Re: [Boost-Users] Problems building Boost 1.27.0 libraries

From: George A. Heintzelman (georgeh_at_[hidden])
Date: 2002-03-22 09:47:41

> That's a bug that's been partly fixed in the latest cvs source -
> the -pthread flag should be -pthreads.
> 
> However that's not the end of it, because the std lib that ships with gcc on
> solaris has a problem with std::runtime_error when you build with -pthreads
> (basically the code doesn't link because the body of
> runtime_error::runtime_error is missing in the headers), I don't know what
> the solution to this is, other than to fix your std lib (or maybe use
> STLPort).
>>
>> I have downloaded Boost and am trying to build the libraries on 
Solaris
>> 8 with g++ 2.95.3:
>>

I ran into this problem just recently. I think John's diagnosis is 
incorrect. In fact, stdexcept includes the body of runtime_error just 
fine, and the class runtime_error exists.

Part of the problem is that the gcc standard library, when compiled 
under -pthreads, typedefs std::string (and all the other default 
std::containers) differently from how it is defined under 
single-threaded compilation. The default allocator is defined to be:

typedef __default_alloc_template<__NODE_ALLOCATOR_THREADS, 0> alloc;

where __NODE_ALLOCATOR_THREADS is a #define which is true if threading 
is enabled, false otherwise.

This in itself is still not the real problem. The problem is that the 
constructor of runtime_error, and all the other stdexcept classes, is 
not actually compiled under pthreads, because stdexcept has a #pragma 
interface "stdexcept" line in it which inhibits the compiler from 
emitting the bodies of these inline functions. Which it also does in 
non-threaded compiling mode, but in that mode the gcc writers have 
conveniently put the non-threaded definitions into the libstdc++, so 
your link works fine.

I think there is a workaround: put

#pragma implementation "stdexcept"

in one of of your source files which includes <stdexcept>. This will 
cause those bodies to be emitted and your link should go fine.

I haven't tested it myself for various local technical reasons, but I 
think it will work.

Good luck, and let me know if it works! If it does, we should maybe 
propose that the boost Threads library include such definitions under 
gcc.

George Heintzelman
georgeh_at_[hidden]