Awesome
<a id="sec-1" name="sec-1"></a>Introduction
NOTE - Because of some mishap with QuickLisp, June 2013 QuickLisp distribution pulled the master branch of Log4CL instead of stable branch as intended.
Very few incompatibilities been reported, and it makes no sense to downgrade version that been already available in QuickLisp, therefore QuickLisp will continue to pull the master branch from now on.
The major difference of Log4CL 1.1.x version from 1.0.x is the Emacs/Slime integration module, called Log4Slime. This document is written with assumption that you will have Log4Slime loaded and enabled in Emacs.
<a id="sec-1-1" name="sec-1-1"></a>Installation
Log4CL is available from QuickLisp. To load it use (ql:quickload :log4cl)
command from REPL. Log4Slime is available in QuickLisp since June 2013.
<a id="sec-1-2" name="sec-1-2"></a>Enabling Log4Slime
(ql:quickload :log4slime)
(log4slime:install)
You should get a message like this:
Wrote ~/quicklisp/log4slime-setup.el
Add the following two statements to your ~/.emacs file
------------------------------------------------------
(load "~/quicklisp/log4slime-setup.el")
(global-log4slime-mode 1)
------------------------------------------------------
Follow the above instructions. The most likely point of failure here
may be log4slime.el
trying to poke the Lisp side to see if log4slime
is loaded, and that for some reason it fails.
If it fails, there should be a message in *Messages*
buffer, that looks
like this: Unable to load log4slime lisp support: <lisp side condition>
In case you did get the above message and you think you fixed the
cause, you can make log4slime try again, by turning M-x global-log4slime-mode
off/on a few times
You can verify if log4slime is enabled, by looking for the Log4CL menu on top menubar in REPL and Lisp mode buffers, or by submitting a log statement from REPL and seeing output colorized.
<a id="The-Basics" name="The-Basics"></a>
<a id="sec-2" name="sec-2"></a>Hello world
(progn
(log:info "I just ate a ~5f, feeling tired" pi)
(when (log:debug)
(dotimes (sheep 3)
(log:debug sheep "zzz")))
(log:warn "doh fell asleep for" (random 10) "minutes"))
Should produce the following output
You can notice several things right away.
-
You can intermix strings and expressions, and expressions will be printed as their un-evaluated form as well as the their value; using a constant
FORMAT
control string as first argument automatically switches toFORMAT
mode. -
Log statements without arguments, turn into conditional expression that tells you if logging is enabled or not
-
The log output contains some headers in addition to the raw log message, such as the log level that message was submitted with, the time of the message as well as what looks like the name of the current package, and (If you have Log4Slime loaded and turned on), these headers use different faces from actual log message.
<a id="sec-2-1" name="sec-2-1"></a>Changing the log level
You can change the log level with by doing
(log:config :debug)
from REPL
. Alternatively, with
Log4Slime enabled in REPL
buffer, you can right-click on the
cl-user part and change the log level from a popup menu.
There is a slight difference between doing it using above two
methods, with log:config
command, the level is changed for the
ROOT
category, but right clicking on the package
name, changes the log level only for that package.
To change the ROOT
category from Emacs, you can use Log4CL
dropdown submenu, or Emacs command log4slime-level-selection
which
is bound to C-c C-g
by default.
After pressing C-c C-g
which invokes the log4slime-level-selection
command
you get the following window.
Pressing "p" to select the package category shows effective log level and allows you to change like so
This concludes the very basic introduction, if you were confused by what various terms such as "category" mean, click on the hyperlink to read more about Log4CL concepts.
Or you can skip the theory and just continue to learn by example.
<a id="naming" name="naming"></a>
<a id="sec-3" name="sec-3"></a>Automatic category naming
Try putting the from the previous section into a DEFUN
instead of a
PROGN
form like so:
(defun hello ()
(log:info "I just ate a ~5f, feeling tired" pi)
(when (log:debug)
(dotimes (sheep 3)
(log:debug sheep "zzz")))
(log:warn "doh fell asleep for" (random 10) "minutes"))
If you run it now, the output under both SBCL and CCL should look like this.
Right click on the blue function name, allows you to change the log level for that specific function.
