Problem with c++ and std::thread comparison

Problem with c++ and std::thread comparison

[Comm Unist]

What happens when you try to call a weak symbol? Do you end up calling 0x000000.....00 depending on your arch?

I ask because I have C++ program that uses std::threads. These use pthreads internally. There were two places where I have problems. One was with condtiion variables and the other with comparsion of std::thread ids:

This changes seems to work, or does c++ not needs to check pthreads being equal?


diff --git a/include/pthread.h b/include/pthread.h
index 99a74a5..e97b9e4 100644
--- a/include/pthread.h
+++ b/include/pthread.h
@@ -85,9 +85,7 @@ __attribute__((const))
 pthread_t pthread_self(void);
 
 int pthread_equal(pthread_t, pthread_t);
-#ifndef __cplusplus
-#define pthread_equal(x,y) ((x)==(y))
-#endif
+inline int pthread_equal(pthread_t a, pthread_t b) { return a == b; }
 
 int pthread_setcancelstate(int, int *);
 int pthread_setcanceltype(int, int *);

Re: Problem with c++ and std::thread comparison

[Jens Gustedt]

[-- Attachment #1: Type: text/plain, Size: 1938 bytes --]

Am Sonntag, den 21.06.2015, 17:10 +1000 schrieb Comm Unist:
> What happens when you try to call a weak symbol? Do you end up calling 0x000000.....00 depending on your arch?

Why do you ask? Could you be more specific in what problem you encounter?

> I ask because I have C++ program that uses std::threads.
> These use pthreads internally.

You mean you are using a C++ library that maps std::threads to
pthreads, or are you talking about a mapping to the musl C11 threads
implementation, that then shares code with musl's pthread
implementation?

(Mapping std::threads to C11 threads should clearly be the more
appropriate choice.)

> There were two places where I have problems. One was with condtiion
> variables

Just annoncing that you have problems doesn't tell us much. Describe
your problem properly, please.

and the other with comparsion of std::thread ids:

> This changes seems to work, or does c++ not needs to check pthreads being equal?
> 
> 
> diff --git a/include/pthread.h b/include/pthread.h
> index 99a74a5..e97b9e4 100644
> --- a/include/pthread.h
> +++ b/include/pthread.h
> @@ -85,9 +85,7 @@ __attribute__((const))
>  pthread_t pthread_self(void);
>  
>  int pthread_equal(pthread_t, pthread_t);
> -#ifndef __cplusplus
> -#define pthread_equal(x,y) ((x)==(y))
> -#endif
> +inline int pthread_equal(pthread_t a, pthread_t b) { return a == b; }

Why wouldn't just a declaration of the function prototype suffice?

The way you are doing this would clearly not work for C since this
would instantiate a version of the function in every translation unit.

Jens


-- 
:: INRIA Nancy Grand Est ::: Camus ::::::: ICube/ICPS :::
:: ::::::::::::::: office Strasbourg : +33 368854536   ::
:: :::::::::::::::::::::: gsm France : +33 651400183   ::
:: ::::::::::::::: gsm international : +49 15737185122 ::
:: http://icube-icps.unistra.fr/index.php/Jens_Gustedt ::





[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 181 bytes --]

Re: Problem with c++ and std::thread comparison

[Szabolcs Nagy]

* Comm Unist <comm.unist@yandex.com> [2015-06-21 17:10:42 +1000]:
> What happens when you try to call a weak symbol? Do you end up calling 0x000000.....00 depending on your arch?

no

you probably encountered weak references in libstdc++
(but hard to tell based on the information you gave us)

extern weak reference will be 0 unless something else
pulls in the definition at link time.

> I ask because I have C++ program that uses std::threads. These use pthreads internally.

that's an internal detail you should not be able to observe

> There were two places where I have problems. One was with condtiion variables and the other with comparsion of std::thread ids:

> This changes seems to work, or does c++ not needs to check pthreads being equal?

if the macro definition is hidden then you get a normal
function call whenever pthread_equal is called.

c++ code must be explicitly made slower by hiding the
macro because it does not allow function like macros
(welcome to c++).

inline semantics is different in c, c++ and gcc so
you cannot make that visible outside of ifdef __cplusplus.

(but diverging code logic under ifdef is not ok
because it makes maintainance problems, the current
code seems to me the most acceptable solution).

> diff --git a/include/pthread.h b/include/pthread.h
> index 99a74a5..e97b9e4 100644
> --- a/include/pthread.h
> +++ b/include/pthread.h
> @@ -85,9 +85,7 @@ __attribute__((const))
>  pthread_t pthread_self(void);
>  
>  int pthread_equal(pthread_t, pthread_t);
> -#ifndef __cplusplus
> -#define pthread_equal(x,y) ((x)==(y))
> -#endif
> +inline int pthread_equal(pthread_t a, pthread_t b) { return a == b; }
>  
>  int pthread_setcancelstate(int, int *);
>  int pthread_setcanceltype(int, int *);

Re: Problem with c++ and std::thread comparison

[Rich Felker]

On Sun, Jun 21, 2015 at 05:10:42PM +1000, Comm Unist wrote:
> What happens when you try to call a weak symbol? Do you end up
> calling 0x000000.....00 depending on your arch?

What weak symbol? If your program is using weak references to pthread
symbols, that's a bug. libstdc++ has this bug upstream but it's fixed
in the patches you should be applying for building a toolchain that
targets musl.

Rich