> Hope I could made it somewhat clearer ...
Thank you for your helpfull sugestion.

> The frame memory is passed from the cwork context to the 
> result pixmap.
I understanded that:
1)My last proposal for a bug fixing is wrong, because valiable result has frame memory same cwork's.
2)Below functions of the create/destroy functions are pairs;
  + GrCreateContext(),... vs GrDestroyContext()
  + GrBuildPixmapFromBits(),... vs GrDestroyPattern()
So, when users call GrBuildPixmapFromBits(..,..,NULL,[NULL]), they must take care about
the allocated memory; cf. calling GrDestroyPattern()

But, I think that GrDestroyPattern can not free the memory created by GrBuildPixmapFromBits() completely.

Oliginal source : Oliginal code can free a top of the frame memory area only.
void GrDestroyPattern(GrPattern *p)
{
	if(p->gp_ispixmap) {
	    if(p->gp_pxp_source.gf_memflags & GCM_MYMEMORY)
		farfree(p->gp_pxp_source.gf_baseaddr);
	    if(p->gp_pxp_source.gf_memflags & GCM_MYCONTEXT)
		free(p);
	    return;
	}
	if(p->gp_bitmap.bmp_memflags) free(p);
}

Modified source : Folowing code is modified that it will free all planes of the frame memory.
void GrDestroyPattern(GrPattern *p)
{
	if(p->gp_ispixmap) {
	    if(p->gp_pxp_source.gf_memflags & GCM_MYMEMORY){                    // 991012
		int ii = p->gp_pxp_source.gf_driver->num_planes;                         // 991012
		while(--ii >= 0) farfree(p->gp_pxp_source.gf_baseaddr[ii]);              // 991012
	    }
// c991012	    if(p->gp_pxp_source.gf_memflags & GCM_MYMEMORY)
// c991012	    farfree(p->gp_pxp_source.gf_baseaddr);
	    if(p->gp_pxp_source.gf_memflags & GCM_MYCONTEXT)
		free(p);
	    return;
	}
	if(p->gp_bitmap.bmp_memflags) free(p);
}

What do you think about above code?

Akira Watanabe