BUGS

BRL-CAD Bugs
============

Bugs should be reported to the bug tracker on the project website at
http://sourceforge.net/tracker/?atid=640802&group_id=105292

The bugs and issues listed in here may or may not be the same as or
related to the bugs reported to the bug tracker.  In general, users
should not look to this file for information regarding the status of
bugs.  Informal bug-related information that is intended for
developers will generally be found here.  This may include short term
issues that are in active development as well as long term and
on-going issues.

Recent Bugs
-----------

* get_regions tcl command doesn't respond well to situations where
  child object names in combs have dashes (the "-" character) in
  names.  It tries to interpret the dash as an argument delimiter.
  Looks like this may be only in situations where the "-" is the
  lead character and on Windows?  Been hard to reproduce...

* dsp does not facetize correctly with zero elevation areas in the
  data - need robustness improvements

* dsp primitives specifying data using Windows style file paths
  don't work.

* attempting to create a dsp using a binary object caused a crash on
  Linux - need to figure out if this was caused by feeding it the wrong
  kind of binary object or some other error...

* screengrab on windows does not work

* Recursive submodel predictably goes infinite, need stack-based
  db_walk_tree(): mged> in submodel submodel submodel grid test.g

* running rt -s20 -F file.pix ... does not result in an 20x20 image
  being written out.  it ends up being 512x512.

* running rt -C255/0/0 -o file.pix ... does not result in the
  background color getting used.  It does work with -F file.pix, going
  through libfb instead of libicv.

* ogl display manager and framebuffer are creating windows with
  transparent backgrounds (the glClear color is not being drawn) on
  Linux.  Presumably a driver-related issue, alpha channel, or expose
  event issue.  Inspecting further, it may be related to environments
  where shmget() fails (or is requested via /dev/oglp) and we get
  private memory -- looks like private memory is not initialized
  (observed on Mac 10.6, Windows, and Ubuntu Linux using 7.26).

* rtedge's overlay mode appears to be generating different results on
  Windows and Linux for the m35 steering wheel (rtwizard rendering
  mode D.)

* pixdiff on Windows is reporting no differences with a pair of files
  that pixcmp can see differences in.

* fblabel seems quite busted, at least talking to a /dev/X fbserv, in
  that it only draws the first character and eventually displays
  rem_read() failures when run in -d debug mode.

* rtedge doesn't refresh a /dev/X fbserv (presumably double-buffered)
  after doing an overlay render.  it does draw, though, as confirmed
  by a forced fb-fb refresh.

* pixborder doesn't account for an edge case where the right edge of
  an image is expected to be colored (except for one pixel on the top
  edge).  Example:

  pixborder -b "255 0 0" -t "255 255 255" < pix/moss.pix | pix-fb

* report that attempting to change units on a read-only file in MGED
  will not change the units in the file but changes them in MGED,
  resulting in problems with nirt missing geometry it shouldn't miss.

* typing 'comb -f -r assembly_name' shuts down mged with the error:
  ERROR: bad pointer 0000000000B4CA50: s/b bu_external(x768dbbd0), was bu_vls(x89333bbb), file ..\..\..\brlcad_src\src\libbu\parse.c, line 2443"
  And on the mged command line:
  rt_db_cvt_to_external5 expecting non-NULL name parameter
  wdb_export((null)): solid export failure

* report of mged command history not working with the down arrow key
  in 7.25.0 (on Windows, unknown on other platforms).  Up arrow works
  fine.

* wgl framebuffer is not lingering on windows (regardless of using
  wgll or default) when it should

* nirt -e "xyz val val val" doesn't work due to the space after the -e
  and how the nirt/rt commands parse quoted args.

* calling "attr set obj key value" should always result in key getting
  set to value, no matter what, as it is the lowest-level attribute
  management interface and what the command (correctly) documents.
  example demonstrated to not work was: attr set obj rgb 123/123/123

* report of LIBBU not working when accessed by a multithreaded
  application on Windows.  reportedly due to some problem with the
  debug memory queue.

