Re: [Boost-users] [phoenix] segmentation fault with valarray

Subject: Re: [Boost-users] [phoenix] segmentation fault with valarray
From: Eric Niebler (eric_at_[hidden])
Date: 2013-01-29 16:36:47

On 01/29/2013 09:31 AM, Jeffrey Lee Hellrung, Jr. wrote:
> On Tue, Jan 29, 2013 at 5:35 AM, Philipp Schwaha  wrote:
> 
>>     Hi all!
>> 
>>     I have worked with phoenix together with valarray and am currently
>>     stumbling over a case of segmentation fault.
>> 
>>     I construct the following phoenix expression
>> 
>>     boost::phoenix::at_c<0>(arg1) += boost::phoenix::at_c<1>(arg1) /
>>     val(mass) * arg2
>> 
>>     I then feed a std::tuple<std::valarray<double>, std::valarray<double>>
>>     to it. It compiles nicely with gcc (4.6 & 4.7) using boost 1.52 as well
>>     as trunk. The resulting executable then proceeds to segfault, however. I
>>     have put together a small example (c++11) demonstrating my problem.
>> 
>>     The valarray's memory should already be reserved from original
>>     initialization. Performing the same operations manually with just the
>>     valarray not in phoenix seems to work fine.
>>     Am I missing something or doing something stupid here?
>> 
>>     thanks for any help!
>> 
>>     cheers
>>             Philipp
> 
> 
> I don't know the internals of Phoenix intimately, but it *might* hold
> subexpressions by reference rather than by value, implying that you'd
> have dangling references when grabbing "mass" upon applying your Phoenix
> function object. Just a guess if you wanted something to look into until
> someone more knowledgeable replies.

I don't believe that's the case, Jeff, but I'm not the Phoenix
implementor. Philipp, I'd suggest you file a bug.

Eric