POV-Ray

Just like radiosity load/save, the photon mapping load/save mechanism should be moved to the frontend and controlled via command-line switch, instead of being SDL-driven in the backend.

The following scene causes a crash:

sphere {
  <0,0,0>, 1
  finish { subsurface { translucency 1.0 } }
}

global_settings {
  subsurface { }
}

For proper render abort/continue support, radiosity cache data must be written to (or read from) disk even if the user does not explicitly opt to have a sample data file written/read. This feature has temporarily been dropped from 3.7 beta and is still pending re-implementation.

To meet high-reproducibility requirements in conjunction with SMP operation, it may be necessary to extend the 3.6 radiosity cache file format.

POV-Ray does not follow common practice on command-line handling; for instance:

povray +i"My File"

entered on a Unix shell would be passed to POV-Ray as

povray
+iMy File

(each line representing a distinct parameter here), which POV-Ray would further dissect, interpreting it as

povray
+iMy
File

To achieve the desired effect, one would actually have to quote the string twice:

povray +i"'My File'"

which the shell would translate to

povray
+i'My File'

which POV-Ray would interpret as

povray
+iMy File

In both cases, this is obviously not what a Unix user would expect.

The further dissecting of individual command-line parameters may have had its valid roots in the peculiarities of DOS’ command-line handling, but to my knowledge all major contemporary operating systems follow a concept akin to Unix, passing a list of parameters instead of a monolithic command line, and burdening the respective command shells with the task of dissecting command lines into parameters.

Therefore I suggest to disable this anachronistic feature in favor of contemporary standards; a compiler flag might be used to allow for easy re-enabling of the feature, for compiling POV-Ray on exotic targets.

- edit -

It has been pointed out that the described behaviour differs from 3.6, so I’m promoting this to a bug and changing the title.

I observe frequent segfaults with POV-Ray 3.7 betas when rendering scenes using photons:

• Debian Linux 4.0r5 “etch” for AMD64
• AMD Phenom X4 9650 2.3GHz, 6 GB RAM
• POV-Ray compiled with Intel icpc 11.0

Segfaults are sporadic but frequent (occurring in roughly 50% of all photon renders).

Using prisms in intersecion or difference CSG objects may cause artifacts in POV-Ray 3.6.2 as well as 3.7.0.beta.34, as demonstrated by the following code:

camera {
  right    -x
  up        y*image_height/image_width
  location  <-24,19,12>
  look_at   <0,0,0>
}

light_source { <100,200,100> color rgb 1 }

plane { y, -2 pigment { color rgb 1 } }


#declare KeyValue = 1.366; // pick any you like

difference {
  prism {
    linear_sweep -0.5,0.5, 4
    
    <-3,20-17>,
    <-3,KeyValue>,
    <-6,-3>,
    <-0,-5>
  }
  intersection {
    cylinder { <-7,-0.51,1>, <-7, 0.51,1>, 4.0 }
    plane { z, KeyValue }
  }
  pigment { color rgb 0.5 } 
}

Apparently the surface of the other object becomes visible when it exactly coincides with a vertex of the prism; probably there is a failure of the inside() test for such values.

Windows UI code should be refactored to use _TCHAR throughout instead of char, as well as the corresponding string function macros, to head for Unicode support.

Code inspection shows that we’re still using fseek() and ftell() in various places (including text file input), which can’t handle file positions of 2GB and beyond (except on 64-bit linux machines); those calls need to be examined and (where appropriate) replaced with the fseek64() macro we’re already defining (but currently not using), and a to-be-defined ftell64() macro.

One potential (untested) error scenario would be a scene file calling a macro that is defined at the end of a > 2GB long include file.

Subsurface Scattering still uses its own rudimentary code to compute illumination from classic light sources; this must be changed to use the standard light source & shadow handling code, to add support for non-trivial light sources (e.g. spotlights, cylindrical lights, area lights), partially-transparent shadowing objects etc.

