pthread question

rotis23 · 03-31-2004, 04:17 AM

Hi All,

Not sure where to put pthread_attr_init and setdetachstate. Should they be called once for many threads as below or each instance of a new thread:

Code:

res = pthread_attr_init(&thread_attr);
                        if (res != 0)
                        {
                                syslog(LOG_ERR,":Error - Server thread error: %s",strerror(res));
                                return 5;
                        }
                        res = pthread_attr_setdetachstate(&thread_attr,1);//PTHREAD_CREATE_DETACHED);
                        if (res != 0)
                        {
                                syslog(LOG_ERR,":Error - Server thread error: %s",strerror(res));
                                return 6;
                        }

                        for(;;)
                        {
                                cli_size = sizeof(cli);
                                client = accept(s, (struct sockaddr *)&cli, &cli_size);
                                if (client == -1)
                                {
                                        syslog(LOG_ERR,":Error - Server thread error: %s",strerror(client));
                                        return 7;
                                }

                                if(cob_silent)
                                        printf("Got client from %s\n", inet_ntoa(cli.sin_addr));

                                res = pthread_create(&tid,&thread_attr,HandleRequest,&client);
                                if (res != 0)
                                {
                                        syslog(LOG_ERR,":Error - Server thread error: %s",strerror(res));
                                        return 8;
                                }
                                pthread_detach(tid);
                        }

Thanks for your help,

rotis23

**Salem** · 03-31-2004, 06:44 AM

> res = pthread_create(&tid,&thread_attr,HandleRequest,&cl ient);
I dunno about the attribute thing, but this line is wrong.
You're passing a pointer to a local (or global) variable, which could well change before the created thread has a chance to get hold of it.

Ideally, you should malloc some 'thread instance' memory, store your client in that memory, and then pass that memory to the thread proper (which will delete it when it finishes).

int *p = malloc ( sizeof *p );
*p = client;
res = pthread_create(&tid,&thread_attr,HandleRequest,p);

rotis23 · 03-31-2004, 07:23 AM

OK, thanks - I'll take a look.

rotis23 · 04-01-2004, 04:53 AM

Salem,

I see.

The client value will change when the next connection is made i.e. with the new connection descriptor. Although I do make a copy immediately within HandleRequest it still could conceivably change during that time.

It's currently a local variable. I've always been unclear about the scope of local variables and whether it's safe to pass their address (using &) to another function. Or even returning the address of a local variable from function. These instances compile but are surely unsafe?

Can anyone confirm?

**Salem** · 04-01-2004, 05:21 AM

Passing &foo to a called function is safe, within the same thread.

These are unsafe - return address of a local, or pass address of local to another thread
return &foo;
pthread_create( ...., &foo );

rotis23 · 04-01-2004, 05:23 AM

Cheers.

rotis23 · 04-06-2004, 09:10 AM

OK, is it safe to:

Code:

struct test mytest;//struct consists of just 2 ints.

mytest.x = 1;
mytest.y = 34;

res = pthread_create(&tid,&thread_attr,HandleRequest,&mytest);

Will this be shared?

**Salem** · 04-06-2004, 09:30 AM

Didn't I answer this already?

rotis23 · 04-06-2004, 11:50 AM

Well yes, but I've also read (on this list I think) that structs are explicitly copied when passed as an argument to a function.

I'm just unsure whether this applies in the special pthread_create case - obviously not.

**Salem** · 04-06-2004, 01:09 PM

> that structs are explicitly copied when passed as an argument to a function.
Yes they are, but you're not passing a struct, your passing a POINTER to a struct.
So the same rules about the scope still apply.

rotis23 · 04-07-2004, 02:57 AM

Fair enough.

