Undefined behavior with structs

[Matthew Gidden]

Hi all,

I'm seeing some undefined behavior with some structs I'm xdressing. My project is on github here. The oddities I'm seeing is that after a purely clean build (removing install and build dirs), I'm getting segfaults from the command line on a simple struct, correct construction with ipython on the same simply struct, and seg faults on both for a more complicated struct. 

Both my xdress and build commands work without error. 

My ipython calls follow below:

In [1]: from cyclopts.execute import SupplyParams
In [2]: params = SupplyParams()
In [3]: from cyclopts.execute import RequestParams
In [4]: other = RequestParams()
Segmentation fault (core dumped)

And my command line calls are:

19:51 ~$ python -c "from cyclopts.execute import SupplyParams; params = SupplyParams();"
Segmentation fault (core dumped)
19:51 ~$ python -c "from cyclopts.execute import RequestParams; params = RequestParams();"
Segmentation fault (core dumped)

Any help would be appreciated.

--

Matthew Gidden
Ph.D. Candidate, Nuclear Engineering

The University of Wisconsin -- Madison
Ph. 225.892.3192

[Matthew McCormick]

Heya Blackbeard,

What tool is being used to parse the C++?

If a default constructor is explicitly added to the C++ declaration,
does it fix the crash?

Cheers,
spidr

> --
> You received this message because you are subscribed to the Google Groups
> "xdress" group.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/xdress/CAHRdPFyfcDtq45QB3ShAXNURa5p0%3DKG6LqsWd184TSrNgm9KGw%40mail.gmail.com.

[Matthew Gidden]

Hi spidr!

On Wed, Apr 2, 2014 at 8:39 PM, Matthew McCormick <matthew.m...@gmail.com> wrote:

Heya Blackbeard,

What tool is being used to parse the C++?

gccxml 

If a default constructor is explicitly added to the C++ declaration,
does it fix the crash?

yes, it does. 

scopz helped me look through some possible issues last night, and we discovered that if you used a class declaration with all public members, seg faults go away. 

the core problem is that  for a struct, its __init__ looks like:

    def __init__(self, ):
        """RequestParams(self, )                                                                                                                      
        """
        self._inst = malloc(sizeof(cpp_execute.RequestParams))
        (<cpp_execute.RequestParams *> self._inst)[0] = cpp_execute.RequestParams()

But a class' __init__ looks like:

    def __init__(self, ):

        """RequestParams(self, )                                                                                                                      

        """

        self._inst = new cpp_execute.RequestParams()

Regards, 

bb

[Anthony Scopatz]

For the record, 

I think what is causing the issue is that structs for C++ in cython were being malloc'd rather than new'd.  I suppose that they should be new'd as well. I don't think that this will be a hard fix.  But we should also add a test case for this.

Be Well

Anthony

To view this discussion on the web visit https://groups.google.com/d/msgid/xdress/CAHRdPFw1_-3iJrs4OUsuHy%3DZKTL176_YdnnTGFtWiJyGfsWYHA%40mail.gmail.com.