* report of g-iges crashing (on Windows) when exporting a 1.2MB BoT
  mesh (presumably imported via stl-g).  worked for smaller models.

* autoview [scale] isn't updating the wireframe when opening a
  database, drawing geometry, then running autoview.  the status bar
  shows the scale change, but the graphics window remains unaffected
  unless you run "view size" to set it.

* overlay command reports an error when trying to load a file
  names edges.pl, but succeeds when the file is called patches.pl
  Need to see why edges.pl is not accepted.

* step-g crashing on mac with lazy symbol load failure missing:
  brlcad::SurfaceTree::SurfaceTree(ON_BrepFace*, bool, int)

* metaball tessellation enters either an infinite loop or an absurdly
  long tessellation process - observed on Linux 64 bit system - quick
  test:

  rm -f metaball.g && mged -c metaball.g "make m.s metaball; facetize m.bot m.s"

* corruption/crash issue rendering terra.g without terra.dsp,
  possibly/probably parallel-related in libbn/noise.c's
  bn_noise_turb() function.  gdb eventually catches a segfault in
  find_spec_wgt() loop with indications of corrupted memory.  valgrind
  confirms corrupted memory (8 bytes) after a build_spec_table()
  realloc within find_spec_wgt() but it's not clear why.  etbl
  book-keeping seems okay..  corrupted read consistently detected on
  spec_wgts = ep->spec_wgts line in bn_noise_turb().  observed on
  64-bit linux system:

  rm -f terra.pix && valgrind bin/rt -o terra.pix -P1 -H3 -s1024 share/brlcad/7.21.0/db/terra.g all.g

* mged 'joint' hold command isn't working. looks like the fp_names
  book-keeping is fuxored or the db_lookup is returning NULL (or isn't
  yet performed).

* idents command is reporting a bogus region count in the region
  output table listing (first column).

* src/librt/primitives/bot/g_bot_include.c:rt_bot_bbox() routine isn't
  calculating the bbox correctly

* 'exists' command documentation claims more than has been implemented

* rtxray crashes if you specify a -o *.bw output file due to
  assumptions in the rt front-end that all output files are .pix

* raytracing a revolved sketch seems to work fine, but rtarea and gqa
  seem to be crashing on that same revolve.  curiously, the brep
  revolve from the csgbrep proc-db has the same failure as well as the
  brep ehy.

* encountered a crash during rtarea on a (default) facetized version
  of the hyp from the proc-db csgbrep tool.

* raytracing rhc crashes (inside malloc, so previous memory badness)
  with normal tolerance set to non-default value (e.g. tol norm 0.1);
  also fails (non-crash) if absolute tolerance is changed (e.g., tol
  abs 0.1)

* revolve raytracing seems to be busted.. at least with the example
  csgbrep revolve shape

* torus with a near zero inner diameter results in zero-length edges
  during tessellation.

* opening a database in archer with the same name as a command doesn't
  work.

* bu_log() with a width specifier does not align like printf().
  Example: bu_log("%24s", str);

* report that "keep" sometimes results in a .g file with unknown units;
  dbconcat also is reported to set the units to unknown if concatting
  a file with unknown units.

* infinite loop in rt_find_backing_dist() raytracing a BoT (without
  pieces) for bot_smooth (i.e., rt_bot_smooth()).  encountered inf
  loop on a jetski fastgen model that was solid but unoriented, with
  about 8k faces.  NOTE: possibly already have a fix for this bug
  pending review and testing.  --CSM

* run make to create all primitives at the origin.  there are raytrace
  differences, crashes, and overall bad behavior between default
  ungrouped "e *" view and adding all of those primitives to a single
  region.  Script for creating all primitives via make command:

  for i in `mged -c test.g make asdf asdf 2>&1 | awk '{print $12}' | sed 's/[<>|]/ /g'` ; do mged -c test.g make $i $i ; done
  mged -c test.g "g primitives *"
  mged -c test.g r primitives.r u primitives
  rt -a 35 -e 25 test.g primitives
  rt -a 35 -e 25 test.g primitives.r

* MGED on Mac OS X 10.6 mouse bindings fail to zoom in incrementally.
  Pan, zoom out, and interactive zoom work.  Shift-grip bindings need
  to be verified.