That is because Log4CL logging macros, automatically determine the
category for logging, based on the context where log statement
appears. In above example the function was defined in the package
CL-USER and function name was HELLO, so the target category of any
logging inside the function, was automatically CL-USER.HELLO
It starts with the package, then function. You can try
putting one of the log statements inside of a LABELS
or FLET
forms, to see what happens.
Also note the farthest to the right in the logging category name, the more specific. The level for "hello" overrides that for "cl-user", which in turn overrides that of the root category.
<a id="sec-3-1" name="sec-3-1"></a>Naming in source files
For the next few examples, it is recommended that you load the
examples come together with Log4CL, by doing (ql:quickload :log4cl-examples)
It should produce the following output:
One thing you should notice, is that source file where function is
defined now appears as part of the log message too. Go to the source
of "greetings". Before you try to use Slime's famous M-.
shortcut, try
clicking on blue "greetings" word with a left mouse button.
If everything went better then expected, it should land you at the first
log statement of the (defun greetings ())
. Cool eh?
<a id="sec-3-2" name="sec-3-2"></a>Naming in CLOS methods
Quickly browse through naming-examples.lisp
. There are a few methods defined,
including :after/:around
methods, as well as some with EQL
specializers.
Run a few of them from REPL, like so:
Log statements inside of methods, are using the category name of the generic function, extended with qualifier, and all non-T specializers.
Try going to the source of the above methods by clicking on them. It should land in the right method, without showing Slime's XREF window.
Note how by changing the level of the foobar
you control all the
methods, but can override them based on their specializers. Try
setting :after
category to different levels, to control all
the :after
methods together.
In addition to playing with methods, try (setf (test.package.one:greetings) "Hey")
too.
<a id="sec-3-3" name="sec-3-3"></a>Context sensitivity
As you browse through source, and are inside of one of the methods, check out the Log4CL dropdown menu. Note that "Defun" submenu changes for each method.
<a id="sec-3-4" name="sec-3-4"></a>Keyboard level selection
Also try C-c C-g
shortcut in the same place. You can configure it
not to show the selection window at all, by customizing the
log4slime-level-selection-single-key
Emacs variable.
After pressing C-c C-g
while inside of the method
You can change keys for the selecting various levels by doing M-x customize-group RET log4slime RET
Note that keyboard selection ignores the Control key so C-c C-g p u
is same as C-c C-g C-p C-u
<a id="sec-3-5" name="sec-3-5"></a>Resetting the mess
If you had forgotten which levels you set for what, and just want to see which levels are set where.
You can display current logging configuration by doing
(log:config
) without any arguments, it willdisplay a tree
If you have had set a lot of custom levels, and now need to get rid
of them, "Reset Children" menu item will nukes the log level from
everything underneath the parent. Doing "Reset Children" on the
ROOT category, gets rid of every other log level that was set
anywhere. Keyboard equivalent is C-c C-g r
<a id="sec-3-6" name="sec-3-6"></a>Logging configurations
After setting the log levels of a few methods, try doing (log:save :foo)
then messing around.. You can restore the named configuration with
(log:restore :foo)
. Configurations are saved in a file in the
home directory, so they survive image restarts
See the Finding needle in a haystack section.
<a id="sec-4" name="sec-4"></a>The magic of (LOG:CONFIG)
Section To be written, for now simply see docstring for LOG:CONFIG
Read the docstring and play with options, below are a few examples:
<a id="sec-5" name="sec-5"></a>Pattern Layout
Section to be written, for now see docstring for docstring for PATTERN-LAYOUT
<a id="sec-6" name="sec-6"></a>Common Practices
Some common recipes.
<a id="sec-6-1" name="sec-6-1"></a>Log levels for production
Generally log levels INFO
and below, are used in normal
operations of software, while levels higher then INFO
are used
by programmers.
-
FATAL
is used for un-recoverable errors, that require restart of an application or major component, theFATAL
messages are to inform the user that something had died in a way that should not normally happen. -
ERROR
is for serious but generally recoverable errors, that occur doing a normal operation of software. File not found, or such. -
WARN
is for "suspicious" things, or to inform the user that some automatic corrective action had failed. Maximum number of retries reached or such. -
INFO
is for informing on major steps that software is performing, and is usually thought of the maximum log level used in normal operations, its "Say what you are doing but don't flood" type of messages.
