Memory leak with Map

[aferust]

The following code causes 40040 bytes of the leak. I mostly used drmemory. But Valgrind yields a similar result too. Maybe, I am wrong but, I somehow detected that while the number of the mallocNew calls for RBNode* is 1001, and the number of the destroyFree calls is 1000 at the end of the scope.

import dplug.core;

// drmemory test.exe -light -count_leaks

 int main() @nogc nothrow
{

    import core.stdc.stdio;

    Map!(int, int) aa0;

    foreach (i; 0..1000){
        aa0[i] = i;
    }

    foreach (i; 0..1000){
        printf("%d \n", i);
        aa0.remove(i);

    }

    // aa0.clearContents(); // the same leak with this also.

    return 0;
}

[p0nce]

Hi @aferust thanks for the report! (edit: reproduced (edit2: no))

[p0nce]

I don't think Map this is the leak you are searching.

clearContents() does not deallocate _rbt itself. If you close the scope and measure allocations everything is returned. Map is lazy-initialized so that Map.init is a valid Map, but an empty Map is still one allocation.

import dplug.core;

 int main() @nogc nothrow
{

    import core.stdc.stdio;
    {
        Map!(int, int) aa0;
        foreach (i; 0..1000){
            aa0[i] = i;
        }

        foreach (i; 0..1000){
            printf("%d \n", i);
            aa0.remove(i);        
        }

        aa0.clearContents();
    } // aa0 destroyed here. After this scope, number of destroyFree and mallocNew is matched
    return 0;
}

[aferust]

I am creating a betterC port of the RedBlackTree and Map. I used struct for RBtree instead of class. It looks like I forgot to add an extra destroyFree(_end) for the last node here. Now there is no leak reported with my port. However, drmemory still reports leaks with dplug. I don't know this one is related also.

[p0nce]

Well, Inspector report leaks with every D program I've seen. I believe there is something in _Dmain In lack of confirmed Dplug bug, I would prefer to close this as I don't have a reproducible issue?

[aferust]

Interesting. Go close it.

Inspector report leaks with every D program I've seen. I believe there is something in _Dmain

I will keep that in my mind.

[p0nce]

It's true small leaks aren't given a high priority in Dplug (vs enterprise software), since a small leak in a plugin instance usually has no impact for the user.

The following D program that prints its arguments:

import std.stdio;

void main(string[] args)
{
    foreach(s; args)
        writefln("%s", s);
}

has this report in Inspector

When built with LDC instead of DMD:

LDC but with Dr memory: dr-report-ldc.txt

For your example and that example + DMD, DrMemory crash...

so not sure what to do with all those reports. Some of the stack traces are inside MS CRT.

[aferust]

Upon our conversation here, I also installed Inspector. I found out that it is very versatile, easy-to-use, and lightweight. I immediately deleted drmemory from my windows machine forever 😁.

Merci Monsieur.

[p0nce]

Yes it is free to use since a few years, you also get Amplifier (though I usually prefer the AMD profiler) :)