Hi,

we had a user who had issues running a python script with matplotlib and x11 forwarding. Other x11 applications had no issues, and we could sometimes get his script to work.
After some digging it seems like matplotlib tries to connect to x11 about 3 times before displaying anything to the user.

So we wrote a small reproducer that will work ok locally, or remotely over ssh but fails if you run it with srun.

----8<-----------

#include <dlfcn.h>
#include <stdlib.h>
#include <stdio.h>

/* Based on Matplotlib src/_c_internal_utils.cpp */

int mpl_xdisplay_is_valid(void)
{
    void* libX11;
    // The getenv check is redundant but helps performance as it is much faster
    // than dlopen().
    if (getenv("DISPLAY")
        && (libX11 = dlopen("libX11.so.6", RTLD_LAZY))) {
        typedef struct Display* (*XOpenDisplay_t)(char const*);
        typedef int (*XCloseDisplay_t)(struct Display*);
        struct Display* display = NULL;
        XOpenDisplay_t XOpenDisplay = (XOpenDisplay_t)dlsym(libX11, "XOpenDisplay");
        XCloseDisplay_t XCloseDisplay = (XCloseDisplay_t)dlsym(libX11, "XCloseDisplay");
        if (XOpenDisplay && XCloseDisplay
                && (display = XOpenDisplay(NULL))) {
	  XCloseDisplay(display);
        }
        if (dlclose(libX11)) {
	  printf("BAD CLOSE\n");
        }
        if (display) {
            return 1;
        }
    }
    return 0;
}

void main(int argc, char **argv)
{
  printf("success %d\n", mpl_xdisplay_is_valid());
  printf("success %d\n", mpl_xdisplay_is_valid());
  printf("success %d\n", mpl_xdisplay_is_valid());
  printf("success %d\n", mpl_xdisplay_is_valid());
}


---->8-----------

[jonst@tetra ~]$ gcc -o xtest xtest.c
[jonst@tetra ~]$ ./xtest
success 1
success 1
success 1
success 1
[jonst@tetra ~]$ srun -n1 --x11 ./xtest
srun: job 43071435 queued and waiting for resources
srun: job 43071435 has been allocated resources
X connection to localhost:94.0 broken (explicit kill or server shutdown).
success 1
success 0
success 0
srun: error: n341: task 0: Exited with exit code 1
srun: Terminating StepId=43071435.0


Adding a sleep between the calls or removing the XCloseDisplay calls, makes it work, but it is nothing we can add to the matplotlib-code the user has. We are trying some other workarounds, but we were wondering if this is a known issue and if you have any recommendations.