Subsurface scattering must be made photon-aware.

Near concave edges, radiosity samples may be re-used at a longer distance away from the edge than towards the edge; there is code in place to ensure this, but it only works properly where two surfaces meet roughly rectangularly, while failing near the junction of three surfaces or non-rectangular edges, potentially causing “cross-talk”.

It should be investigated how the algorithm can be improved or replaced to better cope with non-trivial geometry.

POV-Ray 3.6, upon encountering problems when parsing command line and/or .ini file options, would quote the offending option in the error message.

POV-Ray 3.7 currently just reports that there is some problem with the command line, without providing any details. I suggest changing this, as the information may be helpful at times.

Currently, radiosity does not make use of the fact that pertubed normals would theoretically just require a different weighting of already-sampled rays, leading to the following issues:

• Honoring normal pertubations in radiosity leads to an increased number of samples, slowing down sample cache lookup.
• The increased number of samples is generated from a proportionally higher number of sample rays, slowing down pretrace even further.
• Low-amplitude pertubations tend to be smoothed out; “reviving” these is only possible by increasing the general sample density.
• Handling of multi-layered textures with different normal pertubations is currently poorly implemented.

As a solution, I propose to store for each radiosity sample not only the resulting illumination for a perfectly unpertubed normal, but from the same set of sample rays also compute the illumination for an additional set of about a dozen standardized pertubed-normal directions, and interpolate among these when computing the radiosity-based illumination for a particular point that has a pertubed normal.

For backwards compatibility, this method of dealing with pertubed normals in radiosity might be activated by a different value for the “normal” statement in the radiosity block, say, “normal 2”.

Some features are optional in custom builds of POV-Ray (I’m thinking about OpenEXR in particular); it would be nice to have a syntax for an SDL script to check for support of such features, so it may take some fallback action if the feature is not supported.

Being able to have functions operate on vectors would be pretty nice to have.

User-defined warps would be nice to have, something along the lines of:

warp {
  function { MyFnX(x,y,z) } // function to compute pattern-space x-coordinate from object-space <x,y,z> coordinate
  function { MyFnY(x,y,z) } // ditto for pattern-space y coordinate
  function { MyFnZ(x,y,z) } // ditto for pattern-space z coordinate
}

// a displacement warp:
warp {
  function { x + MyFnX(x,y,z) }
  function { y + MyFnY(x,y,z) }
  function { z + MyFnZ(x,y,z) }
}

Most sample scenes and include files were designed at times when POV-Ray did not to any proper gamma handling, or still used the inferior 3.6 “assumed_gamma” mechanism.

All the scenes and include files should be reviewed, and updated to fit the new 3.7 gamma model.

The primary task will probably be gamma-adjusting literal color values and ambient parameters; I suggest using macros (which ideally should be defined in an include file) to be set according to the #version statement, so the scene/include file could be kept compatible with older versions.

With previous versions of POV-Ray, it was possible to turn off display output, but still assess the output during render by viewing the output file as it was progressively generated. This allowed e.g. to run a long render on a remote machine as a background process, and check the output from time to time via FTP or similar.

The following scene gives a comparison of current conserve_energy handling in standard shadow computations vs. photons.

Note how the rather highly reflective slabs fail to cast shadows, except where the photons target sphere enforces computation of shadow brightness to be done by the photons algorithm.

For more realistic shadowing without the need to enable photons, I suggest do add proper conserve_energy handling to the shadow computation code (which shouldn’t be too much effort).

global_settings {
  max_trace_level 10
  photons { spacing 0.003 media 10 }
}

camera {
  right x*image_width/image_height
  location  <-2,2.6,-10>
  look_at   <0,0.75,0>
}

light_source {
  <500,300,150>
  color rgb 1.3
  photons {
    refraction on
    reflection on
  }
}

sky_sphere {
  pigment {
    gradient y
    color_map {
      [0.0 rgb <0.6,0.7,1.0>]
      [0.7 rgb <0.0,0.1,0.8>]
    }
  }
}

plane {
  y, 0
  texture { pigment { color rgb 0.7 } }
}

#declare M_Glass=
material {
  texture {
    pigment {rgbt 1}
    finish {
      ambient 0.0
      diffuse 0
      specular 0.2 // just to give a hint where the sphere is
    }
  }
  interior { ior 1.0 }
}