* rt -i -s4096 -H16 -F/dev/null boolean-ops.g all

* asc2g of a .asc file containing the first line as a "#" comment
  results in a bu_malloc(0) bomb.  comment lines should be valid in
  .asc files.

* compilation reportedly fails when building Tcl if the path to the
  source directory includes a space.

* mged/tcl "gets" command doesn't work if you run the "source" command
  to read in a script that uses gets on stdin.  it just hangs until
  multiple sigint's are received.

* invoking mged from a system menu seems to be broken.  reported by a
  reliable ArchLinux user.  probably exhibits cross-platform on KDE,
  GNOME, or XFCE (but original environment is unknown).  the bug is
  undoubtedly related to mged's interactive and stdin/stdout input
  detection where mged thinks it's being scripted (beginning on or
  about mged.c:1150)

* Report of copy/paste not working with Tk on Windows Vista - does
  work on Windows XP.

* Update/create .mgedrc results in bogus qray lines if a db is not
  open when the file is created.

* MGED doesn't initialize multipane view to the documented multipane
  defaults if you just open a db with multipane already enabled.

* Creating an 'extrude' via MGED gui crashes MGED.

* edcomb command seems to be busted similar to red, doesn't seem to do
  anything.

* gqa manpage has the wrong indentation on the options write-up.

* Detailed BoT model ray trace image had incorrect background pixel
  values in framebuffer, off slightly by one.

* ogl display manager still crashes on attach in mged on Mac OS X 10.5

* rt is consuming 100% CPU in fb_close() on a lingering ogl
  framebuffer waiting to close.  needs to sleep or select for a while.

* /dev/oglb 'copy mode' fails to refresh the framebuffer

* X display manager (and probably others) will lock up mged if you Z/B
  while the wireframe is being actively drawn (in X_drawVList()).
  reproducible by cloning havoc a dozen or more times, getting
  sub-1sec fps, Z/B after rotating wireframe but before update
  finishes drawing.

* rtedge help shows redirection to a file working, but it does not
  seem to work (zero-file results).

* report of rt aborting randomly on 64-bit with camo shader.

* mged raytrace fails with simple "raytrace failed" message when using
  rt, nirt, etc. on Windows after modifying a region.  See sf bug
  report 2954404.

* mged rrt command is not respecting the view size, just does
  something similar to autoview but even gets az/el wrong too.

* Windows mged build is not generating tclIndex files with all
  entries.  notably the pattern_gui.tcl file is not getting included
  as well as a variety of others.  see missing entries in
  tclscripts/mged/tclIndex for examples:
  https://sourceforge.net/tracker/?func=detail&atid=640802&aid=2923199&group_id=105292

* vdeck truncates the region names in the 'regions' file, probably
  wrong bu_strlcpy() length.

* hyp tessellation routine creates bot that does not raytrace

* if you attach a display manager in tk-mged, kill the window, then
  quit, mged crashes on exit.

* CMD-b and CMD-f navigates the mged command window text widget on Mac
  and trivially causes problems with the cursor and text injection.

* mged segfaults when drawing lots of solids at once (e.g. draw *.s)

* mged command mode doesn't display the output of multiple
  semicolon-separated commands.  Only the output of the very last mged
  command is displayed.  Example:

  mged -c moss.g "tops ; title"
  mged -c moss.g "puts OK ; tops ; puts OK2 ; title ; puts DONE"

  Note that this is consistent with doing the same thing on the MGED
  Command Window's command prompt - if the above is a bug it is
  apparently not restricted to command/classic mode.

  Further investigation suggests that this is a consequence of
  Tcl treating multiple commands in a single " " block as one command,
  insofar as capturing the output is concerned.  It probably relates
  to situations where Tcl_SetResult is called where (in this case)
  Tcl_AppendResult would have been better, but it is not clear whether
  changing this result setting approach would cause other breakage.