By default Log4CL is configured with root category having INFO
log level.
<a id="development" name="development"></a>
<a id="sec-6-2" name="sec-6-2"></a>Log levels for development
DEBUG
is for for informing about detailed steps taken by operations
and printing intermediate values.
TRACE
is for very detailed debugging, like printing variables inside
loops and such.
DEBU1..DEBU9
log levels are numerically around the TRACE
and can be used
if you need more granularity. One possibility is that (log:expr)
macro, can
be configured via LOG:PACKAGE-OPTIONS
mechanism, to use different
log level then DEBUG
and can set to use one of the extra levels.
OFF
log level is very important counter-part for DEBUG
and
TRACE
. Its used for "narrowing things down in reverse", which is
described in the next section
<a id="needle" name="needle"></a>
<a id="sec-6-3" name="sec-6-3"></a>Finding needle in a haystack
Programmers often need to concentrate on a specific area of their software. With traditional non-hierarchical logging system, having a lot of debug sprinkled around the code, flood the programmers with a lot of information they don't need, and makes it hard to find the messages relevant to the problem being debugged.
Because Log4CL is hierarchical, its easy to narrow down the logging, to focus on exactly the right area, by using the following process.
-
Turn
DEBUG
on for the root category, or entire package and then run your code through the functionality that you are focusing on. REPL will fill with a lot of debugging output. -
Right-click on each message that is not related to a problem, and turn the corresponding category
OFF
. You can how go wide or narrow, turn off entire packages or source files, or by individual methods, functions or local functions. If you went too far, use Reset children command on the parent category.If you use CLOS, use the category hierarchy to your advantage, if for example you think problem relates to before or after method, you can can control logging for all :AFTER methods of generic function by clicking :after category in
(<gf name> :after <specializer> ...)
-
Once you narrowed down the logging to your liking, you can quickly save that configuration of log levels with
(LOG:SAVE)
, and later (may be in a different image, or even different machine) restore it with(LOG:RESTORE)
, and you can give these saved configuration names, such as(LOG:SAVE :bug-123)
<a id="sec-7" name="sec-7"></a>Glossary
Very small glossary of Log4CL concepts
<a id="logger" name="logger"></a><a id="category" name="category"></a>
<a id="sec-7-1" name="sec-7-1"></a>Loggers and categories
Loggers are named singleton objects that form a hierarchy, and are
sources of log messages, or more correctly entry points where log
message enter the logging system. Each call to a logging macro like
(log:debug ...)
operates on a specific logger object
(See also naming section).
Logger's unique name is called "logger's category", or "category name". Loggers form a hierarchy, based on their category names, where child loggers have their category name prefixed by that of the parent, followed by a dot. So if we have loggers A, A.B, A.B.C and A.B.D then logger A is parent of A.B, which has two children A.B.C and A.B.D - as shown on below diagram. (Note: ROOT logger category name is empty string)
ROOT---A---A.B---A.B.C
|
\---A.B.D
Because loggers are singletons, logger category name is usually shortened to just CATEGORY and is used inter-changeably with the word LOGGER; the convention is that thing is "a logger" when talking about actual Lisp object, and "category" otherwise.
Each logger can have a log level threshold, or if its does not have one, it inherits one from its parent. To ensure that for any logger, an effective log level can be determined, the ROOT logger always have a level.
Loggers will only pass through messages, if logger's threshold level
is equal or greater verbosity, then log message. For example if in
above example logger A is configured with info log level, then
(log:warn ...)
and (log:info)
messages will be passed through,
but (log:debug)
messages would not.
<a id="appender" name="appender"></a>
<a id="sec-7-2" name="sec-7-2"></a>Appenders
Appenders process log messages by writing them to files, or displaying them on the screen. Appenders attach to a specific logger, and each logger can have many appenders attached.
When a log message passes through a logger that has appenders, they are all called in turn to do appender specific processing, be it writing log message to a file, or a terminal. After all of logger's appenders had processed the message, its passed on to the parent logger.
So log messages inheritance flows in reverse order from the log level one, tricking up from child loggers towards root, with below exception.
Each logger has a property called additivity, which is T
by
default, which controls the above process. When additivity is
NIL
, logger is called non-additive and any messages that reach
it, will not be passed to the parents.
