##################################################
Revision history for Log::Log4perl
##################################################
0.22 8/08/2002
* Threshold settings of appenders:
$appender->threshold($ERROR);
log4j.appender.A.Threshold = ERROR
* Chris R. Donnelly <cdonnelly@digitalmotorworks.com>
submitted two patches:
- extended init() to take obj references (added, also added a test
case and documentation)
- fixed %F and %L if Log4perl is used by a wrapper class (accepted,
but changed variable name to Log::Log4perl::caller_depth as
a tribute to Log::Dispatch::Config, added test case 022Wrap
and documentation
0.21 8/08/2002
* Synopsis shows code samples in Log4perl.pm/README
* Slight Log4j incompatibility but useful: %F{n} lets you
limit the number of entries the source file path is logged
* Erik W. Selberg (erik@selberg.com) suggested having PatternLayout.pm
suppress another \n if the messages already contains a \n and the
format requires a %n. Done.
* Erik W. Selberg (erik@selberg.com) suggested loggers should take
any number of messages and concatenate them. Done.
* Fixed double-init problem and added a test case. Now the entire
configuration is cleared before the second init(). However, this
surfaced a problem with init_and_watch: If a program obtains
references to one or more loggers, rewriting the configuration
file during program execution and re-initing makes these reference
point to loggers which hold obsolete configurations. Fixed that by
code in debug(), info(), etc. which *replaces* (shudder) the
logger reference the program hands in to them with a new one of
the same category. This happens every time if 'init_and_watch' has
been enabled. However, this introduces a small runtime penalty.
This is different from the original log4j, which does some
half-assed re-initialization, because Java isn't expressive enough
to allow for it. Making this thread-safe might be tough, though.
* Added DEBUG statements to Logger.pm and Config.pm to trace execution
(debugging won't work because of "eval"s). Both files define a
constant named DEBUG towards the top of the file, which will
have perl optimize away the debug statements in case it's set to 0.
* A warning is issued now (once) if init() hasn't been called or
no appenders have been defined.
* Added ':levels' target to Log::Log4perl to import $DEBUG, $ERROR,
etc. levels (just like 'use Log::Log4perl::Level' works).
* Added ':easy' target to allow for simple setup
* Code references can be passed in as log messages to avoid parameter
passing penalty
0.20 7/23/2002
* Strip trailing spaces in config file
* Accept line continuations in properties file
* Refactored Logger.pm for speed, defined the logging behavior when
the logger is created, not when a message is logged
* Fixing test suites so that SimpleFormat newline is accounted for
* Fixed a bug with root inheritance where the category name wasn't
coming through
* added init_and_watch
0.19 07/16/2002
* Added Log::Log4perl::TestBuffer back in the distribution, otherwise
regression test suite would fail.
0.18 07/16/2002
* Failed attempt to fix the Log::Dispatch::Buffer problem.
0.17 07/11/2002
* Updated documentation according to Dave Rolsky's suggestions
* Lots of other documentation fixes
* Fixed bug in renderer, %M was displayed as the logger function
bumped up the level by 1
* Fixed %% bug
0.16 07/10/2002
* Updated documentation for CPAN release
* Applied Kevin's patch to limit it to one Log::Dispatcher
0.15 07/10/2002
* There were name conflicts in Log::Dispatch, because we used *one*
Log::Dispatch object for the *all* loggers in the Log::Log4perl
universe (it still worked because we were using log_to() for
Log::Dispatch to send messages to specific appenders only). Now
every logger has its own Log::Dispatch object. Logger.pm doesn't
call Kevin's anti-dupe logic anymore -- is this ok? Maybe there's
some leftovers which need to be cleaned up.
* Kevin fixed t/014ConfErrs.t after last night's Appender.pm change
0.14 07/09/2002
* (!) Added new class Log::Log4perl::Appender as a wrapper around
Log::Dispatch::*. Layouts are no longer attached to the loggers,
but to the appenders instead. $app->layout($layout) sets the
layout. $logger->add_appender($app) is the new syntax to add
an appender to a logger. The $logger->layout method is gone
for that reason.
* Added documentation on categories
* Added documentation on Log::Log4perl::Appender,
Log::Log4perl::Layout::SimpleLayout,
Log::Log4perl::Layout::PatternLayout.
0.13 07/09/2002
* in the config files, 'debug' is not a level, 'DEBUG' is
* expanded the layouts so that we can add subclassess, added
SimpleLayout, note that api usage changes
-$logger->layout('buf',"The message is here: %m");
+$logger->layout(new
Log::Log4perl::Layout::PatternLayout('buf',"The message is
here: %m"));
* did benchmarks, see doc/benchmark*, t/013Bench.t
* further tweaked errors for bad configuration, added a test for those
0.12 07/08/2002
* Log::Log4perl::Logger->get_logger now accessible via
Log::Log4perl->get_logger()
* Log::Log4perl::Config->init now accessible via
Log::Log4perl->init()
* Adapted test cases to new shortcuts
* Constrained some files to 80 chars width
* Added test case t/009Deuce.t for two appenders in one category
via the config file
* Changed default layout in case there's none defined (SimpleLayout)
* Implemented dictatory date format for %d: yyyy/mm/dd hh:mm:ss
0.11 07/07/2002
* added documentation to Log/Log4perl.pm
* added is_debug/is_error/is_info etc. functions to Logger.pm,
test cases to t/002Logger.t
0.10 07/05/2002
* %p should return level name of the calling function, so
$logger->warn('bad thing!!') should print 'WARN - bad thing'
even if the category is set to debug, so took level_str out of
Logger.pm (kg)
0.09 07/03/2002
* %p should return level name, not number, adding level_str to Logger.pm (kg)
* Level.pm - discriminating: priorities are 1-4, levels are
'info','debug',etc (kg)
0.08 07/03/2002
* Non-root loggers are working now off the config file
0.07 07/02/2002
* Updated documentation
* removed "diagnostics"
0.06 07/01/2002
* Bug discovered by Kevin Goess <cpan@goess.org>, revealed
in 004-Config.t: Wrong layout used if Appender is inherited.
