source: trunk/source/geometry/solids/test/SBT/README@ 1350

SBT README      DCW 6/4/99

Solids Batch Test
-----------------

This facility is designed to test the correct behavior of CSG solids.
It features:

      1. Interactive creation of solids. Solid parameters can
         be specified at the UI interface so that varied tests can
         be run on solids with different parameters without recompiling
         the code.
         
      2. Log file. Errors are listed in a log file for later study.
      
      3. Debugging. Errors in a log file can be recreated for later
         offline debugging.
         
      4. Visualization. Errors in a log file can be recreated offline
         for visual display, as an aid in debugging.
         
      5. Grids. There is an option to create test trajectories and 
         points along discrete points in space, to test errors at the
         exact intersections with solid edges and corners.


Installation
------------

The standard make file included with SBT links against the full geant4
libraries (much like a geant4 application). 

It should be possible to produce a makefile that only requires a
subset of libraries, but this makefile is not supplied in the package.

SBT has been tested on SUN-CC and Linux platforms.


General (Interactive) Use
-------------------------

The general use of SBT involves the following commands:
    1. Choosing a solid (with a "/solid/" command)
    2. Specifying geometry test parameters (with a "/test/" command)
    3. Running the geometry test (with command "/test/run")
    4. Specifying voxel test parameters (with a "/voxel/" command)
    5. Running the voxel test (with command "/voxel/run")
    6. Repeating as desired

Example session:

/solid/G4Polycone 0 360 4 (1.0,1.2,1.4,1.2) (-1.0,-1.0,1.0,1.0)
/test/maxPoints 100
/test/errorFileName mylogfile.log
/test/run


Batch Jobs
----------

geant4 command scripts for some solids are supplied. They are given
names <solid>.geant4, e.g. "box.geant4" and "polycone.geant4". Each
script attempts to run a comprehensive test using several variations
of each solid, with and without discrete test points. When used they
produce several log files. The log file are given the names 
<solid>.<test>.log, e.g. "box.a1.log".

The Unix script "runSBT" is included to do the following:
     1. Run SBT using a geant4 script
     2. Tally up in one line the total number of
        reported errors in all log files


Method: geometry
----------------

SBT creates a set of random points in 3D space, in a Gaussian distribution
with adjustable mean and width in each dimension. The default is to
generate points with a distribution centered at zero and with a width
of 1 meter. This is appropriate for solids of dimensions of roughly 1
meter on each side.

Adjustments of the mean is possible with the "/test/target" command
and the widths with the "/test/widths" command. Adjustment of these
parameters are generally not necessary if enough points are generated
but may be desirable in order to make the test more efficient.
Adjustment of the mean may also be desirable for discrete points
(see description later in this file).

For each generated point, SBT first asks the target solid if the
point is inside, outside, or on the surface of the solid. Depending
on the answer, the point is added to a corresponding list of points.
Each of the three lists (of inside, outside, or surface points) contains
at most 100 points. If a new point is added to a list that already 
contains 100 points, it replaces an existing point chosen at random.
The limitation of 100 points is intended to keep SBT running roughly
in a CPU time of roughly proportional to the total number of
points generated.

After being added to its list, each inside point is tested
against all of the points on the outside list. Similarly, each
outside point is tested against all of the points on the inside list.
At the moment, no tests are performed on the surface points.
The tests are described in more detail below.

Any error encountered is written to a log file. The name of the log
file by default is "sbt.log", and can be changed with the
"/test/errorFileName" command. The format of the log file is
described in more detail below. To prevent flooding of the log file
by one type of error, if the same error is uncovered more than five
times, any remaining such errors are suppressed in the log file.

SBT finishes its test when one of the following conditions are met:

     1. The total number of random points reaches a maximum
     2. The total number of errors reported reaches a maximum
     
Value (1) defaults to 10000 points and is set by command "/test/maxPoints".
Value (2) defaults to 100 errors in the log file and is set by
command "/test/maxErrors".


Method: voxel
-------------

To start a voxel test, SBT first generates up to 100000 random
points until it finds 1000 points inside the solid (using method 
G4VSolid::Inside).

