weird behavior in MaterialData Init() [SOLVED]

On 29/06/2015 at 16:51, xxxxxxxx wrote:

thanks Andreas, I will rely on constructor then "for local class members".

On 29/06/2015 at 22:23, xxxxxxxx wrote:

well. doing it in constructor didn't work "simply because it is calling Alloc() ..."
so what I want to achieve:
store some data inside the material "is this possible with local variables? or I have to use descriptions"

On 01/07/2015 at 06:02, xxxxxxxx wrote:

Hi Mohamed,

you already found out, that the constructor is a bad idea.
But why not use Init()? If you have a problem with multiple initializations, you could simply have a flag in your class, which gets set after the initialization is done. Next time you enter Init() you simply check the flag and exit.

On 01/07/2015 at 07:58, xxxxxxxx wrote:

this is horrible!!!, the class is constructed multiple times, how will the flag work?
I wanted to create a singleton flag for any purpose inside the class, but as the class is created everytime it calls matpreview, it doesn't work "all class members are thrown away and I can't rely on Init()"

for example:
Init() { m_update_preview = TRUE;}
and in the mat preview render:
if(m_update_preview) { m_update_preview = FALSE; render();}

this is totally won't work, no class members are working as it is initialized multiple times, what should I do?

On 01/07/2015 at 08:00, xxxxxxxx wrote:

So you have global data, that is used by all instances your material class?
In such cases I tend to use a globally static variable as a flag.

On 01/07/2015 at 08:37, xxxxxxxx wrote:

not global.

On 02/07/2015 at 05:30, xxxxxxxx wrote:

???
You don't want a global flag? Then, why not?
You have no global data? Fine.
But your problem are those repeated Init() calls, right?

Regarding my suggestion of a flag as member variable. This could still work, if you correctly implement it in Read()/Write() and CopyTo().

On 02/07/2015 at 09:23, xxxxxxxx wrote:

I see, so descriptions can solve this? "trying to avoid overriding Read()/Write()"

On 02/07/2015 at 17:45, xxxxxxxx wrote:

to some extent, the problem is half solved, here is what I did:
used CopyTo() , which solved how the buttons work "in the video, I have 2 buttons, one that sets the flag to TRUE and one sets it to FALSE"

now the problem is: I want to set the flag to FALSE after the render, some code:

  
		case MATPREVIEW_GENERATE_IMAGE:
			{
				MatPreviewGenerateImage* image = (MatPreviewGenerateImage* )data;
				if (image->pDoc)
				{
					Int32					w = image->pDest->GetBw();
					Int32					h = image->pDest->GetBh();
  
					RenderData* rdata = image->pDoc->GetActiveRenderData();
					BaseContainer bcRender = rdata->GetData();
  
					bcRender.SetFloat(RDATA_XRES, w);
					bcRender.SetFloat(RDATA_YRES, h);
					bcRender.SetInt32(RDATA_ANTIALIASING, ANTI_GEOMETRY);
					
					image->pDest->Clear(0, 0, 0);
					GePrint("m_update_preview = " + String::IntToString(m_update_preview));
					if(image->bLowQuality)
					{
						bcRender.SetBool(RDATA_RENDERENGINE, RDATA_RENDERENGINE_PREVIEWSOFTWARE);
						image->lResult = RenderDocument(image->pDoc, bcRender, nullptr, nullptr, image->pDest,
															 RENDERFLAGS_EXTERNAL | RENDERFLAGS_PREVIEWRENDER, image->pThread);
					}
					else if(m_update_preview == TRUE)
					{
						((SevenPhotonsMaterial* )node)->m_update_preview = FALSE;
						BaseVideoPost* myRenderer = BaseVideoPost::Alloc(ID_SEVENPHOTONS);
						rdata->InsertVideoPost(myRenderer);
						bcRender.SetInt32(RDATA_RENDERENGINE, ID_SEVENPHOTONS);
  
						image->lResult = RenderDocument(image->pDoc, bcRender, nullptr, nullptr, image->pDest,
															 RENDERFLAGS_EXTERNAL, image->pThread);
					}
				}
				break;
			}  
		case MSG_DESCRIPTION_COMMAND:
			{
				DescriptionCommand* dc = (DescriptionCommand* )data;
				if (dc->id[0].id == SEVENPHOTONS_MAT_PREVIEW_UPDATE)
				{
					m_update_preview = TRUE;
					break;
				}
				else if (dc->id[0].id == SEVENPHOTONS_MAT_PREVIEW_UPDATE_STOP)
				{
					m_update_preview = FALSE;
					break;
				}
			}

so with the 2 commands, it works fine, but it doesn't change to FALSE in the MATPREVIEW_GENERATE_IMAGE message, any ideas?