Fixed.
* Changed Log::Log4perl::TestBuffer to keep track of the
object population -- so we can easily reference them
in the Log::Log4perl test cases. Got rid of get_buffer().
* Added a reset() method to Log::Log4perl and Log::Log4perl::Logger
for easier testing. It resets all persistent loggers to
the inital state.
* Added documentation
0.05 06/30/2002
* Fixed bug with mapped priorities between java/Log::Dispatch
* Java/Perl integration with conf file
0.04 06/30/2002
* Layout tests
* %r to layout
* Added lib4j configuration file stuff and tests
0.03 06/30/2002
* Layout
* Curly braces in Layout first ops
0.02 06/30/2002
* Created Logger and test cases
0.01 06/22/2002
* Where it all began
TODO:
##################################################
* BasicConfigurator() vs. :easy, PropertyConfigurator()
* Log/throw exception (either warn/die or real exceptions)
* get_logger() thread safety (two try to create it at the same time)
* hierarchy-wide threshold (p.14), log4j.threshold = ERROR
* Object rendering
* NDC/MDC
* Internal DEBUG for log4perl (log4perl.debug = 1)
* custom levels (addtl. levels)
* log4j.logger.blah = INHERITED, app
* variable subst: a=b log4j.blah = ${a}
* if appenders are attached in a conf file, previous ones are detached
* log4j.additivity.loggername = true|false
* log4j.renderer.blah = blah
* I wonder if it might not be a mistake to have the base level be
'0', might be safer to make it '1', less chance of those
false-but-is-defined errors I think that's what log4j uses
* Should we canonicalize on a level is a string and a priority is an
int? might make maintenance easier
* permission problems, init() creates the files, maybe read later by
different uid, no way to set umask?
* what happens when init() is called twice?
* if $log->debug is called before init(), should it die saying
"please initialize log engine"?
* I really want a format specifier to print the pid, helps with web apps
* possible bug: was logging to a category, then took my appender off
that category, am still seeing stuff in my log, something didn't
take about the reload
* Custom filters
* XML config files
Kevin:
##################################################
* Bug: If the *same* appender is being attached to several loggers,
appender inheritance doesn't work as expected: Even if the
appender appears twice in the inheritance tree, it's only executed once.
But it actually needs to be run twice.
* Dave Rolsky says: A couple other things: The JavaMap docs mention
that SocketAppender has no Log::Dispatch equivalent. You could
create an IO::Socket object and pass it to Log::Dispatch::Handle.
* I think we should not install Log::Log4perl::TestBuffer, we only use
it for our test cases, we shouldn't infringe on the Log::Dispatch
namespace, should we?
* test cases should use platform independend directory separator
(Mac's pre-OSX ':'?) DONE need level validation on Logger::level()
DONE Create new module negotiating between Java-style config files
and perl-style appenders
* Module for reading XML-style config files
* revisit CategoryFactory idea, give the user opportunity to override
log levels with their own?
OK Please use 80 chars max line width
DONE (is gone) redo level_str() hack
DONE (Dispatch singleton) question - are appenders created just once if
they're reused in the config file?
Mike:
##################################################
* Kevin says:
>WARN in the config file
> init_and_watch(...);
> while (1){
> ...blah...
> $logger->level($INFO);
> $logger->info(...);
> sleep 5;
> }
What's happening if the reload happens right after the level() call
and the config file sets the level to WARN?
* Documentation on Layout class changes
* Document %d{} in layout
* Documentation of Log::Log4perl::Logger and Log::Log4perl::Config
* Layout.pm: 'd' (partially implemented), 't', 'x', 'X'
* Additivity flag -- set/get as API and in cfg file
* README tests (Pod::Tests or something)
* Thread safety
* Just had a wild idea: Could we possibly utilize the compiler
frontend to eliminate log statements that are not going to be
triggered? This would be a HUGE performance increase!
DONE Log::Dispatch wrapper for perl modules just like JavaMap?
DONE Warning in testcase 002
DONE Added Log::Log4perl::Appender class to wrap Log::Dispatch::*
DONE Log::Log4perl::TestBuffer: Delete from MANIFEST
DONE Log::Log4perl->init()
DONE Several appenders in one config line: log4j.rootLogger=debug, stdout, R
DONE Define default if there's no layout:
http://jakarta.apache.org/log4j/docs/api/org/apache/log4j/SimpleLayout.html
##################################################