03-31-2004, 06:44 AM	#2
Salem and the hat of Jobseeking Join Date: Aug 2001 Location: The edge of the known universe Posts: 21,674	> res = pthread_create(&tid,&thread_attr,HandleRequest,&cl ient); I dunno about the attribute thing, but this line is wrong. You're passing a pointer to a local (or global) variable, which could well change before the created thread has a chance to get hold of it. Ideally, you should malloc some 'thread instance' memory, store your client in that memory, and then pass that memory to the thread proper (which will delete it when it finishes). int p = malloc ( sizeof p ); *p = client; res = pthread_create(&tid,&thread_attr,HandleRequest,p); __________________ If you dance barefoot on the broken glass of undefined behaviour, you've got to expect the occasional cut.
Salem is offline

03-31-2004, 07:23 AM	#3
rotis23 Registered User Join Date: Aug 2002 Posts: 351	OK, thanks - I'll take a look.
rotis23 is offline

04-01-2004, 04:53 AM	#4
rotis23 Registered User Join Date: Aug 2002 Posts: 351	Salem, I see. The client value will change when the next connection is made i.e. with the new connection descriptor. Although I do make a copy immediately within HandleRequest it still could conceivably change during that time. It's currently a local variable. I've always been unclear about the scope of local variables and whether it's safe to pass their address (using &) to another function. Or even returning the address of a local variable from function. These instances compile but are surely unsafe? Can anyone confirm?
rotis23 is offline

04-01-2004, 05:21 AM	#5
Salem and the hat of Jobseeking Join Date: Aug 2001 Location: The edge of the known universe Posts: 21,674	Passing &foo to a called function is safe, within the same thread. These are unsafe - return address of a local, or pass address of local to another thread return &foo; pthread_create( ...., &foo ); __________________ If you dance barefoot on the broken glass of undefined behaviour, you've got to expect the occasional cut.
Salem is offline

04-01-2004, 05:23 AM	#6
rotis23 Registered User Join Date: Aug 2002 Posts: 351	Cheers.
rotis23 is offline

04-06-2004, 09:10 AM	#7
rotis23 Registered User Join Date: Aug 2002 Posts: 351	OK, is it safe to: Code: struct test mytest;//struct consists of just 2 ints. mytest.x = 1; mytest.y = 34; res = pthread_create(&tid,&thread_attr,HandleRequest,&mytest); Will this be shared?
rotis23 is offline

04-06-2004, 09:30 AM	#8
Salem and the hat of Jobseeking Join Date: Aug 2001 Location: The edge of the known universe Posts: 21,674	Didn't I answer this already? __________________ If you dance barefoot on the broken glass of undefined behaviour, you've got to expect the occasional cut.
Salem is offline

04-06-2004, 11:50 AM	#9
rotis23 Registered User Join Date: Aug 2002 Posts: 351	Well yes, but I've also read (on this list I think) that structs are explicitly copied when passed as an argument to a function. I'm just unsure whether this applies in the special pthread_create case - obviously not.
rotis23 is offline

04-06-2004, 01:09 PM	#10
Salem and the hat of Jobseeking Join Date: Aug 2001 Location: The edge of the known universe Posts: 21,674	> that structs are explicitly copied when passed as an argument to a function. Yes they are, but you're not passing a struct, your passing a POINTER to a struct. So the same rules about the scope still apply. __________________ If you dance barefoot on the broken glass of undefined behaviour, you've got to expect the occasional cut.
Salem is offline

04-07-2004, 02:57 AM	#11
rotis23 Registered User Join Date: Aug 2002 Posts: 351	Fair enough.
rotis23 is offline

Thread Tools
Show Printable Version Email this Page
Display Modes
Linear Mode Switch to Hybrid Mode Switch to Threaded Mode

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Alice....	Lurker	General Discussions	16	06-20-2005 02:51 PM
The Pthread Hell	matott	Linux Programming	1	04-10-2005 05:59 AM
Debugging question	o_0	C Programming	9	10-10-2004 05:51 PM
Question...	TechWins	A Brief History of Cprogramming.com	16	07-28-2003 09:47 PM
Question, question!	oskilian	A Brief History of Cprogramming.com	5	12-24-2001 01:47 AM