Wizard Chronical

This list contains tidbits of information that are so helpful once you know it, but is generally impossible to find.

• Purify not show enough context to call ??/??/1998
  If you write a fcn that does a new [] in g++, this will eat up the default call chain of 6. This will result in not showing what called your fcn, making it useless to fix.
  
  Solution: Use the -chain-length=7 option to purify at link time to increase the value from 6.
  
• Redefinition of __P() ??/??/1998
  Staff must have changed something recently because whenever I compile with gcc/g++ with full warnings turned on (-Wall), I get complaints about /usr/imports/sys/cdefs.h redefining __P.
  
  Solution: Simply include the offending <sys/cdefs.h> before you include anything else. All later includes check before defining __P.
  
• Unused parameter warnings ??/??/1998
  I always compile with -Wall to turn on all warnings, because I like clean code. Problem is for many fcns, they must have certain paramaters (as in signal handlers), so these almost always generate warnings.
  Solution: Leave the type for the parameter, but remove the variable name from the parameter list. Warnings go away for unnamed parameters.
  
• G++/GCC not warning about shadowing 03/16/2000
  Most people think that -Wall in gcc/g++ takes care of turning on all the warnings the compiler can issue when compiling some code. That's not true, and it's particularly nasty in situations where shadowing may be happening (e.g. a local variable shadows a class member). If this happens, the programs seems to perform in a very weird fashion. Therefore, it's a good idea to prevent these errors by turning on -Wshadow (remember also to read the man page for other important warnings that are not included in -Wall that may be important for you). (added by Henrique Andrade (hcma@cs.umd.edu))
  

• Funny errors happen when I'm running Insure 02/01/2001
  Running your code under Insure seems to cause the program to be executed differently. In some extreme cases the program may die because of that.
  

  Explanation: Insure instrumentation requires creating temporaries in certain cases. Because of that, copy constructor, assignment operator and destructor could be called more often in instrumented code.

  Provided these functions are implemented correctly, there is no change in the (overall) program behavior.

  If you do not want to implement e.g. copy constructor, you can declare it private, and insure will not try to make a temporary (but will not be able to perform certain checks either).

  If you do not make it private, but simply "expect" it not to be called, you are following a bad programming practice: even if you are careful enough never to write any code that will implicitly call the copy ctor, can you make the same promise for all the people who will be maintaining your code 10 years from now?

  Solution: Create your class infra-structure completely (copy contructor, assignment operator, and destructor!) (added by Henrique Andrade (hcma@cs.umd.edu) - based on personal message exchanged by Chialin Chang and Parasoft Tech Support)
  

• Xterm failed to display ??/??/1999
  When remotely displaying an xterm from the class cluster machines, the shell prompt would just return, and a corefile of size zero.
  Solution: There is some very odd interaction with csh limits. Unlimiting filesize, results in no core file, and xterms working.
  
• My X11 login freezes 10/23/2000
  I was running gnome on Solaris 2.6, and when logging in fresh, the X11 stipple background was visible, as was the xconsole window. Many of the processes were in the STOPPED (T) state, and a kill -CONT would wake them up. The question was why were they in this funny state, and how to fix it?
  Solution: I was using a script xallow.ssh that used an ssh shell to copy the new X11 cookie to remote home directories for some of my accounts. One machine I was xallow.ssh'ing into was tsunami. The problem was tsunami was not allowing logins, and the ssh that previously worked for years, was now failing and prompting for password. The call to xallow.ssh was a background process, which is forced into the STOPPED (T) state by the os when tty output is detected. It also appears that all processes in the process group were affected, thus freezing all of X.
  

• Printing a PDF file ??/??/1999
  The output has messed up chars all over the page.
  Solution: Unselect the "download at once" box in the print dialog for acroread. (courtesy uysal@cs.umd.edu)
  
• Java Plugin use on UNIX 11/15/1999
  The plugin seems to start, but fails to start the actual jdk and display the applet.
  Solution: This was indirectly caused by staff installing a version of Sun's plugin in /usr/local/lib/netscape. If this is the first plugin used, the ${HOME}/.java/properties file the plugin creates will not be usable if the JDK entry is not correct. Solution is to rename or delete the properties file, and make sure the desired plugin is being started. (debugged by myself and chialin@cs.umd.edu)
  
• Ssh for remote progs with keyboard input 01/07/2000
  I usually start an xterm and in the xterm, rsh or ssh to the remote machine and run a prog. For rsh, hitting Ctrl+C will cause the remote prog to be send SIGINT and exit. For ssh, the Ctrl+C causes the ssh process on the local machine to exit, but not the remote prog. This left many many processes to kill when I finally noticed them.
  Solution: Apparantly rsh always allocates a pseudo tty for non interactive remote shells. Ssh does not by default. Solution is to pass -t to the ssh command to create such a pseudo tty. In this case a Ctrl+C is instead sent over the pseudo tty and SIGINT is sent to the remote prog process.