NAME

Devel::Trepan -- A new modular Perl debugger

SUMMARY

A modular, testable, gdb-like debugger in the family of the Ruby trepanning debuggers.

Features:

  • extensive online-help

  • syntax highlighting of Perl code

  • context-sensitive command completion

  • out-of-process and remote debugging

  • interactive shell support

  • code disassembly

  • gdb syntax

  • easy extensibility at several levels: aliases, commands, and plugins

  • comes with extensive tests

  • is not as ugly as perl5db

Some of the features above require additional modules before they take effect. See "Plugins" and "Recommended Modules" below.

DESCRIPTION

Invocation

From a shell:

$ trepan.pl [trepan-opts] -- perl-program [perl-program-opts]

For out-of-process (and possibly out-of server) debugging:

$ trepan.pl --server [trepan-opts] -- perl-program [perl-program-opts]

and then from another process or computer:

$ trepan.pl --client [--host DNS-NAME-OR-IP]

Calling the debugger from inside your Perl program using Joshua ben Jore's Enbugger:

# This needs to be done once and could even be in some sort of 
# conditional code
require Enbugger; Enbugger->load_debugger( 'trepan' );

# Alternatively, to unconditionally load Enbugger and trepan:
use Enbugger 'trepan';

# work, work, work...
# Oops! there was an error! Enable the debugger now!
Enbugger->stop;  # or Enbugger->stop if ... 

Or if you just want POSIX-shell-like set -x line tracing:

$ trepan.pl -x -- perl-program [perl-program-opts]

Inside the debugger tracing is turned on using the command set trace print. There is extensive help from the help command.

Basic Commands

The help system follows the gdb classificiation. Below is not a full list of commands, nor does it contain the full list of options on each command, but rather some of the more basic commands and options.

Commands involving running the program

Step into (step)

step[<+|-] [into] [count]

Execute the current line, stopping at the next event. Sometimes this is called "step into".

With an integer argument, step that many times. With an 'until' expression that expression is evaluated and we stop the first time it is true.

A suffix of + in a command or an alias forces a move to another position, while a suffix of - disables this requirement. A suffix of > will continue until the next call. (finish will run run until the return for that call.)

If no suffix is given, the debugger setting different determines this behavior.

Examples:

step        # step 1 event, *any* event obeying 'set different' setting
step 1      # same as above
step+       # same but force stopping on a new line
step over   # same as 'next'
step out    # same as 'finish'

Related and similar is the next (step over) and finish (step out) commands. All of these are slower than running to a breakpoint.

Step over (next)

next

Step one statement ignoring steps into function calls at this level. Sometimes this is called "step over".

Continue execution (continue)

continue [location]

Leave the debugger loop and continue execution. Subsequent entry to the debugger however may occur via breakpoints or explicit calls, or exceptions.

If a parameter is given, a temporary breakpoint is set at that position before continuing.

Examples:

continue
continue 10    # continue to line 10
continue gcd   # continue to first instruction of method gcd

See also step, next, finish commands and help location.

Step out (finish)

finish

Continue execution until the program is about to leave the current function. Sometimes this is called 'step out'.

Terminate gently (quit)

quit[!] [unconditionally] [exit-code]

Gently exit the debugger and debugged program.

The program being debugged is exited via exit() which runs the Kernel at_exit finalizers. If a return code is given, that is the return code passed to exit() - presumably the return code that will be passed back to the OS. If no exit code is given, 0 is used.

Examples:

quit                 # quit prompting if we are interactive
quit unconditionally # quit without prompting
quit!                # same as above
quit 0               # same as "quit"
quit! 1              # unconditional quit setting exit code 1

See also kill and set confirm.

Hard termination (kill)

kill[!] [signal-number|signal-name]

Kill execution of program being debugged.

Equivalent of kill('KILL', $$). This is an unmaskable signal. Use this when all else fails, e.g. in thread code, use this.

If you are in interactive mode, you are prompted to confirm killing. However when this command is aliased from a command ending in !, no questions are asked.

Examples:

kill  
kill KILL # same as above
kill -9   # same as above
kill  9   # same as above
kill! 9   # same as above, but no questions asked
kill unconditionally # same as above
kill TERM # Send "TERM" signal

See also quit

Restart execution (restart)

restart

Restart debugger and program via an exec call.

See also show args for the exact invocation that will be used.

Examining data

Evaluate Perl code (eval)

eval[@$][?] [Perl-code]

Run Perl-code in the context of the current frame.