* mged rotation halts after a few events and after zooming in/out at
  least once (Mac OS X 10.5)
  More info:  This problem appears to be specific to a particular
  X11 version or versions - it has been observed with
  xorg-server-1.4.2-apple33 on 10.5.  It does not occur with the
  community XQuartz 2.5.0 (xorg-server 1.7.6), nor does it occur
  when Tk 8.5.8 is compiled to use the Aqua backend. At least one other
  (possibly earlier) version of X11 from Apple on a 10.5 Mac did not
  exhibit bad rotate behavior. The problem is not specifically the
  zooming action - rather, the zoom triggers
  a Tk focus call on the display manager window (rather than the
  Tk parent window) and the problem manifests when focus is on the
  display manager window.  mged classic mode, where there is no
  parent window, exhibits the bad rotate behavior immediately.

* rtedge fills in nearly entire faces of an object when executing on a
  model that is in perspective mode. Effects really bad on small sized
  renders, and gets better the larger the render size. Does not happen
  in non-perspective ray-traces.

* infinite light sources close to objects result in unexpected/bad
  light behavior.  example is a 1m sph light about 5m from a box about
  1k x 1k x 1k in size.  setting make each light be red, green, and
  blue respectively, set to infinite (and fraction 10) and sides to
  the cube will render sides with a fixed color (green, blue) and top
  magenta.  On one test with all lights at fraction 1.0 and only one
  inf, was getting interleaved partial flat/no color all around cube.

* fastgen plate mode bots are reporting invalid shotline segments.
  depending on whether you shoot from the front or back can result in
  bogus lists visible with nirt in mged, entirely missing the bot, or
  mismatched entry/exit/overlap segments.

* tab completion in classic/command mode sometimes doesn't work
  (presumably if mged can't find tcl resources).  was finally able to
  reproduce this with the result given as:

  invalid command name "bind"

* mged 'facetize -n' command is giving many errors:
    ERROR: color index out of range (-128 > 255)
  and it shouldn't.  seems to be happening on 3-manifold surfaces but
  only for particular faces.  looks like the painting algorithm that is
  being used is flawed (and an incomplete implementation).

* running 'keep' on too many objects (hundreds is causing a usage
  statement (and not doing the keep)

* rtwizard won't display a line overlay if you line object is obscured
  by another.  Example being to display a vehicle as ghost, crew as
  solid, and crew as line.  confirmed even when using the other line
  occlusion settings.

* fbserv is receiving image data very slowly if the framebuffer is a
  remote X24 interface.  could reproduce on mac with the following:

  fbserv -s1024 1 /dev/X &
  rt -F1 -s1024 db/moss.g all.g
  rt -F1 -i -s1024 db/moss.g all.g  # 8x worse

* ls -A and/or e -A reportedly crashes on use, unverified

* File->Export seems to not work on Windows

* xpush fails to properly create a hierarchy killing existing objects
  resulting in dangling references. This may have been fixed by
  r69256, but a model that exposes the issue has to be tested to be
  sure.

* mged on Mac OS X will crash on an X11 call if compiled against
  system Tcl/Tk (at least on 10.4 linking against 8.4), see TODO

* the ray tracer fails to report LOS and sometimes even hits on a BoT
  that is marked as unoriented or is inverted (even though it
  sometimes will report the segments, just without LOS).  this seems
  to happen for bots that are neither lh or rh (they're mixed) and
  causes the raytrace to fail if they're set to anything other than
  'no' for orientation.  this affects rt, nirt, et al.

* g-vrml ignores bots that are created with dxf-g

* Mac OS X installer doesn't include the symbolic links, only
  including the versioned installation directory.

* Mouse cursor turns into a huge yellow-translucent arrow when running
  MGED through remote X11.

  Running this fixes the bug: .id_0t config -cursor "xterm black"

* mged (on debian) is reporting this when the user presses tab:
  Error in bgerror: invalid command name "::swidgets::togglearrow"
  the second time they press tab, they get:
  Error in bgerror: window name "_secErrorDialog" already exists in parent

  The problem was noticed on 7.12.4 and happens regardless of there
  being a db opened or not.  The problem does not occur in classic
  mode.

