Potential TF1Convolution memory leak

[Tialis]

Dear all,

I believe I found a memory leak within the TF1Convolution class, or perhaps the way I’m using it is simply wrong.

This is related also to this thread: Memory increase in repeated fits using TF1Convolution

I see exactly the same behavior. I created a minimal working example to execute TF1Convolution on a loop:

#include <TF1.h>
#include <TF1Convolution.h>

double test_convolution_function(double *x, double *p);

void test_TF1Convolved(){
	TF1 *f1 = new TF1("f1", test_convolution_function, -1, 6, 0);
	while(true) f1->Eval(1.);
}

double test_convolution_function(double *x, double *p) {

	TF1Convolution f_conv("expo", "gaus", -1, 6, true);
	f_conv.SetRange(-1., 6.);
	f_conv.SetNofPointsFFT(40000);

	TF1 f("f", f_conv, 0., 5., f_conv.GetNpar());
	f.SetParameters(1., -0.3, 0., 1.);

	return f.Eval(x[0]);
}

Within this JIRA issue the claim is that it was solved: https://sft.its.cern.ch/jira/browse/ROOT-9361

But I just tested today root v6-13-02 and the problem persists (also with v6-12-06).

I’m running on Ubuntu 16.04 with gcc 4.8.

Am I using the wrong ROOT version? Or is the problem really still there?

Thank you in advance!

[moneta]

Hi,

The issue has been fixed in 6.13.03 and in 6.12 patches, 6.12.07. The version you have tested are just the one before and still contain the bug

Cheers

Lorenzo

[Tialis]

Oh, I assumed the fix would be contained within v6-13-02. I will test the gitHub master and confirm it here.

Thank you Lorenzo!

[Tialis]

Dear Lorenzo,

I just tested the github master version and the memory leak is indeed reduced, but not removed (executing the code I copy-pasted above).

But modifying it slightly, as shown below, the memory leak disappears:

#include <TF1.h>
#include <TF1Convolution.h>

double test_convolution_function(double *x, double *p);

void test_TF1Convolved(){
	TF1 *f1 = new TF1("f1", test_convolution_function, -1, 6, 0);
	while(true) f1->Eval(1.);
}

TF1Convolution* f_conv;
double test_convolution_function(double *x, double *p) {

	if (f_conv) delete f_conv;
	f_conv = new TF1Convolution("expo", "gaus", -1, 6, true);
	f_conv->SetRange(-1., 6.);
	f_conv->SetNofPointsFFT(40000);

	TF1 f("f", f_conv, 0., 5., f_conv->GetNpar());
	f.SetParameters(1., -0.3, 0., 1.);
	return f.Eval(x[0]);
}

I’m not sure why there is such a leak in the code above, but I assume that is my lack of knowledge.

Thank you for the help!