#declare M_PseudoGlass=
material {
  texture {
    pigment {rgbt 1}
    finish {
      ambient 0.0
      diffuse 0.5
      specular 0.6
      roughness 0.005
      reflection { 0.3, 1.0 fresnel on }
      conserve_energy
    }
  }
  interior { ior 1.5 }
}


sphere {
  <1.1,1,-1.3>, 1
  material { M_Glass }
  photons {
    target 1.0
    refraction on
    reflection on
  }
}

// behind target object
box {
  <-0.2,0,-2.3>, <0.0,4,0.3>
  material { M_PseudoGlass }
  rotate z*1 // just to better see the reflection of the horizon
}

// before target object
box {
  <2.4,0,-2.3>, <2.6,4,-0.3>
  material { M_PseudoGlass }
  photons { pass_through }
  rotate z*1 // just to better see the reflection of the horizon
}

The combination of metallic reflection with conserve_energy causes the reflection to lose colour, as demonstrated by the following scene:

global_settings {
  max_trace_level 10
}

camera {
  right x*image_width/image_height
  location  <-2,2.6,-10>
  look_at   <0,0.75,0>
}

light_source {
  <500,300,150>
  color rgb 1.3
}

sky_sphere {
  pigment {
    gradient y
    color_map {
      [0.0 rgb <0.6,0.7,1.0>]
      [0.7 rgb <0.0,0.1,0.8>]
    }
  }
}

plane {
  y, 0
  texture { pigment { color rgb 0.7 } }
}

#declare M=
material {
  texture {
    pigment {rgbt <1.0,0.7,0.2,0.99>}
    finish {
      ambient 0.0
      diffuse 0.5
      specular 0.6
      roughness 0.005
      reflection { 0.8, 1.0 metallic }
      conserve_energy
    }
  }
  interior { ior 1.5 }
}

box {
  <-0.2,0,-2.3>, <0.0,4,0.3>
  material { M }
  rotate z*5
  rotate x*2
}

When using cutaway_textures in a CSG object that has union children, results are not as expected; instead, surfaces in the union children that have no explicit texture will be rendered with the default texture instead. This is not the case for e.g. difference children.

Example:

#default { texture { pigment { rgb 1 } } }

camera {
  right x*image_width/image_height
  location  <0,1.5,-4>
  look_at   <0,1,0>
}

light_source { <500,500,-500> color rgb 1 }

#declare U = union {
  sphere { <0,-0.1,-1>, 0.3 }
  sphere { <0, 0.1,-1>, 0.3 pigment { color red 1 } }
}

intersection {
  sphere { <0,0,0>, 1 pigment { color green 1 } }
  object { U }
  cutaway_textures
  rotate y*90
}

When declaring U as an intersection instead, the results are as expected, with the surface of the first sphere in U being rendered with the texture defined in the outer intersection.

The syntax notation used in the main documentation is different than that used in the quick-reference section. This should be changed for consistency, using the superior quick-reference notation throughout.

The SDL currently provides no way to compute the exact direction of a spline at a given location, even though mathematically this is a piece of cake: The first-order derivative of any spline section gives you the “speed” as a vector function, and is trivial to compute for polynomial splines (which are behind all spline types that POV-Ray supports); the normalized “speed” vector, in turn, gives the “pure” direction.

