• A.6.1 Bugs in vem
• A.6.2 Bugs in pigi
• A.6.3 Bugs in tycho
• A.6.4 Code generation bugs
• A.6.5 Bugs in pxgraph
• A.6.6 HPPA specific bugs
• A.6.7 IBM AIX specific bugs
• A.6.8 Silicon Graphics IRIX5 specific bugs
• A.6.9 Linux specific bugs
• A.6.10 Sun Solaris 2.4 specific bugs
• A.6.11 Sun OS4 specific bugs
• A.6.12 DEC Alpha specific bugs
• A.6.13 GNU compiler bugs

A.6.1 Bugs in vem

• There is no interlock between commands started by Vem and commands started by pigiRpc. If either process requests the other to perform some operation while the other process is trying to request the first one to do something, they end up in a "deadly embrace". The usual result is that pigiRpc exits with a complaint about a protocol error. (Fortunately, Vem continues to run, so you can at least save your schematics before restarting.) Examples of the problem include trying to issue edit-parameters or look-inside commands (which are invoked in a Vem window but require cooperation from pigiRpc) while pigiRpc is compiling a facet or running a simulation with graphical animation turned on (which require Vem to react to requests from pigiRpc). The simplest rule that will keep you out of trouble is "don't do anything in Vem windows while pigiRpc is busy".
• Deleting wires that are too long (i.e., end of wire is past the terminal, but near it) can result in vem dumping core.
• Labels that end with a carriage return are neither displayed nor printed correctly.
• Editing icons stimulates a memory leak in vem that can make the process grow quite big. After editing a few dozen icons, you may wish to exit vem and restart.
• The "layer" command only works when editing icons, not when editing schematics.
• The "copy-objects" command will copy from one window to another only when editing icons, not when editing schematics.
• vem or pigiRpc may fail to start up on 4 bit (16 color) screens. At UC Berkeley, we use 8 bit and 24 bit screens. Black and white (1 bit) support in vem is a little weak. If you are on a black and white screen, you may need to modify $PTOLEMY/lib/pigiXRes9. (Note that $PTOLEMY/lib/pigiXRes9.bw is read when the -bw pigi option is used to create black and white screen dumps from a color monitor. pigiXRes9.bw has very little to do with running on a black and white screen).
• If the VEMBINARY environment variable is set, then bin/pigiEnv.csh will use the contents of that variable as the vem binary. However, the binary must be named vem, or there will be minor problems with X resources, such as small fonts and incorrect snap. Most users will never use the VEMBINARY environment variable. It is primarily used to debug vem.
• If your X server runs out of colors, vem may crash. One workaround is to not run color
  hogs like Netscape, or to limit the number of colors Netscape can use.
• Bug fixes to vem now mean that vem is more picky about text label heights. Currently, the range for text label heights is 0 to 80. If you select text and then use E, to edit the label, you may see an error if your text height is greater than 80. One workaround is to delete the text and reenter it with a height that is within the proper range. Another workaround is to use an old vem binary to edit the height and set it to within the range.
• If you re-read a facet that has an open run window, you won't be able to dismiss the run window. The workaround is to open a new run window.

