Hi! Thanks HG1 for review :) About the 0.85f in the shader code, I can't explain it unless "it does not work correctly without" (It doesn't fit the circle dotted lines)... But maybe the formula isn't correct. I just played with the formula of the link to have a result which sounds good to me but I didn't understand it. Maybe someone could find a better formula or could explain mine :) .
I haven't found where I can add a notifier to update light when we press "S" key (I've not much time this days but I'll ask on IRC).

@Campbell Barton (campbellbarton): Thanks for review! About float spotscale[2] in DNA_lamp_types: I think it's a good idea to keep this attribute for lamp since I'm using it in a lot files and it's not always easy (I don't know an easy way) to access "lamp object" from lamp. But if someone gives me a tip to access "lamp object" from lamp, I can remove this and each time I call lamp->spotscale, I'll replace it with Object *ob = lamp->id->object/data??? (something like that)> But I may have to include DNA_object_types in some files... For the moment, I make the link beetween lamp->spotscale and object->obmat in gpu_material.c in the gpu_lamp_from_blender function line 1912.

About the second point: Can this be removed in KX_Light.cpp: This could be removed but: - This part (up) of the code can be used to change color of a lamp too this way: lamp.color[0] = 1.0.
If you remove this part of the code, you have to write lamp.color = [1.0, 0.0, 0.0] (you can't set each rgb component separately. Same for the scale.

• If you use object scale instead of spotscale, you loose some controls on the lamp children behaviour and the lamp physics (Personnaly, I think that it's better to keep the possibility to 1) scale the object light/ 2) only modifiy the shape of the light itself (it costs nothing, you have more control, Blender internal users don't care about it, it concerns only bge).

About the third point: You're right. It's a mess to add notifiers each time you use SKEY, you release SKEY, you use ALT+S etc... What you can try if you want is to add a notifier only in gpu_material.c in gpu_lamp_from_blender function. example: if (something):
{
lamp->spotvec[0] = len_v3(ob->obmat[0]);
lamp->spotvec[1] = len_v3(ob->obmat[1]);

copy_v2_v2(la->spotscale, lamp->spotvec);

DAG_id_tag_update(&la->id, 0);

WM_main_add_notifier(NC_LAMP | ND_LIGHTING_DRAW, la);
}

If you don't use an if statement, this part of the code will slow down blender (you can try by dupicating spots). The problem is to find this "something" in the if statement. I did not succeed to find what to put in this condition. So I tried to add notifiers in transform.c (applyResize, and line 1255 case RIGHTMOUSE (to cancel transformation) and in other parts of the code...), and in RNA_object.c (rna_Object_internal_update).

Anyway, the last version of the patch needs many improvements. I think I'll delete the ellipse spot support for the moment, and I have to find a proper way to update spotlight shape each time we use SKEY for scaling. I may need help for this (if someone has any suggestion :) ? ). Tell me @Campbell Barton (campbellbarton) if you continue to disagree with the first and the second point :) . I'll try to find a way to satisfy everyone.

Have all a wonderfull day/night :)

It seems I was tired when I tried to update lamp in gpu_material.c:

That works this way:

#include "BKE_depsgraph.h"
#include "WM_api.h"
#include "WM_types.h"

lamp->spotvec[0] = len_v3(ob->obmat[0]);
lamp->spotvec[1] = len_v3(ob->obmat[1]);
if (la->spotscale[0] != len_v3(ob->obmat[0]) || la->spotscale[1] != len_v3(ob->obmat[1]))
{
copy_v2_v2(la->spotscale, lamp->spotvec);
DAG_id_tag_update(&la->id, 0);
WM_main_add_notifier(NC_LAMP | ND_LIGHTING_DRAW, la);
}

Can I do like this? I make a little clean up, remove ellipse support, and update the .diff

No, That doesn't work :( . Sorry, I may have tested with the wrong blender.exe :) ... Or I'm missing something. If you have a tip, it will be welcome!

@Ulysse Martin (youle): now you use the scale from the matrix so all variables for the 2d spot scale can be removed ? stop me if i'm wrong.
I tested the patch, all seems work properly.

So glad people are getting into coding the game engine

I need to finish wrectified, and start getting tutored by one of you.,

Thanks !

the conversion of len_v3 in glsl is : "vector.length()"
so x : imat[0].length() and y : imat[1].length()

@Brecht Van Lommel (brecht) and @Antonis Ryakiotakis (psy-fi): If you've got a moment, could you take a look at the last version of the patch (very simplified) and tell me what do you think about it, please? Thanks very much!

