Re: [Valgrind-users] [drd] race condition in std::istringstream / std::getline

[Jorge M. <jor...@gm...>]

On Thu, May 27, 2010 at 3:42 AM, Bart Van Assche <bva...@ac...> wrote:
> On Thu, May 27, 2010 at 4:29 AM, Jorge Moraleda
> <jor...@gm...> wrote:
>>
>> Dear all,
>>
>> Am I doing something wrong, or is is there a race condition somewhere
>> in std::istringstream and/or std::getline?
>>
>> (Bart, Konstantin, this ended up being the error that I've been trying
>> to track down for a while and that I thought --erroneously-- was
>> caused by std::locale)
>>
>> This program:
>>
>> /// file: thread.cpp
>> #include <pthread.h>
>> #include <string>
>> #include <sstream>
>>
>> void *threadEntry(void *threadid)
>> {
>>
>>         std::string bar("01\n23");
>>         std::istringstream content(bar);
>>
>>                std::string line;
>>                std::getline(content,line);
>>
>>    pthread_exit(NULL);
>> }
>>
>> int main (int argc, char *argv[])
>> {
>>  pthread_t threads[2];
>>  int rc;
>>  long t;
>>
>>        std::ostringstream dummy;
>>        dummy << 0;
>>        for(t=0; t<2; t++) {
>>                rc = pthread_create(&threads[t], NULL, threadEntry, (void *)t);
>>        }
>>        pthread_exit(NULL);
>> }
>>
>> when compiled using g++ -g -pthread thread.cpp
>> and run with valgrind --tool=drd ./a.out
>>
>> produces the following output:
>>
>> ==10635== drd, a thread error detector
>> ==10635== Copyright (C) 2006-2009, and GNU GPL'd, by Bart Van Assche.
>> ==10635== Using Valgrind-3.5.0-Debian and LibVEX; rerun with -h for
>> copyright info
>> ==10635== Command: ./a.out
>> ==10635==
>> ==10635== Thread 3:
>> ==10635== Conflicting load by thread 3 at 0x05134160 size 8
>> ==10635==    at 0x4EBBC2A: std::basic_istream<char,
>> std::char_traits<char> >& std::getline<char, std::char_traits<char>,
>> std::allocator<char> >(std::basic_istream<char, std::char_traits<char>
>> >&, std::basic_string<char, std::char_traits<char>,
>> std::allocator<char> >&, char) (in /usr/lib/libstdc++.so.6.0.13)
>> ==10635==    by 0x400D03: threadEntry(void*) (threads2.cpp:12)
>> ==10635==    by 0x4C32870: vgDrd_thread_wrapper (drd_pthread_intercepts.c:272)
>> ==10635==    by 0x55E8A03: start_thread (pthread_create.c:300)
>> ==10635==    by 0x58DD80C: clone (clone.S:112)
>> [ ... ]

Bart, thank you for all your help.

> Strange. I have tried to reproduce this with gcc 4.4.4 (which includes
> libstdc++.so.6.0.13), but did not see any race reports.

I observe the same thing you do. My original report was using gcc
4.4.1. When I compile the test program with 4.4.4 I don't see the race
reports either. However, my real program (which is not a small test
that I can submit) does produce race conditions even when compiled
with gcc 4.4.4. The race conditions are in different places every
time, and never in the stringstream/getline code, but the "other
segment" contains in all cases a block of code that looks essentially
the same as the one in the test case. I wrote a custom alternative not
using stringstreams or getline in that block of code, and all the race
conditions are gone, so this problem is solved for me, but I still
would like to find out what the problem was. I have tried to refine
the test program so that it would reproduce the race conditions, but I
haven't been able to.

>
> Does inserting two pthread_join() calls between pthread_create() and
> pthread_exit() in main() help ?

I tried this in the gcc 4.4.1 system and it does not help.
>
> Bart.
>