Combining Embedded and Interactive Python in a Debugger: A Case Study
[EuroPython 2011] Caroline Tice - 23 June 2011 in "Track Ravioli""/>
LLDB is an open source compiler currently under development. LLDB incorporates Python in several different ways: From the command line, one can drop into an interactive Python script interpreter, entering normal Python interactive commands and having them interpreted and executed. In addition, the LLDB debugger has a programming API that allows users to call actual debugger functions and access real debugger objects. This API is run through SWIG and converted into a Python module that gets automatically imported into the interactive interpreter inside LLDB. The upshot of this is that, in the Python mode, users can call the API functions as if they were regular Python functions. This is turn means that users can do things like hit a breakpoint, call the appropriate API functions to get the actual current frame object, check to see if the caller function several frames up the stack is a particular function, and if so, perform some other interesting action. LLDB also allows for breakpoint commands to be written in Python (which therefore allows them to call the API functions), as well as having a command that interprets and executes a single line of Python.
Because the API and all of the debugger functionality is available in a Python module, it is also possible to write a debugger front end completely in Python and have it import the LLDB module and call the LLDB API functions (which calls into the LLDB shared library), to create a new debugger front end. Our test suite takes advantage of this to run tests on the LLDB library without having to start up an interactive LLDB debubgger session.
There were many interesting problems encountered and overcome in getting all of this to work. LLDB combines both interactive Python and embedded Python. There were some interesting things worked out so that a single Python environment dictionary is used for the entire debugger session: A user can pop into and out of the interactive interpreter and have access at any time to anything defined in any previous visit to the interactive interpreter (within the same debug session); breakpoint commands have access to everything defined in the interactive session; so do embedded script commands. It gets even more complicated, because LLDB can have multiple debugger sessions alive at once, and each debugger session has its own Python interpreter with its own dictionary and must not be allowed to interfere with another session's state. In this talk I will discuss how we implemented many of these features in LLDB, pointing out some of the more interesting problems we encountered and how we overcame them, and demonstrate it all working in LLDB.