Bringing loggers and listeners together
version 0.15
In a \s-1CPAN\s0 or other module:
package Foo;
use Log::Any qw($log);
$log->error("an error occurred");
$log->debugf("arguments are: %s", \@_)
if $log->is_debug();
my $log2 = Log::Any->get_logger(category => 'My::Class');
In your application:
use Log::Any::Adapter;
# Send all logs to Log::Log4perl Log::Any::Adapter->set('Log4perl');
# Send all logs to Log::Dispatch my $log = Log::Dispatch->new(outputs => [[ ... ]]); Log::Any::Adapter->set( 'Dispatch', dispatcher => $log );
# See Log::Any::Adapter documentation for more options
\*(C`Log::Any\*(C' allows \s-1CPAN\s0 modules to safely and efficiently log messages, while letting the application choose (or decline to choose) a logging mechanism such as \*(C`Log::Dispatch\*(C' or \*(C`Log::Log4perl\*(C'.
\*(C`Log::Any\*(C' has a very tiny footprint and no dependencies beyond Perl 5.6, which makes it appropriate for even small \s-1CPAN\s0 modules to use. It defaults to 'null' logging activity, so a module can safely log without worrying about whether the application has chosen (or will ever choose) a logging mechanism.
The application, in turn, may choose one or more logging mechanisms via Log::Any::Adapter.
\*(C`Log::Any\*(C' supports the following log levels and aliases, which is meant to be inclusive of the major logging packages:
trace debug info (inform) notice warning (warn) error (err) critical (crit, fatal) alert emergency
Levels are translated as appropriate to the underlying logging mechanism. For example, log4perl only has six levels, so we translate 'notice' to 'info' and the top three levels to 'fatal'.
Every logger has a category, generally the name of the class that asked for the logger. Some logging mechanisms, like log4perl, can direct logs to different places depending on category.
The most convenient way to get a logger in your module is:
use Log::Any qw($log);
This creates a package variable $log and assigns it to the logger for the current package. It is equivalent to
our $log = Log::Any->get_logger(category => _\|_PACKAGE_\|_);
In general, to get a logger for a specified category:
my $log = Log::Any->get_logger(category => $category)
If no category is specified, the caller package is used.
To log a message, use any of the log levels or aliases. e.g.
$log->error("this is an error"); $log->warn("this is a warning"); $log->warning("this is also a warning");
You should not include a newline in your message; that is the responsibility of the logging mechanism, which may or may not want the newline.
There are also printf-style versions of each of these methods:
$log->errorf("an error occurred: %s", $@); $log->debugf("called with %d params: %s", $param_count, \@params);
The printf-style methods have a few advantages, besides being arguably more readable:
Any complex references (like \*(C`\@params\*(C' above) are automatically converted to single-line strings with \*(C`Data::Dumper\*(C'.
Any undefined values are automatically converted to the string \*(L"<undef>\*(R".
A logging mechanism could potentially use the unchanging format string (or a digest thereof) to group related log messages together.
To detect whether a log level is on, use \*(L"is_\*(R" followed by any of the log levels or aliases. e.g.
if ($log->is_info()) { ... } $log->debug("arguments are: " . Dumper(\@_)) if $log->is_debug();
This is important for efficiency, as you can avoid the work of putting together the logging message (in the above case, stringifying @_) if the log level is not active.
Some logging mechanisms don't support detection of log levels. In these cases the detection methods will always return 1.
In contrast, the default logging mechanism - Null - will return 0 for all detection methods.
Log::Any::Test provides a mechanism to test code that uses \*(C`Log::Any\*(C'.
To direct logs somewhere - a file, the screen, etc. - you must use Log::Any::Adapter. This is intentionally kept in a separate distributions to keep \*(C`Log::Any\*(C' as simple and unchanging as possible.
Many modules have something interesting to say. Unfortunately there is no standard way for them to say it - some output to \s-1STDERR\s0, others to \*(C`warn\*(C', others to custom file logs. And there is no standard way to get a module to start talking - sometimes you must call a uniquely named method, other times set a package variable.
This being Perl, there are many logging mechanisms available on \s-1CPAN\s0. Each has their pros and cons. Unfortunately, the existence of so many mechanisms makes it difficult for a \s-1CPAN\s0 author to commit his/her users to one of them. This may be why many \s-1CPAN\s0 modules invent their own logging or choose not to log at all.
To untangle this situation, we must separate the two parts of a logging \s-1API\s0. The first, log production, includes methods to output logs (like \*(C`$log->debug\*(C') and methods to inspect whether a log level is activated (like \*(C`$log->is_debug\*(C'). This is generally all that \s-1CPAN\s0 modules care about. The second, log consumption, includes a way to configure where logging goes (a file, the screen, etc.) and the code to send it there. This choice generally belongs to the application.
\*(C`Log::Any\*(C' provides a standard log production \s-1API\s0 for modules. \*(C`Log::Any::Adapter\*(C' allows applications to choose the mechanism for log consumption.
See http://www.openswartz.com/2007/09/06/standard-logging-api/ for the original post proposing this module.
No. \*(C`Log::Any\*(C' does not, and never will, include code that knows how to log to a particular place (file, screen, etc.) It can only forward logging requests to another logging mechanism.
Each of the logging mechanisms have their pros and cons, particularly in terms of how they are configured. For example, log4perl offers a great deal of power and flexibility but uses a global and potentially heavy configuration, whereas \*(C`Log::Dispatch\*(C' is extremely configuration-light but doesn't handle categories. There is also the unnamed future logger that may have advantages over either of these two, and all the custom in-house loggers people have created and cannot (for whatever reason) stop using.
Our intent is to keep \*(C`Log::Any\*(C' minimal, and change it only when absolutely necessary. Most of the \*(L"innovation\*(R", if any, is expected to occur in \*(C`Log::Any::Adapter\*(C', which your module should not have to depend on (unless it wants to direct logs somewhere specific). \*(C`Log::Any\*(C' has no module dependencies other than Test::Simple for testing.
To encourage \s-1CPAN\s0 module authors to adopt and use \*(C`Log::Any\*(C', we aim to have as few dependencies and chances of breakage as possible. Thus, no \*(C`Moose\*(C' or other niceties.
Log::Any::Adapter, the many Log:: modules on \s-1CPAN\s0
This software is copyright (c) 2011 by Jonathan Swartz.
This is free software; you can redistribute it and/or modify it under the same terms as the Perl 5 programming language system itself.