Re: Two pthread libraries on Red Hat 9

Hi Thomas, 

I don't really know what these two libraries are, but I would use some sort 
of sync to make sure that the main thread doesn't exit before the spawned 
one even starts. I do know that LinuxThreads doesn't necessarily immediately 
start the thread spawned by pthread_create(). Maybe the difference in 
behavior is just a race condition due to a subtle timing difference? 

Oh, could the other thread library by NGPT or NPTL? 

Don 

Thomas Bailey writes: 

> Good morning, 
> 
> I've already posted this to the general RedHat list - but it was
> suggested it might be more appropriate to the development list. 
> 
> I was wondering if anyone could explain to me the difference between the
> pthread libraries in /lib and /lib/tls on RedHat 9. I ask because the
> library in /lib/tls (which is linked by default when you supply
> -lpthread to gcc) displays unexpected behavior, whereas the library in
> /lib or /lib/i686, behaves correctly. 
> 
> The code below demonstrates the difference: 
> 
> #include <dlfcn.h>
> #include <stdio.h>
> #include <stdlib.h>
> #include <netdb.h>
> #include <time.h>
> #include <pthread.h>
> #include <unistd.h> 
> 
> #ifdef __cplusplus
> extern "C" {
> #endif 
> 
> /// Our global key for accesing the TSS
> static pthread_key_t gcKey; 
> 
> /// Destructor
> static void DestroyKey (void* pcValue)
> {
>         printf("Freeing TSS\n");
>         abort();
>         free (pcValue);
> } 
> 
> 
> /// Initialization code
> static void*  Init(void *arg)
> {
>         pthread_key_create (&gcKey, DestroyKey);
>         pthread_setspecific(gcKey, malloc(42));
>         return (void *) 0;
> } 
> 
> 
> int
> main (int argc, char* argv[])
> {
>         pthread_t myThread;
>         pthread_create(&myThread, NULL, Init, NULL);
>         return 0;
> }
> } 
> 
> In this example program, we create a thread, and allocate some thread
> specific storage, passing a pointer to the DestroyKey function to
> pthread_key_create. We would expect this function to be called when the
> thread's thread specific storage is destroyed, i.e. on thread exit. We
> would then expect the program to print a message and abort (i.e. dump
> core). 
> 
> When linked against /lib/tls/libpthread-0.29.so, DestroyKey is not
> called. If I change the permissions on /lib/tls/libpthread-0.29.so and
> /lib/tls/libc-2.3.2.so to make them unreadable, so my program gets
> linked against /lib/i686/libpthread.so.0 and /lib/i686/libc.so.6,
> DestroyKey is called on thread exit. 
> 
> I'm curious as to whether anyone knows a) if this is a bug in the
> default threading library and b) if anyone knows a more elegant way to
> force my application to link against the 'right' libraries other than
> changing the permissions on the 'wrong' libraries. I have tried the
> -nostdlib option to gcc. 
> 
> Regards 
> 
> Tom Bailey 
> 
>  
> 
>  
> 
> -----------------------------------------------------------------------
> The information contained in this e-mail is confidential and solely 
> for the intended addressee(s). Unauthorised reproduction, disclosure, 
> modification, and/or distribution of this email may be unlawful. If you 
> have received this email in error, please notify the sender immediately 
> and delete it from your system. The views expressed in this message 
> do not necessarily reflect those of LIFFE Holdings Plc or any of its subsidiary companies.
> ----------------------------------------------------------------------- 
>