Afterwards, it starts generating random voxels. It does this by
throwing a random number for each x,y,z dimension. If this
number is less than 0.2, this coordinate is given no limits.
Otherwise, the remaining 0.8 is spread between +/- the range specified
in command /voxel/widths (default value is 1 meter).

For each random voxel, 20 random offsets are choosen. For these
20 offsets in addition to no offset, the following rotations are tried:

        1. No rotation
        2. Random rotation around z
        3. A further random rotation around x
        4. A further random rotation around y
        5. A further random rotation around z

So, each random sized voxel is tested under 21*5 = 105 different
offsets and rotation.

Each voxel test consists of calling G4VSolid::CalculateExtent for
each of the three coordinate axes.

For each axis, if CalculateExtent returns false, the following tests
are performed:

        1. Make sure none of the 1000 points inside the solid
           (found earlier) are inside the voxel.

Otherwise, if CalculateExtent returns true, the following tests are
performed:

        1. Make sure all 1000 of the points inside the solid
           (found earlier) are inside the limits returned.

        2. Make sure the min limit is smaller or equal to the max limit.

        3. Make sure the min and max limits are not outside the
           original voxel limits (if there were any)
           
        4. Calculate the set of points spread in the following
           pattern:
           
                      1----2----3
                      |    |    |
                      |    |    |
                      4----5----6
                      |    |    |
                      |    |    |
                      7----8----9

           along the two axes not being tested to the current
           voxel limits along those two axes, or to +/- 10 meters
           if the voxel is not limit only an axis. These nine points
           are positions at the min and max points in the voxel.
           SBT then uses G4VSolid::Inside to check whether the
           points are outside or on the surface of the solid.
           

Choosing the Target Solid
------------------------

The following commands are available for creating solids:

/solid/G4Box  <dx> <dy> <dz>

/solid/G4Para  <dx> <dy> <dz> <alpha> <theta> <phi>

/solid/G4Trap  <dz> <theta> <phi> <dy1> <dx1> <dx2> <alpha1> <dy2> <dx3> <dx4> <alpha2>

/solid/G4Trd  <dx1> <dx2> <dy1> <dy2> <dz>

/solid/G4Sphere  <rmin> <rmax> <startPhi> <deltaPhi> <startTheta> <deltaTheta>

/solid/G4Torus  <rmin> <rmax> <rtorus> <startPhi> <deltaPhi>

/solid/G4Tubs  <rmin> <rmax> <dz> <startPhi> <deltaPhi>

/solid/G4Cons  <rmin1> <rmax1> <rmin2> <rmax2> <dz> <startPhi> <deltaPhi>

/solid/G4Hype  <innerRadius> <outerRadius> <innerStereo> <outerStereo> <dz>

/solid/G4Polycone  <phiStart> <phiTotal> <numRZ> <r[]> <z[]>

/solid/G4Polyhedra  <phiStart> <phiTotal> <numSides> <numRZ> <r[]> <z[]>

Units are always meters and degrees.

The arrays r[] and z[] have the following syntax:

       (<value1>[,<values2>...])  | (-)
       
where *no spaces are allowed*. The special array "(-)" is the empty array.

Examples:

/solid/G4Box 1 1 1

/solid/G4Tubs 0.8 1 1 0 90

/solid/G4Polycone 0 90 4 (1.0,1.2,1.4,1.2) (-1.0,-1.0,1.0,1.0)


Log File Format: geometry test
------------------------------

The log file is an ASCII file delimited by newlines. Each line
that begins with "%" is a comment. Other lines have the format:

<error number> <p.x> <p.y> <p.z> <v.x> <v.y> <v.z>

where p is the position and v is the (unit) direction most closely
associated with the error. Before each such error line, SBT adds
a comment describing the error. In addition SBT adds comments
concerning the test at the beginning and end of the file.

The meaning of the specific errors are described below.


Log File Format: voxel test
---------------------------

The log file from the voxel test is not made to be machine readable,
since the types of errors from this test are too varied. Instead,
for each error the word "ERROR" is displayed, and a description of the
error follows. The voxel and transform parameters are then displayed.


Testing Inside Points
---------------------

See routine SBTrun::TestInsidePoint in file src/SBTrun.cc.

