From Bugzilla Helper: User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:0.9.3) Gecko/20010801 Description of problem: Using the ostrstream class in the c++ standard library causes a memory leak. If a program uses a lot of them, then all the computer's memory can be swallowed. I'll include a test program to demonstrate. How reproducible: Always Steps to Reproduce: 1. Make program using ostrstream 2. Use the str() method to get a const char* to the string 3. Destructor doesn't clear memory Additional info: Test program (eg test.cc): #include <iostream> #include <strstream> int main() { while(1) { std::ostrstream o; o << "Hello, world!" << '\0'; std::cout << o.str(); } } Compile with g++ test.cc Run a.out Watch memory usage of program increase. This is a serious bug for c++ developers. There's a workaround. Add o.rdbuf()->freeze(0); before the destructor is called.
Created attachment 26653 [details] Patch to gcc sources to fix problem (untested)
Created attachment 26654 [details] previous patch was reversed
Created attachment 26655 [details] third time lucky!!!
I think libstdc++ is right: http://www.cygnus.com/pubs/gnupro-97r1/4_GNUPro_Libraries/c_The_GNU_C++_Iostream_Library/libio.html sais in ostrstream::str () description that it Implies ostrstream::freeze(). This means that after o.str() the string is managed by user who is responsible for freeing it, unless o.freeze(0) is called. You could e.g. stick the pointer returned by o.str() into global variable and use it far after o has been destructed. Note that g++ 3.0 works exactly the same way with your example, it leaks memory too. But it is because of missing o.freeze(0) in your testcase.
Interesting, it seems there's some debate whether this is a bug or not. For example, in the following link Compaq "fixed" their library to free the memory: http://www.physik.rwth-aachen.de/group/IIIphys/compute/cplusplus/cxx611_release.html I think however, it's not a bug, after doing some research. The solution, I suppose is to use ostringstream, but that's only a recent addition to the standard :-(
My reading of the link you provided is that they weren't freeing the string at all. ISO C++ in [depr.strstreambuf] sais clearly that the string should be deleted only if strmode & allocated != 0 and strmode & frozen == 0. In your testcase, strmode & frozen != 0, so it should not be deleted.