1   otrace FAQ

1.1   What platforms does otrace support?

otrace is written purely in python, but it uses OS-specific calls for file, shell, and terminal-related operations. It has been tested only on Linux and Mac OS X, but the demo program also runs on Windows.

1.2   Can all unix shell commands be used with otrace?

No. Many unix shell commands can be used when the current directory is not /osh/*. However, each unix shell command is executed in a separate shell process. Full-screen commands like vi will not work, but you can use the oshell edit and view commands instead. Unix shell aliases may also not work, but you can use oshell aliases.

1.3   Can I use pdb with otrace?

The trace -a <action> -c <condition> <function> command allows pdb (or ipdb) as one of the action options. This options switches control to pdb when the trace condition is satisfied. (To return control to otrace, use the continue command of pdb.)

1.4   Can otrace be used in production code? Is it secure?

Until tracing is initiated, otrace simply acts like a loaded, but unused, module. There is no performance penalty. The GNU screen utility can be used to provide a detachable terminal for using otrace with server processes. This terminal can only be accessed by the user who owns the server process. Anyone who is able to login with privileges to access the otrace terminal can directly execute unix shell commands in any case. However, the features that make otrace useful also make it easier for someone who has acquired unauthorized login privileges to examine variables in a running program, or even modify code, without having to edit files and restart the program. This can be prevented by using the lock command to password-protect the otrace terminal. (The safe_mode parameter can help prevent inadvertent modification of the running program.)

1.5   Does otrace create true “snapshots” of variables?

No. For efficiency, by default, otrace retains only a “shallow copy” of the dictionary containing arguments and/or local variables. Entities in this shallow copy could change their value as execution progresses. The parameter deep_copy can be set to force otrace to retain deep copies of built-in object types and classes that implement the __deepcopy__ hook.

1.6   Why isn’t TAB completion working for me?

The GNU readline module needs to be installed for TAB completion to work. On Mac OS X, the pre-installed readline module may need to be replaced using the easy_install readline command. On Windows, use easy_install pyreadline instead. (easy_install is available via the setuptools package.)

1.7   Is there a graphical front-end to otrace?

Yes, see GraphTerm (Note: If you browse the code for otrace, you will notice the hooks used by the front-end, which could potentially be used by other graphical front-ends as well. otrace will always remain accessible directly via the command line.)

1.8   What is monkey patching? Is it a good thing?

“Monkey patching” is a term used to refer to modification of code while a program is running. It is generally considered a bad practice to use in production code, although it is (or was) apparently an accepted practice in some programming environments. It is not very commonly used in Python. otrace makes it extremely easy to “monkey patch” your program, but this feature is not meant to be used in production code. However, it can be really useful when debugging a stateful program, because you can keep modifying code as you learn more about the bug you are tracking, while retaining the program state that triggers the bug. (Adding print statements to live code as you track a bug may perhaps be the most useful feature of otrace.)

1.9   What is object tagging?

The idea behind object tagging is that you can put a “marker” on an object, by adding a special attribute, _otrace_tag, indicating that it should be traced. Any function or method invocation where a tagged object is present in the argument list can be automatically tracked by otrace. Thus, we can follow a “rogue object” as it makes its way through the application. If a remote procedure call (RPC) implementation preserves the special tagging attribute, then tagged objects could also be tracked across different hosts running otrace, facilitating the debugging of distributed applications. (It remains to be seen how useful this feature will be in practice.)

1.10   What is /osh/web?

For web servers using bi-directional websockets to communicate with the browser, hooks are provided in otrace to allow the mapping of each currently connected browser to the virtual directory /osh/web/username. Once in that directory, you can type any Javascript expression which will be sent to the browser to be evaluated and the result will be displayed back. So the otrace console temporarily acts like the Javascript console on the user’s browser (e.g., like Firebug). See the Tornado-based demo program tornademos/chat_websock.py for Python and Javascript “glue code” that is needed to accomplish this.

1.11   Does otrace work with Python 3?

It appears to work when ported using the 2to3 tool, but no real testing has been done.