erase() crashes

[AIIleG]

erase() crashes

[mattreecebentley]

Unfortunately impossible to diagnose without some dummy code which replicates the issue.
I could spend hours mocking up tests but without the exact match, hard to know if I'd hit the issue.
Any idea if it's a memory leak issue?

[EDIT: only thing I can see that you might be doing wrong there is that if your code erases the last element in the colony, it's going to iterate past .end() and cause a segmentation fault. To fix:

for (it = nclients.begin(); it != nclients.end(); ) {
	if (it->lastcheck < cutoff)
		it = nclients.erase(it);
        else ++it;
}

[mattreecebentley]

Ah, I see. BTW using indexes in that way will always be slower than using an iterator,
because it creates two operations (begin() +i, ++i) instead of one (++it).
Merry Christmas!

[AIIleG]

Thanks for the quick reply. I tested the loop issue you mentioned:

struct nclient {
	string ip;
	unsigned port;
	time_t lastcheck;
};

int main() {
	plf::colony<struct nclient> nclients;
	plf::colony<struct nclient>::iterator it;
	struct nclient dummy;

	dummy.ip = "1.2.3.4";
	dummy.port = 53;
	dummy.lastcheck = 381846;
	nclients.insert(dummy);
	dummy.ip = "1.2.5.4";
	dummy.port = 5333;
	dummy.lastcheck = 3846;
	nclients.insert(dummy);
	printf("count: %u\n", nclients.size());

	for (it = nclients.begin(); it != nclients.end(); ++it) {
		printf("%u\n", it->port);
		it = nclients.erase(it);
	}

	printf("count: %u\n", nclients.size());
	nclients.clear();
	printf("count: %u\n", nclients.size());
	exit(0);
}

Result:

count: 2
53
count: 1
count: 0

And no crash. After all, that kind of loop is the same as you have in the examples on your website. I guess I have to add more debug logging and investigate further. Thanks either way so far!

[mattreecebentley]

That's not the loop I wrote.
For what you've got there, not only will your loop skip the element after the one you've erased during the loop, it will crash in the manner I've described if you end up erasing the back element.
Closing until you've re-written your code with a working loop (I don't have any examples like yours in my code) and can prove that it's not the result of going past end().

[AIIleG]

The loop you wrote there: https://plflib.org/colony.htm#functions

 for (plf::colony<int>::iterator it = i_colony.begin(); it != i_colony.end(); ++it)
  {
    it = i_colony.erase(it);
  }

The purpose of my test was to see whether this causes a crash and it did not. Since that's a quick and simple test I did that first before trying the different loop you suggested in this very thread here previously.

[mattreecebentley]

Yeah, that loop explicitly states that it's trying to erase half the elements, in a use-case where I know that the number of elements is even, therefore it will reach the end condition.

[AIIleG]

Yes okay, my test above with 3 elements does crash. Might not hurt to make that more clear on the website.
Either way, taking the safe route you suggested above seems to prevent the crashes. Thanks!

[mattreecebentley]

For any std:: container, erase returns the element after the element you're erasing, so the situation will be the same for any standard container. It just might take longer to reach a crash state for a vector because non-back erasure is so much slower. Cheers-

[AIIleG]

Thanks for the insight. That might be why it didn't crash for me with a vector because I didn't catch the return. I did it like this:

vs = nclients.size();

for (i = 0; i < vs; ++i) {
	nclients.erase(nclients.begin() + i);
}

Running this with 3 elements doesn't crash unlike the colony (or other container) version with an iterator.

[AIIleG]

Thanks again for the info and Merry Christmas to you, too :-)