A.6.2 Bugs in pigi

• It is possible to make more than one icon to represent a given star. For instance, the icon And and And.input=2 refer to the same star, but the former can have any number of inputs, while the latter has exactly two inputs. As of this writing, this facility does not work for galaxies. You should create only one icon for each galaxy.
• $PTOLEMY/bin/pigi is a link to $PTOLEMY/bin/pigiEnv.csh which may add to your path. Under certain circumstances, you may get the message `Warning: ridiculously long PATH truncated'. To work around this problem, try shortening your path. One trick is to use shorter pathnames in your path, perhaps using symbolic links.
• If your X server runs out of colors, pigi may crash. One workaround is to not run color hogs like Netscape, or to limit the number of colors Netscape can use.
• The compile-SDF target fails if a galaxy has incrementally linked in stars. The problem is that the .h file for the incrementally linked in star cannot be found by compile-SDF. As a workaround try editing the makefile in ~/PTOLEMY_SYSTEMS and adding the directory where the incrementally linked stars are. Then build the binary by hand.
• The compile-SDF target cannot handle star parameters that use the Tcl Interpreter to evaluate Tcl commands.
• On some machines, the SDF Matlab demonstrations will appear to hang Ptolemy. This is a known bug in the Matlab external interface library that requires that the Matlab process be attached to a terminal. There are two workarounds: (1) run Ptolemy in the foreground, or (2) manually control the Ptolemy interface to Matlab by using the matlab command in ptcl (see 3.9.10 on page 3-16) via the Ptolemy console. In both workarounds, the Matlab process will be attached to a terminal.
• Ptolemy0.7 will not work with Matlab5 or Mathematica3. We are working on upgrading the interface. In the short term, you must use Matlab4 or Mathematica2.2.
• The Networks Of Workstations active messages (NOWam) CGC target is slow, but it does run under Solaris. However these demos will not work in an environment that requires Kerberos for rsh-ing jobs.
• To use the Dynamic Dataflow Code Generation (cgddf) CGC target, a file must be recompiled. The changes to parScheduler.cc are a fairly radical shift in how Ptolemy handles disconnected graphs. These changes could break other parallel schedulers. The script $PTOLEMY/bin/mkcgddf will build a pigiRpc.ptrim.debug binary that contains the cgddf target. This script is basically untested, but should work.
• Mixed CGC/VHDL/TclTk demos leave ptvhdlsim running after exiting.
• Tom Lane pointed out the following problem in 0.7:
  An initializable delay attached to any multiporthole output will fail. For example "Fork -> Printer", if you use the double-arrowed form of the Fork icon and put an initializable delay on the arc.

A.6.3 Bugs in tycho

• If you start up tycho with Ptolemy in the background (i.e, tycho -ptrim &), then exiting from the Matlab console may hang. The workaround is to place your tycho process into the foreground. This is a known bug with Matlab.
• See $PTOLEMY/tycho/doc/bugs.html for a more complete list of Tycho bugs.

A.6.4 Code generation bugs

• The CGC multirate filterbank and CGC chaoticBits demos bring up messages about initializable delays.
• If you try to run the CGC demos that generate sound, and you do not have a properly configured audio device, or you do not have write permission to /dev/audio, then you get a cryptic and uninformative error message, rather than something useful. The CGC tremolo demo will probably only work on a Sun SPARCstation running SunOS4.1.3 or Solaris2.x. The CGC alive demo will only work on an SGI.
• The use of initializable delays and variable delays in the code generation domains may result in incorrect code being generated. Variable delays are delays which depend upon the value of a galaxy or universe parameter. The workaround is to use regular delays of a constant value (i.e. "4") in the code generation domains. Note that initializable and variable delays work fine in the simulation domains.
  
  For example, the CGC:multirate:filterbank demo produces initializable delay warnings. The output from the CGC:multirate:filterbank demo is different than the output from the SDF:multirate:filterbank demo in that the original and reconstructed signals don't overlap as much in the CGC version.

A.6.5 Bugs in pxgraph

• If pxgraph is given exceptional numeric input, such as the IEEE floating point Inf,
  -Inf, or NaN, then it displays a blank window only.
• If pxgraph is given an empty file to plot, all it does is send a message "problems with input data" to the standard output.
• There is no way to produce hardcopy without running pxgraph interactively.
• Specifying the colors for data sets using X resources does not work.

A.6.6 HPPA specific bugs

• If you are rebuilding Ptolemy on the hppa, you must have X11 installed. Apparently, HP ships machines without most of the X11 include files. The prebuilt binaries also use libraries from X11R6. See ftp://ptolemy.eecs.berkeley.edu/pub/ptolemy/contrib/hpux.
• If you are building the GNU tools under HPUX10.20, you will also need GNU sed, which is downloadable from the Ptolemy ftp site in ftp://ptolemy.eecs.berkeley.edu/pub/gnu/hppa and you may need to patch HPUX. See $PTOLEMY/src/gnu/README for details.
• If you are using the prebuilt binaries under HPUX10.20, then you will need to install X11R6 shared libraries. See ftp://ptolemy.eecs.berkeley.edu/pub/ptolemy/ptolemy0.7/README.hpux
• The CGC demo animatedLMS demo dumps core upon start-up.
• On the hppa, under the HP C++ compiler, Xhistogram may have rounding problems.
• On the hppa, the CGC animatedLMS demo requires editing of Target Parameters to compile.
• On the hppa, the CGC animatedLMS demo gets an Arithmetic Exception and core dumps.
• On the hppa, the SDF animatedLMS demo generates a "non-numeric value" error in Tcl if the step size is so large that the system is made to go unstable.
• Compilation under older versions of hppa.cfront may fail. The problem seems to be with the +A ld command line argument. You must use the +A ld argument when compiling pigiRpc and ptcl, or incremental linking of new stars will fail. However, use of the +A argument produces the warning:
  