* mged command window is showing up as an empty 0x0 dimensioned window
  on freebsd using xfce and metacity (single display, 1680x1050 res).
  dm shows up just fine.

* g2asc of a v4 is apparently writing out the wrong s_cgtypes and
  otherwise parsing v4 .g files incorrectly

* bump shader seems to be busted.  only renders as flat by itself and
  stacked with phong it seems to ignore the uv settings.

* the graphics window does not acquire keybinding focus in mged on
  ubuntu linux.  key events go to the command window even with the
  graphics window has focus.

* MGED File->Preferences->Fonts results in error "named font "button
  _font" doesn't exist on Linux.

* mged tab-completion doesn't work if the object names have spaces

* Open dialog in MGED seems rather busted for scrolling and selection
  (seems to be Mac OS X specific)

* turning on Framebuffer->Rectangle Area in mged disables the
  embedded framebuffer display on Mac OS X when you sweep a raytrace
  rectangle.  sweeping still works if you leave it at All.

* invalid shaders result in an "Unable to locate where BRL-CAD ... is
  installed" message that refers to being unable to find a
  /lib/libSHADER.so file.  Quite a bogus message and probably not
  something that should abort the ray-trace.

* rtedge seems to get stuck in a loop at the end of rendering, never
  terminates (unless you rtabort/kill the process); works with -P1

* nirt/query_ray reports intersection messages in triplicate if the
  shot routines miss but still print out messages

* mged's plot tool (the gui one, not the 'plot' command) outputs a
  plot file that has an erase as the last command, causing empty plots

* interactive mouse edits in mged can go haywire on Mac OS X with
  snap-to-grid enabled

* overlap tool in mged encourages very inefficient CSG operations
  (e.g. a simple bolt with an entire hull armor subtracted)

* a character is (still) sometimes captured by the cursor box

* permute (in mged) requires apply or save before 'new' vertex values
  will function properly as keypoints.

* facedef then move face undoes the facedef in mged

* dsp primitive is going into what seems to be an unbounded memory
  allocation loop during prep, inside dsp_layers().

* crashes during photonmap raytrace of m35 where right node is invalid
  rt -o test.pix -V 1.0 -P4 -J0 -l7 -A0 ~/Desktop/m35.g all.g box.r sph.r

* libfb cannot utilize standard output/error "device" files without
  path trickery.  it recognizes the /dev/ prefix and presumes it's a
  libfb device instead of a filesystem device.

   OK: rt -F/./dev/stdout
  BAD: rt -F/dev/stdout

* libfb cannot open a file descriptor when piping/redirecting output.

   OK: rt -F/./dev/stdout moss.g all.g
  BAD: rt -F/./dev/stdout moss.g all.g > test.pix

* ray-tracers assume a seekable output stream.  libfb reports numerous
  fseek errors and outputs slightly corrupted pix data.

   OK: rt -o test.pix moss.g all.g && pix-png test.pix > test.png && rm -f test.pix
  BAD: rt -o /dev/stdout moss.g all.g | pix-png > test.png

* running any of the various mged commands (e.g. clicking the raytrace
  button many times quickly, or running rtarea on a large model) that
  output a lot of data will hang mged (idle cpu utilization).

* solids.sh regression test fails on Mac OS X with one pixel
  off-by-many on the edge of the middle ARB8 when using an optimized
  -O3 compile (-fno-unsafe-math-optimizations makes no difference)

* triangulation of some pipe in toyjeep.g fails (rather reliably) on
  some given bend.  fails in nmg_triangulate model().

* rtedge is not respecting the -c"set" options.  needs to give some
  indication that they were parsed too since it ignores unrecognized
  values apparently.

* fbhelp sends some of the output to stdout and some to stderr...

* debugbu 2 immediately reports a bu_vls_free() error.. apparently
  been a problem since 4.5 days.  this occurs because
  bu_memdebug_add() is not called unless memory checking is enabled
  via bu_debug.  so when bu_debug is set to 2 or 3, it ends up
  reporting errors during bu_free() for items that were allocated and
  were not yet being tracked.

