GDB-like Python Debugger in the Trepan family
This is a gdb-like debugger for Python. It is a rewrite of pdb from the ground up. It is both a high-level debugger as well as a lower-level bytecode debugger. By lower-level debugger, I mean that it understands a lot about byte code and will try to make use of that in its normal higher-level instructions.
A command-line interface (CLI) is provided as well as an remote access interface over TCP/IP.
This package is for Python 3.2 and above. See trepan2 for the same code modified to work with Python 2.
There’s a lot of cool stuff here that’s not in the stock Python debugger pdb, or in any other Python debugger that I know about.
Python reports line information on the granularity of a line. To get more precise information, we can (de)parse into Python the byte code around a bytecode offset such as the place you are stopped at.
So far as I know, there is no other debugger that decompile code at runtime.
See the deparse command for details.
We use information in the line number table in byte to understand which lines are breakpointable, and in which module or function the line appears in. Use info_line to see this information.
In the future we may allow specifiying an offset to indicate which offset to stop at when there are several choices for a given line number.
You can pass the debugger the name of Python bytecode and many times, the debugger will merrily proceed. This debugger tries very hard find the source code. Either by using the current executable search path (e.g. PATH) or for some by looking inside the bytecode for a filename in the main code object (co_filename) and applying that with a search path which takes into account directory where the bytecode lives.
Failing to find source code this way, and in other situations where source code can’t be found, the debugger will decompile the bytecode and use that for showing source test. This allows us to debug `eval`’d or `exec’’d code.
But if you happen to know where the source code is located, you can associate a file source code with the current name listed in the bytecode. See the set_substitute command for details here.
Terminal source code is colorized via pygments . And with that you can set the pygments color style, e.g. “colorful”, “paraiso-dark”. See set_style . Furthermore, we make use of terminal bold and emphasized text in debugger output and help text. Of course, you can also turn this off. You can use your own pygments_style, provided you have a terminal that supports 256 colors. If your terminal supports the basic ANSI color sequences only, we support that too in both dark and light themes.
GNU readline command completion is available. Command completion is not just a simple static list, but varies depending on the context. For example, for frame-changing commands which take optional numbers, on the list of valid numbers is considered.
We can adjust debugger output depending on the line width of your terminal. If it changes, or you want to adjust it, see set_width .
If you want to evaluate the current source line before it is run in the code, use eval. To evaluate text of a common fragment of line, such as the expression part of an if statement, you can do that with eval?. See eval for more information.
Many Python debuggers only allow setting a breakpoint at a line event and functions are treated like line numbers. But functions and lines are fundamentally different. If I write:
def five(): return 5
this line means has three different kinds of things. First there is the code in Python that defines function five() for the first time. Then there is the function itself, and then there is some code inside that function.
In this debugger, you can give the name of a function by surrounding adding () at the end:
Also five could be a method of an object that is currently defined when the breakpoint command is given:
Sometimes you want small steps, and sometimes large stepping.
This fundamental issue is handled in a couple ways:
There are now step event and next event commands with aliases to s+, s> and so on. The plus-suffixed commands force a different line on a subsequent stop, the dash-suffixed commands don’t. Suffixes >, <, and ! specify call, return and exception events respectively. And without a suffix you get the default; this is set by the set different command.
By default the debugger stops at every event: call, return, line, exception, c-call, c-exception. If you just want to stop at line events (which is largely what you happens in pdb) you can. If however you just want to stop at calls and returns, that’s possible too. Or pick some combination.
In conjunction with handling all events by default, the event status is shown when stopped. The reason for stopping is also available via info program.
I’m not sure why this was not done before. Probably because of the lack of the ability to set and move by different granularities, tracing calls and returns lead to too many uninteresting stops (such as at the same place you just were at). Also, stopping on function definitions probably also added to this tedium.
Because we’re really handling return events, we can show you the return value. (pdb has an “undocumented” retval command that doesn’t seem to work.)
There are debugger macros. In gdb, there is a macro debugger command to extend debugger commands.
However Python has its own rich programming language so it seems silly to recreate the macro language that is in gdb. Simpler and more powerful is just to use Python here. A debugger macro here is just a lambda expression which returns a string or a list of strings. Each string returned should be a debugger command.
We also have aliases for the extremely simple situation where you want to give an alias to an existing debugger command. But beware: some commands, like step inspect command suffixes and change their behavior accordingly.
We also envision a number of other ways to allow extension of this debugger either through additional modules, or user-supplied debugger command directories.
We do more in the way of looking at the byte codes to give better information. Through this we can provide:
a skip command. It is like the jump command, but you don’t have to deal with line numbers.
disassembly of code fragments. You can now disassemble relative to the stack frames you are currently stopped at.
Better interpretation of where you are when inside execfile or exec. (But really though this is probably a Python compiler misfeature.)
Check that breakpoints are set only where they make sense.
A more accurate determination of if you are at a function-defining def or class statements (because the caller instruction contains MAKE_FUNCTION or BUILD_CLASS.)
Even without “deparsing” mentioned above, the ability to disassemble where the PC is currently located (see info pc), by line number range or byte-offset range lets you tell exactly where you are and code is getting run.
Commands that take integer arguments like up, list, or disassemble allow you to use a Python expression which may include local or global variables that evaluates to an integer. This eliminates the need in gdb for special “dollar” debugger variables. (Note however because of shlex parsing, expressions can’t have embedded blanks.)
You can now debug your program in a different process or even a different computer on a different network!
Related, is flexible support for remapping path names from file system, e.g. that inside a docker container or on a remote filesystem with locally-installed files. See subst for more information.
Can be installed via the usual pip or easy_install. There is a source tarball. How To Install has full instructions and installing from git and by other means.
The Debugger plays nice with other trace hooks. You can have several debugger objects.
Many of the things listed below doesn’t directly effect end-users, but it does eventually by way of more robust and featureful code. And keeping developers happy is a good thing.(TM)
Commands and subcommands are individual classes now, not methods in a class. This means they now have properties like the context in which they can be run, minimum abbreviation name or alias names. To add a new command you basically add a file in a directory.
I/O is it’s own layer. This simplifies interactive readline behavior from reading commands over a TCP socket.
An interface is it’s own layer. Local debugging, remote debugging, running debugger commands from a file (source) are different interfaces. This means, for example, that we are able to give better error reporting if a debugger command file has an error.
There is an experimental Python-friendly interface for front-ends
more testable. Much more unit and functional tests. More of pydb’s integration test will eventually be added.
trepan2 : trepan debugger for Python 2
pydbgr : previous incarnation of the Python 2 debugger
pydb : even older incarnation of debugger (for very old Python 2)
Tutorial: Tutorial for how to use
https://pypi.python.org/pypi/uncompyle6 : Python decompiler
https://pypi.python.org/pypi/decompyle3 : Python 3.7 and 3.8 decompiler
https://pypi.python.org/pypi/xdis : cross-platform disassembler
Release history Release notifications | RSS feed
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.