  CC: error: could not find __head symbol. You must use CC to link. If your main is not in C++, a call to _main() is required.(740)

  Even if this warning is printed, a viable binary is still produced. The +A ld argument is not necessary for other binaries, such as vem and pxgraph.
• If you are running an older version of the HP Cfront compiler, you may need to add
  -DPOSTFIX_OPT= to your c++ command line. See config-hppa.cfront.mk.
• If you are having problems incrementally linking in stars, and you are getting messages like:
  /bin/ld: (Warning) Inter-quadrant branch in XXX
where XXX is the name of the .o file you are trying to link in,
  then you may need to be compiling your stars with the proper level of optimization. See "Dynamic linking and makefiles" on page A-31 for more information.
• If you have problems with ld under HPUX, you should try patching your operating system with a patch from HP. Try http://europe-support.external.hp.com if you are in Europe, http://us-support.external.hp.com if you are anywhere else.
• If, under HPUX9.x, you get a message like:
  ld: fatal: relocations remain against allocatable but non-writable sections
Then you may need to apply a patch to ld.
• To incrementally link new stars under HPUX9.x, you will probably need to supply a make.template file. See "Dynamic linking fails" on page A-30.
• The multilink command seems to trigger Inter-quadrant branch messages, and then hang pigi. We are not sure why.
• HPUX10.x will build a pigiRpc with the Process Network (PN) domain, but the event loop is broken. The PN domain is not part of the default hppa build, so this problem will not affect most users.
• If you are under HPUX10, then building the PN domain requires Distributed Computing Environment (DCE) threads. You will need to install the DCE development set of the OS CDs. If you don't have a /usr/include/pthread.h, then you probably don't have the DCE development set installed.

A.6.7 IBM AIX specific bugs

To generate gcc-2.7.2 under AIX3.2.5 with cc, you need the PTF U436313. The PTF U432238 is suggested in the README.RS6000 file from the gcc-2.7.2 distribution, but this PTF is superseded by the PTF U436313. This PTF allows you to upgrade the cc IBM compiler from the release 1.3.0.0 (delivered with AIX 3.2.5) to the release 1.3.0.33 which compiles gcc-2.3.6.

A.6.8 Silicon Graphics IRIX5 specific bugs

• Linking of a full size pigiRpc binary can fail with a `GOT Overflow' message. The problem is that pigiRpc has too many symbols. The Irix dso man page suggests using shared libraries, but gcc-2.7.2 may not support C++ shared libraries under Irix. The workaround is to run ptrimRpc rather than pigiRpc.
• Installation of the xv man pages will fail. Irix does not have nroff and tbl.
• To build the CGC demos, you may need to exit pigi, set the environment variable SGI_CC to -cckr:
  setenv SGI_CC -cckr
and restart pigi. See the sgi cc man page for more information.
• sdf:image:motionCompensation demo has some odd looking blocks in the McompOutput.1 image. Some of the images from the irix5 run of this demo do not look like images in the sun4 version of this demo.

A.6.9 Linux specific bugs

• Linux uses GNU ar, which seems to have problems if it is run on two files whose names are not unique in the first 13 characters. We have renamed a number of stars to workaround this problem. The make checkjunk command will report the names of files that are not unique in the first 13 characters. You should only have problems with this if you are creating stars of your own and placing them in libraries.
• If you are running the Linux version on a standalone machine, then please refer to section "Mr. Ptolemy window does not come up" on page A-16.
• Richard Nicholls <R.Nicholls@mmu.ac.uk> reports that the ld with GNU binutils 2.5.2 does not support dynamic loading (See the TODO file distributed with the GNU binutils source). One solution is to use an older version of GNU ld. Another solution would be to use dlopen() style incremental linking, see $PTOLEMY/src/kernel/Linker.sysdep.h.

