I like DTrace and wanted to have at least part of its power for Perl progams, beyond the DTrace probes already provided by perl. So I used Aspects to create dip.pm. Allow me to quote from the manpage:

$ dip -e 'aspect Profiled => call qr/^Person::set_/' myapp.pl
$ dip -s toolkit/count_new.dip -- -S myapp.pl
$ dip -e 'before { count("constructor", ARGS(1), ustack(5)); $c{total}++ }
    call qr/URI::new$/' test.pl

$ cat quant-requests.dip
# quantize request handling time, separated by request URI
before { $ts_start = [gettimeofday] }
    call qr/Dancer::Handler::handle_request/;
after { quantize ARGS(1)->request_uri => 10**6*tv_interval($ts_start) }
    call qr/Dancer::Handler::handle_request/;
$ dip -s request-quant.dip test.pl
...
/
       value  ------------------ Distribution ------------------ count
        1024 |                                                   0
        2048 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@    95
        4096 |@@                                                 4
        8192 |                                                   0
       16384 |@                                                  1
       32768 |                                                   0

/login
       value  ------------------ Distribution ------------------ count
         512 |                                                   0
        1024 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@                70
        2048 |@@@@@@@@@@@@@@@                                    30
        4096 |                                                   0

DESCRIPTION

dip is a dynamic instrumentation framework for troubleshooting Perl programs in real time. dip can provide fine-grained information, such as a log of the arguments with which a specific function is being called.

Conceptually, dip sits on top of Aspect and uses pointcuts and advice - to use Aspect-oriented programming jargon - to define dynamic instrumentation. These instruments are applied to the program from the outside, without having to change the program code at all. While most dip scripts will consist of aspect-oriented instrumentation, they can also use the full power of Perl.

dip aims to bring some of the power of DTrace to perl. Therefore it is useful to stick to DTrace terminology. dip pointcuts resemble DTrace "probes"; dip advice resembles DTrace "actions".

Whenever the condition for a probe is met, the associated action is executed; the probe "fires". A typical probe might fire when a certain function is entered or exited. The probe's action may analyze the run-time situation by accessing the call stack and context variables and evaluating expressions; it can then print out or log some information, record it in a database, or modify variables - an action is, after all, pure Perl code. Using variables allows probes to pass information to each other, allowing them to cooperatively analyze the correlation of different events. For example, a probe that fires when a function is entered could record the current time; another probe that fires when that function is exited could record how much time the function took.

Because of the nature of Aspect-oriented programming in Perl, you only pay for what you use. When probes are defined, all existing possible locations for running the action are examined, and the probe is only activated for those locations that match the probe's condition.

Output

At the end of your program run, during END time, all aggregators - see below - will dump their results. Also any other hashes you have written to in your dip scripts will be dumped.

For example, if you simply wanted to know which kinds of objects have been instantiated at least once, you could use:

before { $c{total}++ } call qr/::new$/

and then %c will be dumped.

Aggregating functions

dip provides aggregating functions that help in understanding a set of data. You can keep counts of occurrences, or quantize data, much like with DTrace.

The quantize aggregating function generates a power-of-two distribution - see its documentation.

FUNCTIONS

import

Remembers the dip script given on the command-line so we can run it in instrument(). Complains if there was no dip script. The --delay option is passed in this way as well.

instrument

Evaluates the dip script we remembered in import(). Dies if there was a problem evaluating it.

Normally this function will be called automatically during INIT time, but you can delay by giving the --delay option to dip; you would use this if your program loads other code at runtime - using do(), for example - that needs to be instrumented as well. In that case you have manually activate the instrumentation using:

$dip::dip && $dip::dip->();

run

Convenience function that takes a filename and runs the file via do(). This is what dip -s uses. For example:

dip -s myscript.dip myapp.pl

is turned into:

dip -e 'run q!$file!' myapp.pl

and ultimately

perl -Mdip='run q!$file!' myapp.pl

ustack

Returns a concise stack trace. Takes an argument of how many levels deep the stack trace should be; the default is 20 levels. Stack frames that point to a package name in the Aspect:: or dip namespace are omitted.

Example: count how many times a XML::LibXML::NodeList object is created, and keep a separate counter for each place it is created from, remembering three stack frames for each place:

before { count "constructor", ARGS(0), ustack(3) }
    call qr/XML::LibXML::NodeList::new$/

cluck

Returns what Carp's cluck() would return, again with Aspect:: and dip namespaces omitted.

count

This aggregator function takes a counter name and a value and keeps a count of how often this value was seen for this counter.

You can pass several values; they will be concatenated using newlines. See the example for ustack().

Example: For each class, count how many objects are created. Also keep a total count.

before { count("constructor", ARGS(0)); $c{total}++ }
    call qr/::new$/

dump_var

Convenience method to dump a variable like Data::Dumper does.

Example: Show all requests a Dancer web application handles:

before { dump_var ARGS(1) }
    call qr/Dancer::Handler::handle_request/

rtrim

Convenience function to right-trim a string.

rref

Convenience function that, if given a string - for example, a package name -, just returns the string, but if given an object, it returns that object's class.

Useful if objects you want to instrument are sometimes created by calling new() on existing objects:

before { count("constructor", rref ARGS(0)) } call qr/::new$/

ARGS

Convenience function to access the arguments of a function that you are instrumenting. ARGS(0), for example, returns the first argument. You can use several argument indices; in this case the indicated function arguments will be stringified and concatenated with a space.

ARGS(0) is equivalent to $_->{args}[0]; ARGS(1,2) is equivalent to join ' ' => ARGS(0), ARGS(1) - see Aspect for the kind of context information that is passed to advice code.

quantize

This aggregator function takes a name, or an reference to a list of names, and a value. For each name, it keeps track of a power-of-two frequency distribution of the values of the specified expressions. Increments the value in the highest power-of-two bucket that is less than the specified expression.