Re: About weird behavior with boost.asio 1.63 on linux 2.6.15

From: Andrey Semashev (andrey.semashev_at_[hidden])
Date: 2023-12-20 18:22:38

On 12/20/23 18:27, Scott Zhang via Boost wrote:
> Dear All:
>     I have been using boost over 10+ years. The asio library work
> quite well along with document.
> Today I get a problem:
> My development environment:
> gcc 4.7.3 OABI
> Linux 2.6.15
> boost 1.63
> 
> The code looks like below:
> void myTask(const char *text)
> {
>     sleep(5);
>     /*for(;;)
>     {
>         //sleep(1);
>     }*/
>         cout<<"text is "<<text<<endl;
>   //  cout<<"out"<<endl;
> }
> boost::asio::io_service ioService;
> ioService.post(boost::bind(myTask,"hello"));
> ioService.run();
> 
> I get 2 problems:
> 1. If I run program as above, the mytask get executed correctly, but
> as soon as it returns, the program get "illegal instruction", and
> program dies.
> If I call myTask directly, then everything is ok.
> So what's wrong with io_service.post
> 
> 2. If I run ioservice within thread using
> threadpool.create_thread(
>                     boost::bind(&boost::asio::io_service::run, &ioService)
> Then after myTask executed, I get "Killed", using strace says program
> get SIGKILL.
> 
> As the test code is so simple, I can't tell where goes wrong.
> 
> I also noticed, if I don't call io_service.post, everything looks ok.
> And the threads, in linux ps, shows as many processes.
> 
> I reread the document of asio 1.63, it says had beed tested against
> linux 2.4 and 2.6 kernel. so it should be ok.

Were you able to collect the stacktrace of the crash? You should be able
to do it by running your code under gdb.

Does this reproduce with optimizations disabled?