[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Writing symbols files for C++ libraries.



On Tue, Apr 07, 2009 at 01:22:51AM +0200, Daniel Kobras <kobras@debian.org> wrote:
> Hi!
> 
> I'm currently fighting with deb symbols files for a C++ library I'm
> packaging, and I'd like to get some insight in how others are coping
> with this task. In particular, I wonder how to get rid of symbols from
> standard template instances that leak into the ABI, eg.
> 
> $ cat test.cpp 
> #include <string>
> 
> std::string foome(const std::string& bar)
> {
> 	return "foo" + bar + "foo";
> }
> 
> $ g++ -fPIC -c test.cpp
> $ readelf -Wds test.o | grep -v UND | grep -v HIDDEN | grep -E
> '(FUNC|OBJECT)'
>     18: 0000000000000000    23 FUNC    WEAK   DEFAULT    9 _ZNSt11char_traitsIcE6lengthEPKc
>     22: 0000000000000000   155 FUNC    WEAK   DEFAULT   11 _ZStplIcSt11char_traitsIcESaIcEESbIT_T0_T1_EPKS3_RKS6_
>     30: 0000000000000000   104 FUNC    WEAK   DEFAULT   14 _ZStplIcSt11char_traitsIcESaIcEESbIT_T0_T1_ERKS6_PKS3_
>     33: 0000000000000000   124 FUNC    GLOBAL DEFAULT    5 _Z5foomeRKSs
> 
> I can hide the operator+() and related instances from with a version
> script, of course, but due to C++ name mangling they tend to be
> arch-specific and thus hard to maintain. Does anyone know of a more
> lightweight solution to limit visibility of template instances and
> ensure a clean ABI for C++ libraries? What are the current best
> practises when writing symbols files for C++ libraries, or is it simply
> not recommended?

Unfortunately, gcc upstream consider this as a feature, and not a bug:
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36022

I found nothing better than using a version script. I'm lucky that the
library in question (WebKit) only really exports C symbols, and C++ is only
internal details, so I'm filtering everything that starts with _Z, now.

Mike


Reply to: