[BUG] Fixed init of the main thread

[najlkin]

This PR fixes initialization of the main thread object, which was sometimes happening in a non-main thread first and as it calls SDL_Init(), it was causing crashes on Mac (including CI runners ❗ ).

[v-dobrev]

I have not noticed this issue on Mac. How can I try to reproduce it? Also, which CI jobs failed due to this issue?

[najlkin]

It is random, depends what thread (main or worker) is first, but have a look at one of the many attempts in #296 , but it did not appear there first, I have seen that before, but just re-running the job saved the day usually.

[tzanio]

I also haven't noticed this, does it happens with reading from files only?

[justinlaughlin]

thank you! LGTM

(I do not know how to reproduce this bug but I've seen it pop up in other tests)

[justinlaughlin]

Here is an example of this failure on glvis-4.3-dev

https://github.com/GLVis/glvis/actions/runs/10083308934/job/27879591928?pr=284#step:23:602

[najlkin]

I believe it can happen with any input, we just reproduced that with @v-dobrev on his Mac with an extra sleep(). The thing is that SDLMainLoop(), which calls this GetMainThread() is called just after the thread (or Session object which does it internally) is created.

[v-dobrev]

@tzanio, here's a diff on master that allowed me to reproduce the issue (when loading a file with -saved):

diff --git a/glvis.cpp b/glvis.cpp
index 67d755e..9c255e9 100644
--- a/glvis.cpp
+++ b/glvis.cpp
@@ -1498,6 +1498,7 @@ int main (int argc, char *argv[])
          return 1;
       }
 
+      std::this_thread::sleep_for(std::chrono::milliseconds{100});
       SDLMainLoop();
       return 0;
    }

[v-dobrev]

Looks good to me. Thanks @najlkin!

[tzanio]

Okay, thanks guys. I trust you