* mged View menu says I and O are keybindings for zoom in/out -- they
  are apparently not.

* raytracers report the wrong amount of cumulative time (reports as
  0.0 seconds elapsed) on amd64 linux; most likely the same pthread
  accounting bug visited a couple years ago.

* Create menu.. create "part" boo hiss..

* wavelet isn't working correctly on a simple image decomposition &
  reconstruction.

* query_ray can't find nirt if not installed

* fbed is looking for /usr/lib/vfont/nonie.r.12

* mged doesn't check if html_dir does not exist when searching for
  docs/browser.

* permute then translate doesn't refresh the graphics window
    permute 4378
    translate menus
    p 0 0 0
    p 0 0 0

* constrained axis translation bindings on OS X don't work
    object/matrix edit
    shift-alt mouse (left/mid/right for x/y/z respectively) should work

* the findfont directive in gv postscript is reporting font not found

* concave arbs give correct wireframe but do not raytrace correctly.

* rtwizard/rtedge creates incorrect "fuzzy" edge overlay when creating
  a "Ghost Image with Insert and Lines" that includes a light source
  and object close to surface (e.g. moss.g; select plane, light, and
  cone for ghost; select cone for insert and lines.)

* mged's solid illuminate doesn't work when Lighting is turned on

* raytrace of zoomed images (e.g., -s64) doesn't work for if_X24
  framebuffers.  zooming logic seems to be broken (was crashing in
  memcpy, but that was fixed) for at least -s96 and smaller:

  ./libtool --mode=execute gdb --args src/rt/rt -F/dev/X -P1 -s64 db/moss.g all.g

* g2asc exports attr lines but will not grok them on asc2g

* mged's matrix selection does not modify the correct matrix if there
  are multiply referenced non-unique paths (e.g. referencing the same
  object N times in a combination).

* facetize -t doesn't work (tnurb support)

* on os x, the -fast option results in an odd optimization/aliasing
  bug where rtip contents are lost in rt after returning from
  rt_gettrees().

* tra in console mode outputs a warning about mged_players

* mged primitive editor doesn't accept/apply values to disk for
  certain primitives (e.g. sph)

* photon map cache file doesn't work (seems to crash rt on use)

* enabling perspective mode horks shaded mode, also Z clipping
  shouldn't be required to get the shading.

* an underlay framebuffer does not underlay with shaded mode

* rtweight chokes on a .density file that has less than 3 values per
  line (infinite loop).  also gives infinite areas when presented with
  an empty .density file.

* rtarea gives incorrect areas when using -H hypersample flag

* rtweight/gqa volumes seem wrong with simple sphere test case, with
  values consistently and quickly converging less than the actual
  volume regardless of the grid size.

* setting src/librt/db5_io.c's AVS_ADD off in db5_import_attributes
  causes bad things to happen and it really shouldn't.

* vrml exporter (and probably others) doesn't export primitives
  directly as one might expect like the ray-tracers, only regions.

* sketches extruded with non-square AB vectors results in inner sketch
  components not getting scaled properly.

* rtcheck is ignoring fastgen tagged geometry since no multioverlap
  handler is set.

* bot-bldxf fails in db_walk_tree(), interface may have changed


Older Bugs
----------

* several manpages are missing or out of date for various tools

* X11 framebuffer often fails to display on certain middle bit depth
  displays. (e.g. 15 or 16 bit depth)

* mk_lcomb fails on large combinations due to mk_tree_pure() building
  a left-heavy tree and wdb_put_internal() then recursively performing
  lookup and put calls.


Annoyances
----------

* matrix edits on objects in mged require you to select a primitive
  for a coordinate system reference.  this can be counterintuitive and
  confusing.

* there are frequently modality errors when entering edit mode as
  control and middle mouse rebind to model edit instead of view edit.

---
Bugs should be reported to the bug tracker on the project website at
http://sourceforge.net/tracker/?atid=640802&group_id=105292

BUGS should be formatted to column 70 (M-q in emacs), no tabs.

// Local Variables:
// mode: Text
// fill-column: 70
// End: