On Fri, Nov 17, 2017 at 04:45:27PM +0300, Nikolay Shirokovskiy wrote:
On 17.11.2017 16:40, Daniel P. Berrange wrote:
> On Fri, Nov 17, 2017 at 04:31:13PM +0300, Nikolay Shirokovskiy wrote:
>>
>>
>> On 17.11.2017 16:24, Daniel P. Berrange wrote:
>>> On Fri, Nov 17, 2017 at 04:17:37PM +0300, Nikolay Shirokovskiy wrote:
>>>> If one of the libraries is compiled with tcmalloc then
>>>> the latter will add GLIBCPP_FORCE_NEW and GLIBCXX_FORCE_NEW to
>>>> environment at startup and thus break commandtest.
>>>
>>> How are they getting those envs into our environment after we clean it
>>> out ? We strongly aim to prevent any non-whitelisted env variable
>>> leakage into children we spawn, so I would really like to kill these
>>> env vars instead of changin the test.
>>
>> They inserted at process startup I guess [1]. They are cleared out by
commandtest
>> but visible in commandhelper.
>
> Hmm, so is comandhelper getting linked to tcmalloc by mistake then ?
> If so, how easy is it to stop it being linked
It is not liked directly. In my case the chain is
libdevmapper.so -> libudev.so -> libtcmalloc.so. It is distro specific
but I guess other can step across this issue and for a different chain. One
just need to link on of the libraries libvirt uses to tcmalloc.
Ah I see. I think this smells like a bug in the tests/Makefile.am
The commandhelper binary should not link to anything at all except
for libc (and perhaps gnulib, but possibly even that is redundant)
Regards,
Daniel
--
|:
https://berrange.com -o-
https://www.flickr.com/photos/dberrange :|
|:
https://libvirt.org -o-
https://fstop138.berrange.com :|
|:
https://entangle-photo.org -o-
https://www.instagram.com/dberrange :|