
USING glBSP
===========

by Andrew Apted.  13th September 2005.


Running glBSP
-------------

There are now two versions of glBSP: normal 'glbsp' (the text
version), and 'glBSPX' (offering a graphical interface).  This
documentation applies to the text version only.  Under Windows,
the text version must be run from the command shell.

There are two main ways of using glBSP:

  1. Create the GL nodes and store them in a separate file with the
     GWA extension.  This is called "GWA mode".

  2. Create the GL nodes and store them in the WAD file itself, just
     like how a normal node builder works.

These ways are described in more detail below.


1. GWA Mode
-----------

For general use, when you just want to use some wad (e.g. FOO.WAD)
with a compliant OpenGL DOOM port, then this method is the recommended
way.  It is also the simplest method, just type:

   glbsp foo.wad

and then glBSP will create the file "foo.gwa".  A compliant OpenGL
DOOM port (such as EDGE) will notice the GWA file automatically and
load it appropriately.  Nothing more needs to be done.

You can also specify multiple wads here, even wildcards should
work.  For example:

   glbsp foo.wad fred.wad simon*.wad

Note !  You should not give these GWA files to the DOOM port using
the command line (i.e. with the -file option), doing so will confuse
it and it'll probably crash.  The port will notice the GWA file
automatically.

Note 2: when using GWA mode, the output file does not contain any
information for things, linedefs, sidedefs, sectors, reject, or
blockmap, all it contains is the GL node information.  This is
where the name "GWA" comes from, it is short for "GL WAD" (in
case you were wondering).


2. Non-GWA Mode
---------------

This method is intended for WAD authors, so they can create a single
WAD file for distribution that contains all the needed information.
Using this method looks like this:

   glbsp foo1.wad -o foo2.wad

and then glBSP will build the GL nodes and output the new file
"foo2.wad" containing them.  It will also build the normal node
information if it is absent.


Response Files
--------------

New in version 2.20 is support for response files.  These are
files containing a list of options.  You specify the response
file by prefixing it with '@'.  For example:

   glbsp @arg_list.rsp

The "@arg_list.rsp" on the command line will be replaced with
the contents of that file.  New-line characters are treated like
spaces.  Recursion (using '@' inside a response file) is not
supported.


Option Descriptions
-------------------

The following options are recognised:

  -q           Quieter output.  Information about each level (like
               the number of linedefs, blockmap size, etc) is not
               displayed when this option is given, and a few other
               messages are skipped.  Important messages, like
               failure to build a certain level, are still shown.

  -fast        Lets glBSP can cheat a bit and re-use the original node
               information to create the GL nodes, doing it much faster.
               Use this option to enable this feature.  The message
               "Using original nodes to speed things up" will be shown.

               The downside to reusing the original nodes is that they
               may not be as good as the ones glBSP normally creates,
               e.g. the special checks to minimise slime-trails don't
               kick in, and the -factor value doesn't have any effect.

  -warn        Shows extra warning messages, which detail various
               non-serious problems that glBSP has while analysing the
               level structure.  Often these warnings show a real
               problem in the level (e.g. a non-closed sector or
               invalid sidedef), so they are worth checking now and
               then.

  -normal      glBSP usually detects if the normal node info (i.e.
               the non-GL variety) is present: when yes, it is left
               untouched, otherwise glBSP creates it.  This option
               forces glBSP to replace the normal node data with
               newly constructed nodes.

  -factor <num>
               Changes the cost assigned to seg splits.  Factor can
               be any number greater than 0, larger values make seg
               splits more costly (and thus glBSP tries harder to
               avoid them), but smaller values produce better BSP
               trees.  See the section below for more info.  The
               default value is known to be a good compromise.

  -pack        Pack sidedefs, by detecting which sidedefs are
               identical and removing the duplicates, producing a
               smaller PWAD.

               NOTE: this will make your level a lot harder to edit!
               Therefore this is most useful when producing your final
               WAD for public release.

  -noreject    Normally glBSP will create an simple REJECT map for
               each level.  This options prevents any existing
               REJECT map, such as one time-consumingly built by
               a dedicated reject builder, from being clobbered.

These options are rarely needed:

  -v1 .. -v5   Specify the version of the "GL Nodes" spec to use.
               The default is -v2.  Using -v1 produces an obsolete
               format.  Giving -v3 or -v5 will force certain lumps
               to use the new formats, but is only useful for testing
               since glBSP will automatically switch to V5 format
               when certain limits are exceeded.

  -loadall     glBSP normally tries to copy lumps from the input
               WAD file to the output file instead of loading them
               into memory.  This allows you to run glBSP on very
               large WADS (e.g. 15 MB for DOOM II) on a low memory
               machine.

               This option causes everything from the input file to be
               loaded into memory.  This allows you to run glBSP using
               the same file for both input and output, but I strongly
               recommend _against_ it: you could lose your original
               WAD if something goes wrong (and you know Murphy...).