Some minor stuff, will check functionality tomorrow

Immediate mode means when you draw your graphic with glBegin(), glEnd() directly to the graphic card. Which is very slow.

glBegin(GL_LINES);
    glVertex3f(0.0f, 0.0f, 0.0f);
    glVertex3f(z_abs, z_abs, x);
    glVertex3f(0.0f, 0.0f, 0.0f);
    glVertex3f(-z_abs, -z_abs, x);
    glVertex3f(0.0f, 0.0f, 0.0f);
    glVertex3f(z_abs, -z_abs, x);
    glVertex3f(0.0f, 0.0f, 0.0f);
    glVertex3f(-z_abs, z_abs, x);
glEnd();

Here an example which is drawing a cube using vertex arrays.
Actually I don't know if Blender now use vertex arrays or vertex buffer objects to draw the objects on the viewport.
So it is best to ask Antony which mode Blender actually use.

// draw a cube drawn with triangles. (initialize arrays)
GLfloat vertices[] = {1.0f, 1.0f, 0.0f,  // 8 of vertex coords oft the cube
                      0.0f, 1.0f, 0.0f,
                      0.0f, 0.0f, 0.0f,
                      1.0f, 0.0f, 0.0f,
                      1.0f, 0.0f, 1.0f,
                      1.0f, 1.0f, 1.0f,
                      0.0f, 1.0f, 1.0f,
                      0.0f, 0.0f, 1.0f};
GLfloat vertices_col[] = {1.0f, 0.352f, 0.0f, 1.0f,  // 8 of vertex colors oft the cube (rgba) if neccesary.
                          1.0f, 0.352f, 0.0f, 1.0f,
                          1.0f, 0.352f, 0.0f, 1.0f,
                          1.0f, 0.352f, 0.0f, 1.0f,
                          1.0f, 0.352f, 0.0f, 1.0f,
                          1.0f, 0.352f, 0.0f, 1.0f,
                          1.0f, 0.352f, 0.0f, 1.0f,
                          1.0f, 0.352f, 0.0f, 1.0f}; 
GLubyte indices[] = {0,1,2, 2,3,0, // 36 of indices first triangle drawn from vertex 0 to 1 to 2 and so on. 
                     0,3,4, 4,5,0,
                     0,5,6, 6,1,0,
                     1,6,7, 7,2,1,
                     7,4,3, 3,2,7,
                     4,7,6, 6,5,4};

// ----Draw call----
// activate and specify pointer to vertex array
glEnableClientState(GL_VERTEX_ARRAY);
glEnableClientState(GL_COLOR_ARRAY);
glVertexPointer(3, GL_FLOAT, 0, vertices);
glColorPointer(4, GL_FLOAT, 0, vertices_col); //GL_FLOAT for 0.0f to 0.1f, GL_UNSIGNED_BYTE for 0 to 255 

// draw the cube
glDrawElements(GL_TRIANGLES, 36, GL_UNSIGNED_BYTE, indices);

// deactivate vertex arrays after drawing
glDisableClientState(GL_VERTEX_ARRAY);
glDisableClientState(GL_COLOR_ARRAY);

@Porteries Tristan (panzergame): Hi! The problem is that if I move "vertices" and "indices" before "x *= y", the lines aren't drawn correctly... I'll discuss it with you tonight if you're on IRC :)

With uniform scale this narrows down the lamps beam on both X and Y axis. this should be resolved.

Select all objects in the attached file and scale, this will change the cone angle.

@Campbell Barton (campbellbarton): Thanks very much for the test! Is this this behaviour you are waiting for: https://youtu.be/ismfJWxIdxU ? (Uniform scale does not change the light beam)

@Ulysse Martin (youle), looks good, for now can you match the behavior of Blender-Internals 'Halo' (view in preview for eg).
Then at least its consistent (and it looks like its doing what you've got working already).

Even though you've likely figured this out - noting that the XYZ scale values should just define a relative aspect for the lamp - if you check Halo behavior - this is whats happening there too.

Looks good, noticed opportunity to avoid calculating all 3 axis lengths twice.

Think this is ready to apply to master. (@Ulysse Martin (youle). please *commandeer* back)

@Campbell Barton (campbellbarton): I trust you (much) more than I trust myself so let's apply this patch :) I definitely approve the decision to remove me from command :) Thanks very much for review, tests and corrections.

Commited rBe60d535443824f44cd6e18280570cd066c1b3dc2
Thanks @Ulysse Martin (youle) & everyone who helped review :)