Re: [PATCH] Implement non-trivial std::random_device::entropy (PR libstdc++/67578)

[Xi Ruoyao <ryxi at stu dot xidian dot edu dot cn>]

On 2017-05-22 14:00 +0100, Jonathan Wakely wrote:
> On 18/05/17 19:27 +0800, Xi Ruoyao wrote:
> > This patch use ioctl to get entropy of std::random_device using
> > /dev/random and /dev/urandom.
> 
> This is a nice addition, thanks.
> 
> N.B. I couldn't apply your patch, several lines had U+00A0 (i.e.
> NO-BREAK SPACE) characters where normal spaces should be. I don't know
> if this is something your mail client did, but maybe attaching the
> patch as a multipart message rather than including it inline in the
> body will prevent it in future.

Oh no...  Seems my email client destroyed my patch. I'll attach the
patch next time.  Please wait for PATCH v2.

> > diff --git a/libstdc++-v3/config/abi/pre/gnu.ver b/libstdc++-v3/config/abi/pre/gnu.ver
> > index 3e6e70b..39ad330 100644
> > --- a/libstdc++-v3/config/abi/pre/gnu.ver
> > +++ b/libstdc++-v3/config/abi/pre/gnu.ver
> > @@ -1963,6 +1963,9 @@ GLIBCXX_3.4.23 {
> >      _ZNSt13__future_base13_State_baseV211_Make_ready6_M_setEv;
> >  #endif
> >  
> > +    # std::random_device::_M_getentropy() const
> > +    _ZNKSt13random_device13_M_getentropyEv;
> > +
> >  } GLIBCXX_3.4.22;
> 
> The 3.4.23 version is now "closed" because it was used for the GCC 7.1
> release, and so we can't add new symbols to it. That means we need to
> bump the library version and create a new GLIBCXX_3.4.24 version node.
> The required steps are documented in doc/xml/manual/build_hacking.xml
> (and done by the attached patch which I'm testing).

I'll do it.  Glad to add the first symbol in 3.4.24.

> > @@ -161,6 +169,34 @@ namespace std _GLIBCXX_VISIBILITY(default)
> >      return _M_mt();
> >    }
> >  
> > +  double
> > +  random_device::_M_getentropy() const noexcept
> > +  {
> > +#if defined _GLIBCXX_HAVE_SYS_IOCTL_H && \
> > +    defined _GLIBCXX_HAVE_LINUX_RANDOM_H
> 
> Is RNDGETENTCNT guaranteed to be defined by all versions of
> <linux/random.h>?
> 
> Would it be better to check for it explicitly? i.e.
> 
> #if defined _GLIBCXX_HAVE_SYS_IOCTL_H && defined RNDGETENTCNT

It has been there since the git repo of Linux kernel was created.
I have to check the ancient history of kernel code to know.

> 
> It would also be nice to have a test for this functionality, but since
> the available entropy might be zero sometimes we can't guarantee that
> a test for a non-zero value will always pass.

I've thought about it, but I couldn't find out how to do the test.
-- 
Xi Ruoyao <ryxi@stu.xidian.edu.cn>
School of Aerospace Science and Technology, Xidian University