A.6.10 Sun Solaris 2.4 specific bugs

• If you are building gcc under Solaris, you must have /bin in your path before /usr/ucb see $PTOLEMY/src/gnu/README.
• If, while linking, you get a message like
  unable to locate archive symbol table: Format error: archive fmag
then you have probably run out of swap space. See the Solaris swap command for how to add more swap with mkfile.
• Under Solaris 2.5 with SunSoft CC 4.1, src/kernel/MatrixParticle.cc fails to compile. The fix for this may require substantial reimplementation of the Ptolemy Matrix class, so the changes did not make it into the 0.7 release.
• The Utah Raster Toolkit does not come with a Solaris2.4 configuration file. We are running SunOS binaries in-house. See "Other useful software packages" on page A-14 for more information.
• The GNU make binary built for Solaris2.5 will not work with Solaris2.4. If you try to run the Solaris2.5 GNU make binary under Solaris2.4, you will get an error message like:
  ld.so.1: make: fatal: relocation error: symbol not found: setlinebuf: referenced in make

A.6.11 Sun OS4 specific bugs

• If you are using the prebuilt GNU binaries in pt-0.7.gnu.sun4.tar.gz, then you may have problems if your Ptolemy distribution is not at /users/ptolemy, and you try setting the GNU environment variables with the $PTOLEMY/bin/g++-setup script. We are working towards a solution, but you might find the following variables will work for you:
  unsetenv GCC_EXEC_PREFIX
unsetenv C_INCLUDE_PATH
unsetenv CPLUS_INCLUDE_PATH
setenv LIBRARY_PATH $PTOLEMY/gnu/$PTARCH/lib
setenv COMPILER_PATH $PTOLEMY/gnu/$PTARCH/lib/gcc-lib/$PTARCH/2.7.2.2
setenv GCC_INCLUDE_DIR $PTOLEMY/gnu/$PTARCH/lib/gcc-lib
We are working on a solution for this.
• Under SunOS4.x, we use the older BSD ld style incremental linking because of problems with g++ shared libraries. This means that the SunOS4.x binaries must be linked statically instead of dynamically, so the binaries are much larger than on platforms that use dlopen() style incremental linking. It also means that only the symbols that are used by stars present at link time are actually present in the binary. This can be a problem if your incrementally linked star uses a symbol from libg++.a that is not used by a star present at link time. See Appendix D, "Shared Libraries" for more information.

A.6.12 DEC Alpha specific bugs

• The DEC Alpha port is a new port that has had very little testing, so there are bound to be 64bit vs. 32bit bugs
• The CGC fixed-point demos give incorrect results. This is most likely due to the fact that the DEC Alpha is 64 bits.
• SDF/DDF Wormholes cause SIGFPE signals, which crash pigi. The DDF ifThenElse demo uses such SDF/DDF Wormhole.
• The full pigiRpc may fail to start with messages like:
  /sbin/loader: Fatal Error: lazy_text_resolve: symbol malloc should not have any relocation entry
The workaround is to do:
  setenv LD_BIND_NOW yes
See the DEC Unix loader man page for more information about LD_BIND_NOW. You may see similar error messages when you run the CGC fixed point demos.
• Vem produces lots of Unaligned Access messages. You can run
  uac p noprint
to turn them off in the current shell.

A.6.13 GNU compiler bugs

• If you write your own stars, then be aware that gcc-2.7.2 may not choose the expected cast. In particular, g++ has been known to cast everything to Fix if the explicit cast is omitted. The arithmetic is then performed using fixed-point computations. This will be dramatically slower than double or integer arithmetic, and may yield unexpected results. It is best to explicitly cast states to the desired form. For more information, see the ptlang chapter in the Ptolemy Programmers Manual.
• If you already have GNU make installed, and the binary is called gmake, then the installation of the Ptolemy GNU tools may fail. The workaround is to create a link in your path from make to gmake.