If no string is given after the word "eval", we run the string from the current source code about to be run. If the "eval" command ends ? (via an alias) and no string is given we try to pick out a useful expression in the line.

Normally eval assumes you are typing a statement, not an expression; the result is a scalar value. However you can force the type of the result by adding the appropriate sigil @, or $.

Examples:

eval 1+2 # 3
eval$ 3   # Same as above, but the return type is explicit
$ 3       # Probably same as above if $ alias is around
eval $^X  # Possibly /usr/bin/perl
eval      # Run current source-code line
eval?     # but strips off leading 'if', 'while', ..
          # from command 
eval @ARGV  # Make sure the result saved is an array rather than 
            # an array converted to a scalar.
@ @ARG       # Same as above if @ alias is around
use English  # Note this is a statement, not an expression
use English; # Same as above
eval$ use English # Error because this is not a valid expression 

See also set auto eval to treat unrecognized debugger commands as Perl code.

Recursively Debug into Perl code (debug)

debug Perl-code

Recursively debug Perl-code.

The level of recursive debugging is shown in the prompt. For example ((trepan.pl)) indicates one nested level of debugging.

Examples:

debug finonacci(5)   # Debug fibonacci function
debug $x=1; $y=2;    # Kind of pointless, but doable.

Making the program stop at certain points

Set a breakpont (break)

break [location] [if condition]

Set a breakpoint. If location is given use the current stopping point. An optional condition may be given.

Examples:

break                  # set a breakpoint on the current line
break gcd              # set a breakpoint in function gcd
break gcd if $a == 1   # set a breakpoint in function gcd with 
                       # condition $a == 1
break 10               # set breakpoint on line 10

When a breakpoint is hit the event icon is xx.

See also help breakpoints.

Set a temporary breakpoint (tbreak)

tbreak [location]

Set a one-time breakpoint. The breakpoint is removed after it is hit. If no location is given use the current stopping point.

Examples:

tbreak
tbreak 10               # set breakpoint on line 10

When a breakpoint is hit the event icon is x1.

See also break and help breakpoints.

Add or modify a condition on a breakpoint (condition)

condition bp-number Perl-expression

bp-number is a breakpoint number. perl-expresion is a Perl expression which must evaluate to true before the breakpoint is honored. If perl-expression is absent, any existing condition is removed; i.e., the breakpoint is made unconditional.

Examples:

condition 5 x > 10  # Breakpoint 5 now has condition x > 10
condition 5         # Remove above condition

See also "break", "enable" and "disable".

Delete some breakpoints (delete)

delete [bp-number [bp-number...]]

Delete some breakpoints.

Arguments are breakpoint numbers with spaces in between. To delete all breakpoints, give no arguments.

See also the clear command which clears breakpoints by line number and info break to get a list of breakpoint numbers.

Enable some breakpoints (enable)

enable num [num ...]

Enables breakpoints, watch expressions or actions given as a space separated list of numbers which may be prefaces with an 'a', 'b', or 'w'. The prefaces are interpreted as follows:

a -- action number
b -- breakpoint number
w -- watch expression number

If num is starts with a digit, num is taken to be a breakpoint number.

Disable some breakpoints (disable)

disable bp-number [bp-number ...]

Disables the breakpoints given as a space separated list of breakpoint numbers. See also info break to get a list of breakpoints

Set an action before a line is executed (action)

action position Perl-statement

Set an action to be done before the line is executed. If line is ., set an action on the line about to be executed. The sequence of steps taken by the debugger is:

1. check for a breakpoint at this line
2. print the line if necessary (tracing)
3. do any actions associated with that line
4. prompt user if at a breakpoint or in single-step
5. evaluate line

For example, this will print out the value of $foo every time line 53 is passed:

Stop when an expression changes value (watch)

watch Perl-expression

Stop very time Perl-expression changes from its prior value.

Examples:

watch $a  # enter debugger when the value of $a changes
watch scalar(@ARGV))  # enter debugger if size of @ARGV changes.

Examining the call stack

backtrace [count]

Print a stack trace, with the most recent frame at the top. With a positive number, print at most many entries.

In the listing produced, an arrow indicates the 'current frame'. The current frame determines the context used for many debugger commands such as source-line listing or the edit command.

Examples:

backtrace    # Print a full stack trace
backtrace 2  # Print only the top two entries

Select a call frame (frame)

frame [frame-number]

Change the current frame to frame frame-number if specified, or the most-recent frame, 0, if no frame number specified.

A negative number indicates the position from the other or least-recently-entered end. So frame -1 moves to the oldest frame.

