video processor: gfx_fillrect() erratic behavior in some transparency contexts(fixed)

[papagirafe]

gfx_fillrect() seems to be confusing the roles of gfx_a and gfx_a2. Let suppose we have a track with an image item and we gfx_fillrect() a rectangle with a gradual envelope on gfx_a (0->1) and gfx_a2=1. If we gfx_blit() this track on top of another one using mode 0x10000, the part covered by gfx_fillrect() will do weird jumps in transparency with a partial mix of gfx_a2 transparency. In non-transprency mode, there is no problem (obviously). I made a project to illustrate the problem which includes gfx_blit() and gfx_fillrect() emulations made with gfx_evalrect() which do behave correctly.

https://stash.reaper.fm/45284/fillrect_erratic.zip




p.s. I was using v6.66 on win10 but I have encountered this problem months ago before I understood properly how alpha works


*** UPDATE 2022-10-24: I confirm, it has been fixed in v6.69 ****

[Justin]

ah yes, there is a bug in the alpha handling of LICE_FillRect() which causes it to write the alpha value rather than the pixel's alpha value. Fixing, though this might be some time before it goes into a release because it could break appearances subtly. Thanks for the report!

[papagirafe]

Originally Posted by Justin

ah yes, there is a bug in the alpha handling of LICE_FillRect() which causes it to write the alpha value rather than the pixel's alpha value. Fixing, though this might be some time before it goes into a release because it could break appearances subtly. Thanks for the report!

Thanks for looking at it! I am sure there are lots video projects that have been patched around this bug (even in mines wihout inderstanding why). At least in contexts where it causes a problem, we can use its evalrect() emulation.

//gfx_fillrect() emulation for proper transparency behavior
gfx_set(r,g,b,a1,mode,dest,a2);
gfx_evalrect(x,y,w,h,"r=r*(1-gfx_a)+gfx_a*gfx_r*255;g=g*(1-gfx_a)+gfx_a*gfx_g*255;b=b*(1-gfx_a)+gfx_a*gfx_b*255;a=gfx_a2*255");