This is the mail archive of the gcc-help@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: g++ 4.7 with g++0x threads & profiling


>> I have a C++ application using the new std::thread functionality.
>> At random I see all kinds of crashes, mostly in std::string but also
>> in std::map and friends.
>> I have profiling (-pg) enabled. It seems that if I disable profiling,
>> also the random errors (SIGSEGV) disappear.
>> Are there any known issues with profiling, c++0x and gcc version 4.7
>> with this? (tried g++ 4.6 as well)
>
> Neither std::map, nor std::string, is thread-safe, AFAIK. Individual
> instances of these, and pretty much all other classes, must be appropriately
> protected and accessed in a safe thread context, using std::mutex or a
> reasonable facsimile. You have not mentioned whether or not you may be
> potentially accessing same instances of std::string or std::map from
> concurrent threads, and whether you're employing std::mutex to do so.

That's all behind mutexes so that should be fine.

> Profiling modifies runtime behavior, by its inherent nature. If there's
> ill-formed, non-deterministic behavior, such as thread-unsafe access; it
> would not be unusual for this to manifest itself with more or less
> frequency, either in presence or in the absence of, runtime profiling.

hmmm ok.
So the profiling code is threadsafe.

--
www.vanheusden.com

bitcoin account: 14ExronPRN44urf4jqPMyoAN46T75MKGgP
msn address: spam@vanheusden.com


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]