Finding Stubborn Bugs with Meaningful Debug Info
Bug tracking is often one of the most difficult processes in software development. Users may have situations different from developers, and bugs that are a big problem for users may not even be visible on developers' machines. Sometimes bugs can come and go, or networked programs may encounter bugs only when talking to specific servers or clients. In this article, I discuss techniques software developers can employ to help track down bugs more easily.
First, I discuss two ways to make it easier to receive and manage bugs, and then I show how to make your programs generate more useful debugging output. Then, I talk about tracking down troublesome bugs. Finally, I cover some practices that can help prevent bugs in the first place. Many of the techniques described in this article are employed in OfflineIMAP (see “Fast Convenient Mail for Travel: OfflineIMAP”, LJ, March 2004).
Before examining how to make better bug reports possible, a critical first step is making sure you can deal with the bug reports you receive. For some small projects, simply publishing an e-mail address is sufficient. However, most projects need something more. Developers often get busy and forget about things. Bugs may be complicated to solve, requiring input from several people, or there simply may be a lot of bug reports.
A bug-tracking system (BTS) is a great way to help ensure that bugs are not forgotten. Most BTS implementations provide a way to track correspondence, handle attached files and delegate responsibility to particular people. Some also support categorization based on things such as severity, user environment and specific components.
If your project is hosted at a project hosting site such as SourceForge or Savannah, you already have a BTS available for your use. You should use it and encourage your users to submit bugs through that interface rather than to a mailing list.
If you need more flexibility, you can find BTS programs for Linux. Some of the most popular free software BTS programs are:
Bugzilla, the BTS used by the Mozilla Project, is a flexible system primarily used through its Web interface.
Request Tracker can be used as both a bug-tracking system and a support-tracking system. It features both Web and e-mail interfaces, though some administrative functions can occur only through the Web interface.
Jitterbug is the BTS used by the Samba Project. It is similar in concept to Bugzilla but is more lightweight.
Debbugs is the BTS used by the Debian Project. Debbugs has a Web interface, though it is read-only; all manipulations occur by e-mail. Debbugs is best suited for large projects with clearly identifiable components and responsibility for those components.
I personally prefer Request Tracker, because it seems to have a nice blend of features for a BTS. Your own requirements may differ.
Sometimes I find a nasty bug in a program and want to report it. But to do so, I have to fill out a detailed questionnaire and perhaps divulge information I'd rather not. It should be easy for people to submit bugs and the information needed to track them down. If you take submissions on the Web, make the process simple. Don't require too much information, and accept submissions even if people don't know some information. Don't expect users to know anything about the different components of the project or which developers are responsible for a given problem.
When tracking down problems, you often want to know what state the program is in. Other times, you may want to know what actions were carried out prior to triggering the bug. Because users of programs don't necessarily have expertise with your code and a debugger, logging often is called for. Logging simply means writing out a record of the actions carried out. Simple programs might merely print out information, but usually you'll want something a little more capable.
Non-interactive programs, such as network servers, do not have a screen on which to display information. These programs often maintain a log file or use the syslog facility built in to Linux and UNIX systems.
Interactive programs may display information on-screen or also may generate a file. Having a log file available can make bug reporting easier, because the user simply can attach it to a bug report.
Sometimes, you might need quite a bit of data to figure out what's going on with a specific problem. However, all this data may be overkill for a normal session—it could flow right off a user's screen or fill up a hard disk. Therefore, many programs have a notion of a log level. The user can set, at runtime, how much information should be logged. Some programs even may have log categories, where users can configure which types of information are logged. OfflineIMAP uses this approach. For troublesome problems, users can turn on a communications log, which logs all data sent to or received from the IMAP server.
Python 2.3 introduced a useful module called logging. The logging module provides a uniform interface to several different ways of logging messages. Its supported logging methods include writing messages to files, network services, syslog, e-mailing messages and several others. The following is a simple example that illustrates use of the logging module:
#!/usr/bin/env python import logging, sys # Create the logger object l = logging.getLogger('testlog') # Create a handler and assign it to the object handler = logging.StreamHandler(sys.stderr) l.addHandler(handler) # Levels are DEBUG, INFO, WARNING, ERROR, CRITICAL. # Set the default level here. Any log messages # beneath that level are dropped. l.setLevel(logging.INFO) # Try it out. l.debug("Debug message -- system initialized.") l.info("Here's some info. I've just debugged.") l.warning("I don't have many messages left.") l.error("Only one more message to go.") l.critical("Nothing else to do!")
This program begins by initializing the logger. It uses the StreamHandler to write logged text to standard error. It also sets the log level to INFO. Then it logs five messages. When you run this program, you see only the last four. The debug message was filtered out by setting the level to INFO. Many programs have a configuration or command-line option to set the level at runtime. You can use different logging methods simply by adding a different handler to your Logger object. The Python documentation has a reference for all the available handlers.
Practical books for the most technical people on the planet. Newly available books include:
- Agile Product Development by Ted Schmidt
- Improve Business Processes with an Enterprise Job Scheduler by Mike Diehl
- Finding Your Way: Mapping Your Network to Improve Manageability by Bill Childers
- DIY Commerce Site by Reven Lerner
Plus many more.
- Non-Linux FOSS: Snk
- diff -u: What's New in Kernel Development
- Building a Multisourced Infrastructure Using OpenVPN
- Server Hardening
- 22 Years of Linux Journal on One DVD - Now Available
- Giving Silos Their Due
- Controversy at the Linux Foundation
- Don't Burn Your Android Yet
- What's New in 3D Printing, Part III: the Software