For exact direction/speed computations, I propose to extend the SDL invocation syntax as follows to allow for evaluating a spline’s derivative:

    SPLINE_INVOCATION:
        SPLINE_IDENTIFIER ( FLOAT [, SPLINE_TYPE] [, FLOAT] )

or

    SPLINE_INVOCATION:
        SPLINE_IDENTIFIER ( FLOAT [, FLOAT] [, SPLINE_TYPE] )

where the second FLOAT will specify the order of derivative to evaluate (defaulting to 0). In order to compute the position, direction, and acceleration of an object traveling along a certain spline, one could then for instance use:

    #declare S        = spline { ... }
    #declare Pos      = S(Time);
    #declare VSpeed   = S(Time,1);
    #declare VAccel   = S(Time,2);
    #declare Dir      = vnormalize(VSpeed);
    #declare Speed    = vlength(VSpeed);
    #declare AccelDir = vnormalize(VAccel);
    #declare GForce   = vlength(VAccel) / 9.81;

Alternatively, a mechanism may be devised to create a spline representing another spline’s derivative; however, it would be debatable whether the syntax should be parameter-like (being an added information that could be overridden again when creating other splines from such a derived spline), or operation-like (converting the spline), and in the latter case how it should affect spline type (and consequently control points); so the spline invocation parameter approach might be more straightforward, with less potential surprises for the user.

In the photons map, the direction of each photon is stored as separate latitude & longitude angles (encoded in one byte each), causing the longitudinal direction component’s precision to be unnecessarily high for directions close to the “poles” (Y axis); in addition, encoded value -128 is never used. For better overall precision as well as precision homogenity, the following scheme could be used instead:

• Encode the latitude (-pi/2 to +pi/2) into LatCount=226 distinct values (= 256*sqrt(pi)/2) rounded to the next even number) from 0 to LatCount-1 using

latCode = (int)((LatCount-1) * (lat/M_PI + 0.5) + 0.5)

• For each latitude code, define a specific number of encodable longitude values, LngCount[latCode] = approx. cos(lat)*pi*65536/(2*LatCount); this can be a pre-computed table, and may need slight tweaking for optimum use of the code space. Encode the longitude (-pi to +pi) into a value from 0 to (LngCount[lat]-1) using

LC = LngCount[latCode];
lngCode = (int)(LC * (lng/(2*M_PI) + 0.5) + 0.5) % LC;

• Besides LngCount[latCode], also store the sum of LngCount[i] with i < latCode as LatBase[latCode]; encode the direction as

dirCode = LatBase[latCode] + lngCode;

• For decoding, a simple lookup from a precomputed list of directions could be used (2^15 entries, i.e. one hemisphere, will suffice). To conserve space, direction vectors could be scaled by (2^N-1) and stored as (N+1)-bit signed integer triples rather than floating point values; due to the limited precision of the lat/long information, 8 bits per coordinate might be enough, giving a table size of 96k. A full double-precision table would require 786k instead.

circular area lights exhibit some anisotropy, being brighter along the diagonals than on average, as can be demonstrated with the following scene:

//+w800 +h800
#version 3.7;
global_settings{assumed_gamma 1}
plane{-z,-10 pigment{rgb 1} finish{ambient 0 brilliance 0}}
disc{0,z,10000,0.5}
camera{orthographic location z look_at 10*z up y*12 right x*12}
light_source{-10*z rgb 10 area_light 10*x 10*y 257 257 adaptive 4 circular}

When resolution is not specified (neither via POVRAY.INI nor via QUICKRES.INI nor via command line or custom .ini file), random values are displayed for image resolution in the Image Output Options message output. (The actual render will be performed at the default size of 160×120 pixels though.)

When a command-line parameter in an .ini file cannot be parsed (such as “+a.3”), POV-Ray reports a “Problem with setting”, quoting the command line, rather than indicating that the problem occurred in an .ini file. This leads the user to think that the problem is with the command line itself, unnecessarily confusing him.

Front- & Back-end code should be refactored for full Unicode support in scene files and strings.

When using OpenEXR output file format, POV-Ray erroneously reports it as “24 bpp EXR” in the message output, while in fact it generates a 3×16 = 48 bpp file.