath ("conftest.d//./..",
NULL);
- if (! (name1 && name2 && strcmp (name1, name2) != 0))
+ if ((name1 && name2 && strcmp (name1, name2) != 0))
Not quite. Paul's (buggy) fix was trying to ensure that strcmp() can't
be called on NULL, but the correct fix is:
if (! (name1 && name2) && strcmp (name1, name2) != 0)
or, by deMorgan's law,
if (! name1 || ! name2 || strcmp (name1, name2))
--
Eric Blake eblake redhat com +1-919-301-3266
Libvirt virtualization library
http://libvirt.org