  -noprog      Turn off the progress indicator.

  -nonormal    Forces glBSP to not create the normal node information
               when it detects that it is absent.

  -forcegwa    Normally glBSP will automatically use GWA mode if the
               output filename is missing (i.e. no -o option) or if
               the output filename has the ".gwa" extension.  This
               option forces glBSP into GWA mode.

  -prunesec    Removes any unused sectors found in the level.  Only
               works when normal nodes are being built.  This is
               not done by default since some scripting languages
               (Fragglescript, old versions of RTS) refer directly
               to the sector numbers and removing even one will
               change all the numbering.

  -mergevert   Merge duplicate vertices at the same location into a
               single vertex.  This is usually safe, but is not done
               by default because some engines (e.g. Risen3D) need the
               duplicate vertices to stay separate for special effect.

  -maxblock <num>
               Sets the limit of the number of blocks the BLOCKMAP may
               contain before we truncate it.  Default is 44000.  When
               the level is too large to fit, glBSP will truncate the
               blockmap, so it covers less area on the level.  This
               means that in the parts it doesn't cover (at the outer
               edges) there is no collision detection: you can walk
               through walls and other objects and bullets/missiles
               don't hit anything.  On very large but sparse levels,
               using a larger value (e.g. 60000) may help.

               A more serious problem is when the blockmap overflows.
               The blockmap created would be invalid, and could crash
               the DOOM engine when used.  glBSP will create an empty
               blockmap instead, causing modern ports to build their
               own blockmap.


ZDBSP Format Nodes
------------------

When the normal nodes overflow, older versions of glBSP would simply
write out the invalid node data.  glBSP 2.20 will write out the node
data in the ZDBSP format (originally created for the ZDoom engine).


Interaction with other tools
----------------------------

As far as I know, none of the various WAD tools that exist (such
as DSHRINK, CLEANWAD, DEUTEX, etc..) are "glBSP aware", they will
rearrange or even remove some of the special GL entries, and
everything goes pear shaped.

When using a reject building tool (such as RMB), you need to give
the -noreject to glBSP to prevent the good REJECT data from being
overwritten.

*** DO NOT: ***

  + Run DSHRINK on your map WADs at any time!

  + Run CLEANWAD on your map WADs *after* you have compiled your GL
    friendly nodes!  (The GL nodes will be removed).

  + Use Wintex/DEUSF to merge map WADs with GL friendly nodes in them!
    (The GL node entries will be rearranged, causing problems).


How glBSP works
---------------

A node builder works like this: say you are looking at your level in
the automap or in the level editor.  The node builder needs to pick a
line (stretching to infinity) that divides the whole map in two halves
(can be rough).  Now cover up one half with your hand, and repeat the
process on the remaining half.  The node builder keeps doing this
until the areas that remain are convex (i.e. none of the walls can
block the view of any other walls when you are inside that area).

Those infinite lines are called "partition lines", and when they cross
a linedef, the linedef has to be split.  Each split piece is called a
"seg", and every split causes more segs to be created.  Having fewer
segs is good: less to draw & smaller files, so partition lines are
chosen so that they minimise splits.  The "-factor" value controls how
costly these splits are.  Higher values cause the node builder to try
harder to avoid splits.

So, each "partition" line splits an area (or "space") of the level
into *two* smaller spaces.  This is where the term "Binary Space
Partition" (BSP) comes from.

Another thing: having a good BSP tree is also important, and helps for
faster rendering & smaller files.  Thus the node builder also tries to
make the BSP tree good (by making it balanced, and keeping it small).
If the "-factor" value is too high, it will care too much about the
splits, and probably make a bad BSP tree.  How good the BSP tree is
can be gauged by the output line that reads:

   Heights of left and right subtrees = (12,24)

Lower values are better (the BSP tree is smaller), and values that
are closer together are also better (the BSP is more balanced).


Differences to BSP 2.3
----------------------

As mentioned in the readme file, glBSP was originally based on BSP 2.3.
Most of the code has been rewritten, however, and some features of BSP
were changed or abandoned.  Features that are different:

  +  When the output file is not specified (i.e. no -o option), then
     the default output file will be a GWA file with the same name.
     Under BSP 2.3, the default output file would be "tmp.wad".

  +  All code for doing visplane checks has been removed.  It was very
     complex stuff, and for OpenGL DOOM ports this checking is not
     needed.  Thus glBSP does not accept the following options that
     BSP 2.3 supports: -thold, -vp, -vpmark, -vpwarn.

  +  glBSP works on big-endian platforms (like MacOS X).

  +  The "just for a grin" special feature where a linedef with tag
     999 would cause an angle adjustment was removed.

  +  glBSP has Hexen support.

  +  glBSP packs the blockmap, and can merge identical sidedefs.

  +  glBSP has a much more modular architecture, and can even serve as
     a plug-in for other programs.

  +  The GUI version, glBSPX, is completely new.



---------------------------------------------------------------------------