Each inside point is tested against all points in the outside list.
The tests are, in order:

   1. DistanceToOut(p) is invoked, where p is the inside point being
      tested. The result must be greater than 0, or the following
      error is issued in the log file:
      
        "TI: DistanceToOut(p) <= 0"

   2. The vector v is calculated from the inside point to the
      next point in the list of outside points. DistanceToOut(p,v,...)
      is then invoked. The value must be greater than zero or
      the following error is issued:
      
        "TI: DistanceToOut(p,v) <= 0"

      The result must not be "kInfinity" or the following error is issued:
      
        "TI: DistanceToOut(p,v) == kInfinity"

      The result must not be less than the value from step (1), 
      or the following error is issued:
      
        "TI: DistanceToOut(p,v) < DistanceToIn(p)"

   3. If the normal returned by step (2) is valid, the dot product
      of the normal and v must be greater than zero, or the
      following error is issued:
      
        "TI: Outgoing normal incorrect"

   4. A new point p2 = p + dist*v is calculated, where dist is the
      value found in step (2). Inside(p2) is then invoked. If the
      results is "kInside", the following error is issued:
      
        "TI: DistanceToOut(p,v) undershoots"

      If the result is "kOutside", the following error is issued:
      
        "TI: DistanceToOut(p,v) overshoots"

      In both cases, the value of p is stored in the log file (not p2).

   5. DistanceToIn(p2) is invoked. If the result is not *exactly* zero,
      the following error is issued:
      
        "T02: DistanceToIn(p) should be zero"

   6. DistanceToOut(p2) is invoked. If the result is not *exactly* zero,
      the following error is issued:
      
        "T02: DistanceToOut(p) should be zero"

   7. DistanceToIn(p2,v) is invoked. If the result is not *exactly* zero,
      the following error is issued:
      
        "T02: DistanceToIn(p,v) should be zero"

   8. DistanceToOut(p2,v,...) is invoked. If the result is zero, we have
      encountered the deadlock condition (track passing through a 
      corner) that currently plagues the CSG solid specifications.
      The test on this particular outside point is suspended, and step
      (2) is skipped to for the next outside point on the list.

      If the result is "kInfinity", the following error is issued:
      
        "T02: DistanceToOut(p,v) == kInfinity"

   9. If the normal calculated in step (8) is valid, the dot product
      of the normal and v must be greater than zero, or the following
      error is issued:
      
        "T02: Outgoing normal incorrect"

  10. The vector p3 = p2 + dist*v is calculated, where dist is the
      value returned in step 8. Inside(p3) is then invoked. If the
      result is "kInside", the following error is issued:
      
        "T02: DistanceToOut(p,v) undershoots"

      If the result is "kOutside", the following error is issued:
      
        "TO2: DistanceToOut(p,v) overshoots"

  11. If the result of step (8) has a valid normal, DistanceToIn(p3,v)
      is invoked. If the result is not "kInfinity", the following
      error is issued:
      
        "TO2: DistanceToOut incorrectly returns validNorm==true (line of sight)"

      (validNorm is to be returned true only if the solid is entirely
      behind the surface being exited.)

  12. If the result of step (8) has a valid normal, the vector
      p3top = pi - p3 is calculated for all points pi on the inside list.
      The dot production of p3top with the normal (from step (8)) is
      then calculated. If this dot product for any inside point
      is positive, then the following error is issued:
      
        "T02: DistanceToOut incorrectly returns validNorm==true (horizon)"

      (validNorm is to be returned true only if the solid is entirely
      behind the plane perpendicular to the normal at the exit point.)


Testing Outside Points
----------------------

See routine SBTrun::TestOutsidePoint in file src/SBTrun.cc.

