TCanvas size is inconsistent in batch vs non-batch mode

[ferdymercury]

• Checked for duplicates

Describe the bug

I see one small issue when saving a 600x600 TCanvas window as png.

I get:

• Interactive mode: 598x571 pixels
• Batch mode: 596x572 pixels

Expected behavior

No pixel difference is found between running batch or not. Right now there is an offset of 1 up to 2 pixels.

To Reproduce

void test() {
    TCanvas *c = new TCanvas("c1","c1",600,600);
    c->SaveAs("c1.png");
}
//root -l test.C -b -q
//root -l test.C -q

Setup

   ------------------------------------------------------------------
  | Welcome to ROOT 6.27/01                        https://root.cern |
  | (c) 1995-2021, The ROOT Team; conception: R. Brun, F. Rademakers |
  | Built for linuxx8664gcc on Jan 11 2022, 18:39:08                 |
  | From heads/master@v6-25-01-2870-gdac9b6398d                      |
  | With c++ (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0                     |
  | Try '.help', '.demo', '.license', '.credits', '.quit'/'.q'       |
   ------------------------------------------------------------------

[couet]

@ferdymercury : I do not think there is a lot we can do. The png file is not a screen copy and some rounding issues in completely different code might be possible (one or 2 pixels). I propose to close this issue.

[ferdymercury]

In my other computer, I get now a difference of 4 pixels using latest ROOT master. So it seems to me that this is not a small rounding error.

• Interactive mode: 598x576 pixels
• Batch mode: 596x572 pixels

Could it be just an outdated calibration from many years ago? See this c3de92f

Check out this code below. It seems there is always the same offset on my current computer. Whenever I run in batch mode, I get 4 pixels difference wrt batch mode. My hypothesis is that this could solved if those numbers are adapted, potentially depending on Windows / Mac with an if. It could be even added as a roottest.

// Run either as root -l test_pixels.cpp -b -q
// Or as xvfb-run root -l test_pixels.cpp -q
#include <TCanvas.h>
#include <TROOT.h>
#include <iostream>
void test_pixels() {
    for(Int_t w=500;w<=900;w++) {
        for(Int_t h=500;h<=600;h++) {
            TCanvas *c = new TCanvas("c1","c1",w,h);
            Int_t diffw = w-c->GetWw();
            Int_t diffh = h-c->GetWh();
            if(!gROOT->IsBatch()) {
                if (diffw !=2 || diffh !=24)
                    std::cout << diffw << " " << diffh << " " << w << " " << h << std::endl;
            }
            else {
                if (diffw !=4 || diffh !=28)
                    std::cout << diffw << " " << diffh << " " << w << " " << h << std::endl;
            }
            delete c;
        }
    }
}

[couet]

ok, I'll look again ...
Why do you need this endless loop to show the problem ?
In this last example there no png generation involved as in the original question... are you pointing an other issue ?

[ferdymercury]

Sorry Olivier, I missed your reply.

The loop was just to see if the offset of 4 pixels appeared always, or only if the canvas was big enough. Or also to see if the offset was proportional to canvas size or not.

I am not creating PNGs in the snippet above, but rather looking at the graphics are size without borders, which is what the PNG should become later. I took the logic out of the ROOT documentation:

 To define precisely the graphics area size of a canvas in the interactive mode, the following four lines of code should be used:
{
   Double_t w = 600;
   Double_t h = 600;
   auto c = new TCanvas("c", "c", w, h);
   c->SetWindowSize(w + (w - c->GetWw()), h + (h - c->GetWh()));
}

and in the batch mode simply do:
c->SetCanvasSize(w,h);

[couet]

I am not sure we need to spend to much time on this issue as the new default graphics will be soon the one in the Web Browser. In the case the size of the canvas is completly different. There is a new PR on the way. In the master only for the time being.

[dpiparo]

I cannot understand from the conversation: is the issue fixed now?

[ferdymercury]

No, it's not fixed, the script reproducing it is #11004 (comment)

Though low priority ...

[ferdymercury]

Another occurrence in the forum: https://root-forum.cern.ch/t/dimension-changes-in-saveas-canvas-for-batch-mode/63385

[rlalik]

As I just learned about this thread, I see this is long standing issue I was also concerned, I preapred a draft for now for a solution to the problem: #21156 - now TCanvas arguments would be true canavs size, not window size.