UNIX for Advanced & Expert Users

I wonder if someone knows what is the rationale behind linking function calls of the POSIX threads library at link-time vs. run-time.

For example, if I create the following program:

Code:

#include <pthread.h>

void noop() {
  return;
}

int main() {
  pthread_self();
  pthread_atfork(noop, noop, noop);
  return 0;
}

and compile it on Ubuntu version 10.10 with gcc version 4.4.5 using:

Code:

gcc -o test -lpthread -pthread test.c

then the pthread_self symbol is linked (dynamically) at run-time, while the pthread_atfork symbol is linked (statically) at link-time.

Further, there is the case of pthread_equal, code of which might be inlined (assuming __USE_EXTERN_INLINES is defined).

My project plans to use a run-time interposition based wrapper for the POSIX threads library functions to monitor and control interaction of a program with the pthreads library. Cases such as pthread_atfork and pthread_equal are giving me a headache :-).

--Jiri

It's extremely difficult to track down how nptl works internally sometimes. It's full of pointless-looking stubs like

Code:

void pthread_function_thing(...) { return(___internal_pthread_function_thing(...)); }

and they'd never ever put the ___internal junk in the same file, that would be far too easy.

But if I had to guess, it's a way to get at your program's internal symbols in a way a mere shared library wouldn't be able to. Adding a callback to fork() can't have been easy.