Each outside point is tested against all points in the inside list.
The test are, in order:

   1. DistanceToIn(p) is invoked, where p is the target outside
      point. The result must be greater than zero, or the following
      error is issued:
       
        "T0: DistanceToIn(p) <= 0"

   2. The vector v is calculated from the outside point to the
      next point in the list of inside points. DistanceToIn(p,v)
      is then invoked. The value must be greater than zero or
      the following error is issued:

        "T0: DistanceToIn(p,v) <= 0"

      The result must not be "kInfinity", or the following error
      is issued:
      
        "T0: DistanceToIn(p,v) == kInfinity"

      The result must be greater than the value from step (1),
      or the following error is issued:
      
        "T0: DistanceToIn(p,v) < DistanceToIn(p)"

   3. A new point p2 = p + dist*v is calculated, where dist is the
      value found in step (2). Inside(p2) is then invoked. If the
      results is "kOutside", the following error is issued:
      
        "TO: DistanceToOut(p,v) undershoots"

      If the result is "kInside", the following error is issued:
      
        "TO: DistanceToOut(p,v) overshoots"

      In both cases, the value of p is stored in the log file (not p2).


Discrete Points
---------------

There is an option of generating random points on a grid. The origin
of the grid always corresponds to the mean value (as set by "/test/target").
The spacing of the grid is adjusted by command "/test/gridSizes".


Debugging: geometry
-------------------

Once a test is run and a log file created, it is possible to read the
values from the log file to recreate an error. This is useful in the
debugger to discover the precise reason for the error.

No check is made that the current solid is identical to the solid
used to make the log file. It is up to the user to ensure this
(using the appropriate "/solid/" command if necessary).

    NOTE:
    This has changed. The current solid is written in the log file
    So the ErrorView script could convert a log file to a visualization
    script that could be run by SBT.

The following commands are supported:

   1. /test/draw <error number>

      Draw the error and the solid (if possible) using the currently
      selected visualization.

   2. /test/debugInside <error number>
   
      Invokes the "Inside" method of the test solid.

   3. /test/debugToInP <error number>
   
      Invokes the "DistanceToIn(p)" method of the test solid.
      
   4. /test/debugToInPV <error number>
   
      Invokes the "DistanceToIn(p,v)" method of the test solid.
      A new point p2 = p + dist*v is calculated, and then the
      "Inside" method is invoked for that point.
      
   5. /test/debugToOutP <error number>
   
      Invokes the "DistanceToOut(p)" method of the test solid.
      
   6. /test/debugToOutPV <error number>
   
      Invokes the "DistanceToOut(p,v,...)" method of the test solid.
      A new point p2 = p + dist*v is calculated, and then the
      "Inside" method is invoked for that point.

These commands are intended to be in combination with a debugger (such
as dbx). To debug a particular error, one sets a break point in the
relevant routine(s) and issues the appropriate "/test/" command.
The "/test/draw" command may be useful to isolate the problem.

Clearly, to use the debugger requires that the geant4 libraries linked
to SBT are compiled with the debug flag on (environment variable G4DEBUG).


Debugging: voxel
----------------

Included in SBT is a package for displaying voxels and the
results of ::CalculateExtent. The commands for this are under
the tree /voxel/picture.

     1. /voxel/picture/voxel
     
        Takes six double numbers (xmin xmax ymin ymax zmin zmax)
        to specify the voxel dimensions. If any min > max, that
        dimension is taken to be unlimited.

     2. /voxel/picture/translate
     
        Takes three double numbers to specify the translation in
        the voxel.

     4. /voxel/picture/rotate
     
        Takes four doubles numbers (axisx axisy axisz rotation) to
        specify the axis and rotation amount to specify the
        rotation of the voxel.

     5. /voxel/picture/point
     
        Takes three numbers for an optional point to be displayed
        along with the voxel and solid shape.

     6. /voxel/picture/limit
     
        Specifies the voxel limits to be drawn inside the
        voxel.

     7. /voxel/picture/draw
     
        Draws the items as specified in the above 1-6 commands.
        
     8. /voxel/picture/debug
     
        Invokes ::CalculateExtent and ::Inside for the items
        as specified in the above 1--6 commands. Useful
        for running the solid code inside a debugger.


Problems
--------

Visualization is used in a manner that may not be entirely consistent
with its design. This may introduce errors.

SBT uses the random number generator G4UniformRand. It does not appear
that this random number generator is reproducible across platforms.

Recovering an error from a log file sometimes does not reproduce the
problem, presumably due to a subtle inaccuracy in translating the double
values to and/or from the ASCII log file. This despite writing out
14 digits for the values in the ASCII log file.