[Python-Dev] Patch review: [ 1098732 ] Enhance tracebacks and stack traces with vars

[Phillip J. Eby]

At 08:20 PM 2/9/05 +0100, BJörn Lindqvist wrote:
>Does Skip's idea have
>any merit?

Yes, but not as a default behavior.  Many people already consider the fact 
that tracebacks display file paths to be a potential security problem.  If 
anything, the default traceback display should have less information, not 
more.  (E.g., display module __name__ instead of the code's __file__).

Also note that the stdlib already has a cgitb module that does this sort of 
display for CGI scripts, so the technique isn't new, and cgitb provides a 
good example for people to create their own advanced traceback formatters with.

If there were another command line option added to Python for this, I'd 
personally prefer it be an option to enter the debugger when a terminal 
traceback is printed.  Currently, I use 'python -i' so that I get an 
interpreter prompt, then use 'import pdb; pdb.pm()' to enter the debugger 
at the point where the error occurred.  One can then print whatever local 
variables are desired, go up and down the stack, list code, and even 
perform calculations on the values on the stack.

About the only place I can think of where such an extremely verbose 
traceback would be useful and safe, is inside of unit tests.  I believe 
that the py.test package uses traceback introspection of this kind in order 
to display relevant values when an assertion fails.  So, it might be useful 
in the context of a unit test error report to get some of that information, 
but even there, there is a question of how much is relevant for display.