Usually only root logger, or non non-additive loggers will have any appenders attached to them.
<a id="layout" name="layout"></a>
<a id="sec-7-3" name="sec-7-3"></a>Layouts
When appender decide they want to process the log message, they format the log message by means of a layout. Layout is a separate object, that attaches to each appender, and is responsible for the textual formatting of the message.
So while appender provides and manages any serialization for the stream to write to, the layout is actually formatting the log message into that stream.
Log4CL provides two layouts, SIMPLE-LAYOUT which is well, simple, and a very configurable PATTERN-LAYOUT, which specifies the formatting of log messages by mean of printf/format like control string.
Easiest way to use the pattern layout, is by using LOG:CONFIG command to select between several predefined formats.
Or you can look for list of all supported format documentation for the PATTERN-LAYOUT class. Please note that if you are drafting your own format, that Log4SLime fontification relies on regular expressions and the log messages being in a certain order. If your layout is not a minor modification of an built-in one, the Log4Slime fontification may stop working. You can of course adjust the regular expressions used by Log4Slime to match your own custom layout to compensate.
<a id="level" name="level"></a><a id="log-level" name="log-level"></a>
<a id="sec-7-4" name="sec-7-4"></a>Log Levels
In Log4CL log levels are numeric constants, in order of increased verbosity:
-
Turn off logging
0=OFF
-
Standard log levels
1=FATAL
,2=ERROR
,3=WARN
,4=INFO
,5=DEBUG
-
Extra debug levels
6..9
namedDEBU1
throughDEBU4
-
Standard log level
10=TRACE
-
Extra debug level
11..15
namedDEBU5
throughDEBU9
<a id="effective" name="effective"></a>
<a id="sec-7-5" name="sec-7-5"></a>Effective log level
Effective log level of the logger X is determined as follows.
-
If logger has level threshold set, then this level is the effective log level.
-
If logger is not first child of a parent, whose category is same as the package name logger was instantiated from, the effective log level of X is the effective log level of its parent logger.
-
If logger is first child of a parent P named same as package, and there exists a sibling logger S, with the last part of category name equal to that of a source file logger X was instantiated from, and S has a level threshold set, that level is effective level of logger X
-
Otherwise effective level of logger X is effective level of its parent.
ROOT logger always has a level threshold set, so above steps always result in a valid log level.
Effective log level is returned by the function (log4cl:effective-log-level LOGGER)
<a id="sec-8" name="sec-8"></a>FAQ
<a id="sec-8-1" name="sec-8-1"></a>I don't see log messages from other threads.
The *TERMINAL-IO*
value bound in the other threads is probably different and points
to other place (likely *inferior-lisp*
buffer under Slime)
-
(log:config :sane2)
will copy messages from other threads toREPL
while continuing output to thread specific*TERMINAL-IO*
(REPL
thread will still only log toREPL
) -
(log:config :sane :this-console)
will redirect all logging to current console regardless of thread local values of*TERMINAL-IO*
<a id="sec-8-2" name="sec-8-2"></a>Why Log4CL starts its own thread, and how I get rid of it
Its a flusher thread to flush the appenders, it increases performance greatly when there is a lot of logging.oe
You can stop it by calling (log4cl:stop-hierarchy-watcher-thread)
On SBCL Log4CL uses *EXIT-HOOKS*
and *SAVE-HOOKS*
to
automatically flush all appenders on exit, so that last second of
logging is not lost, and to terminate the watcher thread when
saving image, which can't be done with multiple threads running.
<a id="sec-8-3" name="sec-8-3"></a>I'd like just the log messages, and not all the extra stuff
Use pattern layout with just %m%n format (message + newline)
<a id="sec-8-4" name="sec-8-4"></a>How do I log into a file
(log:config :daily "file.txt")
which will be backed up each day to
file.txt.YYYYMMDD
<a id="sec-8-4-1" name="sec-8-4-1"></a>I want both log file and backup log file to have YYYYMMDD prefix or roll once per week
(log:config :daily "file.txt.%Y%m%d")
file will roll when %Y%m%d expansion
changes.
<a id="sec-8-4-2" name="sec-8-4-2"></a>What about just one plain file, without rolling
(log:config :daily "file.txt" :backup nil)