On 03/07/2015 at 05:45, xxxxxxxx wrote:

Hm? My questions may be stupid...
Are those cases in the same switch? Then why do you access m_update_preview directly in MSG_DESCRIPTION_COMMAND and via a node in MATPREVIEW_GENERATE_IMAGE?
Otherwise without knowing your class layout it's difficult to tell. Maybe you are resetting the flag in a copy?

On 03/07/2015 at 07:47, xxxxxxxx wrote:

not via a node
this: ((SevenPhotonsMaterial* )node)->m_update_preview = FALSE;
is the same as this: m_update_preview = FALSE;
it is a class member.
what I wanted to achieve is: when I hit update button,it renders "only once" //this one didn't work,
as it is perma updating, that's why I tried setting the flag to false once it enters "old singleton style"
when I hit cancel button,it breaks any rendering. //I achieved this

On 06/07/2015 at 09:01, xxxxxxxx wrote:

Hi Mohamed,

the reason you can't reset the flag is indeed, that you are working on a copy. MATPREVIEW_GENERATE_IMAGE is send to a clone of your material for rendering purposes.

That being said, I have to say, I have no solution, yet. I'm currently looking into MATPREVIEW_GET_DIRTY_COUNT and MATPREVIEW_COPY_USERDATA, if these might offer a solution. I will also discuss your case with the rest of the team tomorrow.

I'm sorry, this thread is taking so long and I still can't offer a solution. But I'm still optimistic, we'll find a solution.

On 23/07/2015 at 08:21, xxxxxxxx wrote:

Hi Mohamed,

finally I found the time to play around with this. Sorry, it took so long.
As said before, the problem is, that MATPREVIEW_GENERATE_IMAGE is send to a clone of your material. So you can't reset the flag, because you only reset it in the clone, not the original material.

My admittedly quite hacky approach is to have a BaseLink member (_linkOrig) in the material. Plus another boolean member (_flagRequestPreview) storing the "render request" flag.

In Init() I initialize as follows:

_linkOrig->SetLink(nullptr);
_flagRequestPreview = true; // assuming that a first initial preview is always wanted

When the material is copied (CopyTo()), I set the link. Either to the source material (if not previously set) or to the stored link. Like so:

MaterialPreviewExample* sdata = static_cast<MaterialPreviewExample*>(snode->GetNodeData());
BaseList2D* bl2d = sdata->_linkOrig->GetLink(snode->GetDocument());
if (bl2d)
  static_cast<MaterialPreviewExample*>(dest)->_linkOriginalMaterial->SetLink(bl2d);
else
  static_cast<MaterialPreviewExample*>(dest)->_linkOriginalMaterial->SetLink(snode);

In MSG_DESCRIPTION_COMMAND I do basically the same as you, simply setting _flagRequestPreview. Of course one could use a checkbox instead of two buttons, but that may be a matter of taste.

Then on MATPREVIEW_GENERATE_IMAGE I simply evaluate the link and use/reset the flag of the original material. Like so:

BaseList2D* bl2d = _linkOrig->GetLink(node->GetDocument());
if (!bl2d)
  bl2d = _linkOrig->ForceGetLink();  // this will be the most probable path, as preview is rendered in another document
  
Bool flagRender;
  
if (bl2d)
{
  MaterialPreviewExample* dOrig = static_cast<MaterialPreviewExample*>(bl2d->GetNodeData());
  flagRender = dOrig->_flagRequestPreview;
  dOrig->_flagRequestPreview = false;
}
else
{
  flagRender = _flagRequestPreview;
  _flagRequestPreview = false;
}
// Here: flagRender is true, only once after the button got pushed

As I said, it's a bit hacky. But as it is only used for rendering of previews, I think, it should work ok.
Let me know, what you think.

Another option would be the use of message MATPREVIEW_PREPARE_SCENE. As far as I can see, this message is still send to the original material. But if you get this message, heavily depends on your preview implementation.

On 23/07/2015 at 09:01, xxxxxxxx wrote:

Hi Andreas,

thanks a lot for the help
I will check them tonight and see , most probably the link approach looks good "I will need that link for other stuff later :smile:

On 23/07/2015 at 22:18, xxxxxxxx wrote:

it works very well, thanks Andreas
you can consider this solved.