Examples:

frame     # Set current frame at the current stopping point
frame 0   # Same as above
frame .   # Same as above. 'current thread' is explicit.
frame . 0 # Same as above.
frame 1   # Move to frame 1. Same as: frame 0; up
frame -1  # The least-recent frame

Move to a more recent frame (up)

up [count]

Move the current frame up in the stack trace (to an older frame). 0 is the most recent frame. If no count is given, move up 1.

Move to a less recent frame (down)

down [count]

Move the current frame down in the stack trace (to a newer frame). 0 is the most recent frame. If no count is given, move down 1.

Syntax of debugger commands

Overall Debugger Command Syntax

If the first non-blank character of a line starts with #, the command is ignored.

Commands are split at whereever ;; appears. This process disregards any quotes or other symbols that have meaning in Perl. The strings after the leading command string are put back on a command queue.

Within a single command, tokens are then white-space split. Again, this process disregards quotes or symbols that have meaning in Perl. Some commands like eval, macro, and break have access to the untokenized string entered and make use of that rather than the tokenized list.

Resolving a command name involves possibly 4 steps. Some steps may be omitted depending on early success or some debugger settings:

1. The leading token is first looked up in the macro table. If it is in the table, the expansion is replaces the current command and possibly other commands pushed onto a command queue. See the "help macros" for help on how to define macros, and "info macro" for current macro definitions.

2. The leading token is next looked up in the debugger alias table and the name may be substituted there. See "help alias" for how to define aliases, and "show alias" for the current list of aliases.

3. After the above, The leading token is looked up a table of debugger commands. If an exact match is found, the command name and arguments are dispatched to that command. Otherwise, we may check to see the the token is a unique prefix of a valid command. For example, "dis" is not a unique prefix because there are both "display" and "disable" commands, but "disp" is a unique prefix. You can allow or disallow abbreviations for commands using "set abbrev". The default is abbreviations are on.

4. If after all of the above, we still don't find a command, the line may be evaluated as a Perl statement in the current context of the program at the point it is stoppped. However this is done only if "auto eval" is on. (It is on by default.)

If "auto eval" is not set on, or if running the Perl statement produces an error, we display an error message that the entered string is "undefined".

Debugger Command Examples

Commenting

# This line does nothing. It is a comment and is useful
# in debugger command files.
     # any amount of leading space is also ok

Splitting Commands

The following runs two commands: info program and list

info program;; list 

The following gives a syntax error since ;; splits the line and the simple debugger parse then thinks that the quote (") is not closed.

print "hi ;;-)\n" 

If you have the Devel::Trepan::Shell plugin, you can go into a real shell and run the above.

Command Continuation

If you want to continue a command on the next line use \ at the end of the line. For example:

eval $x = "This is \
a multi-line string"

The string in variable $x will have a \n before the article "a".

Command suffixes which have special meaning

Some commands like step, or list do different things when an alias to the command ends in a particular suffix like ">".

Here are a list of commands and the special suffixes:

command   suffix
-------   ------
list       >
step       +,-,<,>
next       +,-,<,> 
quit       !
kill       !
eval       ?

See help on the commands listed above for the specific meaning of the suffix.

BUGS/CAVEATS

Because this should be useful in all sorts of environments such as back to perl 5.008, we often can make use of newer Perlisms nor can we require by default all of the modules, say for data printing, stack inspection, or interactive terminal handling. That said, if you have a newer Perl or the recommended modules or install plugins, you'll get more.

Although modular, this program is even larger than perl5db and so it loads a little slower. I think part of the slowness is the fact that there are over 70 or so (smallish) files (rather than one nearly 10K file) and because relative linking via rlib is used to glue them together.

AUTHOR

Rocky Bernstein

SEE ALSO

My Devel::Trepan blogs and wiki

Plugins

Other Debuggers

  • perldebug is perl's built-in tried-and-true debugger that other debuggers will ultimately be compared with

  • Devel::ebug

  • DB is a somewhat abandoned debugger API interface. I've tried to use some parts of this along with perl5db.

COPYRIGHT

Copyright (C) 2011, 2012 Rocky Bernstein <rocky@cpan.org>

This program is distributed WITHOUT ANY WARRANTY, including but not limited to the implied warranties of merchantability or fitness for a particular purpose.

The program is free software. You may distribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation (either version 2 or any later version) and the Perl Artistic License as published by O'Reilly Media, Inc. Please open the files named gpl-2.0.txt and Artistic for a copy of these licenses.