[Rd] C++ debugging help needed
Duncan Murdoch
murdoch.duncan at gmail.com
Mon Oct 7 18:48:38 CEST 2013
On 07/10/2013 12:18 PM, Martyn Plummer wrote:
> Yes, on reflection it's an ABI problem on Linux (use of PIC code in
> shared libraries means that any symbol can be interposed). Using
> namespaces isn't really the answer because that's an API issue. I think
> what you really need to do is control the visibility of your classes and
> functions so that everything is hidden except for the entry points you
> call from R (Writing R Extensions section 6.15). This should stop the
> symbol collision because hidden functions are resolved inside the shared
> object instead of going through a lookup table that can be overwritten
> by someone else's package.
That's not possible on all platforms, but on platforms that allow it,
rgl (on R-forge) is now doing it. It only exposes R_init_rgl(), which
registers all the other entry points.
I think the work that I did to add the namespace was also worthwhile for
those platforms that don't allow you to hide things; it makes collisions
less likely, though not impossible.
For those unfamiliar with this: it is a little tedious to add external
registration, but you do gain extra checks on your calls. Once you have
that it is usually very easy to hide everything except the registration
function. (One exception: if your package has its own configuration
script, it's a bit more work.)
Duncan Murdoch
>
> Martyn
>
>
>
> On Fri, 2013-10-04 at 15:05 -0400, Duncan Murdoch wrote:
> > I have now got two "solutions" to this. The rgl version currently on
> > CRAN does a simple rename to avoid the name clash. A later version,
> > still only on R-forge, puts most objects into a namespace called "rgl".
> > (The old code had two small namespaces "gui" and "lib"; they are gone now.)
> >
> > I am not yet confident that the current version with namespaces will
> > compile on all platforms; it seems much more fragile this way, with
> > errors showing up on Linux that were not errors on Windows. (E.g.
> > sometimes I included a header with declarations in the rgl namespace
> > followed by system header files, and the latter acted differently than
> > they did when placed before the rgl header file, apparently declaring
> > the system functions to be in a new anonymous namespace.)
> >
> > rgl also includes some C code from the gl2ps project and some C++ code
> > from FTGL; I didn't put those into the rgl namespace. So there are
> > still possibilities for clashes if anyone else uses those.
> >
> > I'm still surprised that anything with plugins works on Unix-alike
> > systems with such bizarre linking rules. This is one of those few cases
> > where the Windows design seems clearly superior.
> >
> > Duncan Murdoch
> >
> >
> > On 02/10/2013 10:50 AM, Duncan Murdoch wrote:
> > > I've had reports lately about segfaults in the rgl package. I've only
> > > been able to reproduce these on Linux. I am not so familiar with C++
> > > details, so I have a couple of questions way down below. But first some
> > > background info.
> > >
> > > One recipe to recreate the crash works with a new version 5.0-1 of the
> > > mixOmics package:
> > >
> > > > library(mixOmics)
> > > > example(pca)
> > >
> > > This crashes with messages like this:
> > >
> > > Program received signal SIGSEGV, Segmentation fault.
> > > 0x00007ffff28aafd9 in __exchange_and_add (__mem=0x7f7fffff7f7ffff7,
> > > __val=<optimized out>) at /usr/include/c++/4.7/ext/atomicity.h:48
> > > 48 { return __atomic_fetch_add(__mem, __val, __ATOMIC_ACQ_REL); }
> > >
> > > The call stack ends with this:
> > >
> > > #0 0x00007ffff28aafd9 in __exchange_and_add (__mem=0x7f7fffff7f7ffff7,
> > > __val=<optimized out>) at /usr/include/c++/4.7/ext/atomicity.h:48
> > > #1 __exchange_and_add_dispatch (__mem=0x7f7fffff7f7ffff7,
> > > __val=<optimized out>) at /usr/include/c++/4.7/ext/atomicity.h:81
> > > #2 _M_dispose (__a=..., this=0x7f7fffff7f7fffe7)
> > > at /usr/include/c++/4.7/bits/basic_string.h:242
> > > #3 ~basic_string (this=0x15f8770, __in_chrg=<optimized out>)
> > > at /usr/include/c++/4.7/bits/basic_string.h:536
> > > #4 Shape::~Shape (this=0x15f8760, __in_chrg=<optimized out>) at
> > > Shape.cpp:13
> > > #5 0x00007ffff22df50b in ~Background (this=0x15f8760,
> > > __in_chrg=<optimized out>) at Background.hpp:15
> > > #6 Background::~Background (this=0x15f8760, __in_chrg=<optimized out>)
> > > at Background.hpp:15
> > >
> > > Up to entry #4 this all looks normal. If I go into that stack frame, I
> > > see this:
> > >
> > >
> > > (gdb) up
> > > #4 Shape::~Shape (this=0x15f8760, __in_chrg=<optimized out>) at
> > > Shape.cpp:13
> > > warning: Source file is more recent than executable.
> > > 13 blended(in_material.isTransparent())
> > > (gdb) p this
> > > $9 = (Shape * const) 0x15f8760
> > > (gdb) p *this
> > > $10 = {_vptr.Shape = 0x7ffff2d8e290, mName = 6, mType = {
> > > static npos = <optimized out>,
> > > _M_dataplus = {<std::allocator<char>> =
> > > {<__gnu_cxx::new_allocator<char>> =
> > > {<No data fields>}, <No data fields>},
> > > _M_p = 0x7f7fffff7f7fffff <Address 0x7f7fffff7f7fffff out of
> > > bounds>}},
> > > mShapeColor = {mRed = -1.4044474254567505e+306,
> > > mGreen = -1.4044477603031902e+306, mBlue = 4.24399170841135e-314,
> > > mTransparent = 0}, mSpecularReflectivity = 0.0078125,
> > > mSpecularSize = 1065353216, mDiffuseReflectivity = 0.007812501848093234,
> > > mAmbientReflectivity = 0}
> > >
> > > The things displayed in *this are all wrong. Those field names come
> > > from the Shape object in the igraph package, not the Shape object in the
> > > rgl package. The mixOmics package uses both.
> > >
> > > My questions:
> > >
> > > - Has my code somehow got mixed up with the igraph code, so I really do
> > > have a call out to igraph's Shape::~Shape instead of rgl's
> > > Shape::~Shape, or is this just bad info being given to me by gdb?
> > >
> > > - If I really do have calls to the wrong destructor in there, how do I
> > > avoid this?
> > >
> > > Duncan Murdoch
> >
> > ______________________________________________
> > R-devel at r-project.org mailing list
> > https://stat.ethz.ch/mailman/listinfo/r-devel
>
> -----------------------------------------------------------------------
> This message and its attachments are strictly confiden...{{dropped:8}}
More information about the R-devel
mailing list