Random Chitchat 2012-2016

[NiLSPACE]

So today I tested the cWorld::DoExplosiontAt function with the cLineBlockTracer. This is the result:

[bearbin]

Cool!

[xoft]

I knew there would be cool stuff
Now make it animate in time - napalm thrower

[xoft]

I'm having trouble with Linux's sem_timedwait() function. For some reason it returns a "timeout" immediately after being called with a 0.5sec timeout. Therefore the DeadlockDetect thread misfires on Linux.

Does anyone have any experience with this?

---

For those of you with Linux, can you compile and run this quick test program?

#include <time.h>
#include <stdlib.h>
#include <stdio.h>
#include <semaphore.h>
#include <errno.h>

int main(void)
{
    sem_t sem;
    if (sem_init(&sem, 0, 0))
    {
        printf("sem_init failed\n");
        return 1;
    }
    timespec timeout;
    timeout.tv_sec = 1;
    timeout.tv_nsec = 500 * 1000 * 1000;
    clock_t Begin = clock();
    int res = sem_timedwait(&sem, &timeout);
    clock_t End = clock();
    printf("sem_timedwait: %d, errno %d, %llu ticks\n", res, errno, (long long)(End - Begin));
    sem_destroy(&sem);
    return 0;
}

Compile with (assuming test.cpp is the filename):

Code:

g++ test.cpp -lpthread

When run, it should print something like

Code:

sem_timedwait: -1, errno 110, 0 ticks

I'm interested in the results - are you getting zero ticks as well? Where's the bug?

[bearbin]

I get the same as you.

---

Are the set and nsec values supposed to be different like they are?

110 seems to indicate that it timed out locking the thing. but idk.

[xoft]

It is expected to time out, because no-one is signalling the semaphore, but it is expected to time out after the specified timeout value, which in this case is 1.5 sec (1 sec + 500M nsec). For some reason it "timeouts" immediately.

[bearbin]

Why not try signalling it and see if it works then?

[xoft]

Oh, duh! So stupid, didn't read the docs. The timeout is specifiead absolute, not relative, so it needs to be set to time(NULL) + nsec

[bearbin]

Derp. Great that you got it working then.

[xoft]

Still, some weirdness going on. The test program stops correctly for the 1.5 sec, but still reports 0 ticks. Anyway, it's good enough for me now; I have more pressing issues:

1, Pickups
2, Hunger

---

It seems that PerWorldThreads has brought in a lot of unexpected resource contention; the current debugging version is no longer capable of running the world tick thread, not even for a single player. The release version is doing better, but still I need to fix this, because we can't do much without the debugging version.