IMPORTANT: cl-format has been moved into clojure.contrib. The code here is no longer maintained. This README describes the version in clojure.contrib (which is at http://code.google.com/p/clojure-contrib/.)
To use pretty printing in clojure.contrib, it is necessary to build clojure.contrib with compiled classes. To do this, you must tell ant where to find clojure.jar. For me, this looks like:
ant -Dclojure.jar=../clojure/clojure.jar
because I keep clojure source and clojure.contrib source right next to each other. Just point the pathname to wherever you keep clojure.jar.
This library adds two new features to Clojure: a generalized pretty printer and a Common Lisp-compatible format function.
The pretty printer is easy to use:
user=> (println (for [x (range 10)] (range x)))
(() (0) (0 1) (0 1 2) (0 1 2 3) (0 1 2 3 4) (0 1 2 3 4 5) (0 1 2 3 4 5 6) (0 1 2 3 4 5 6 7) (0 1 2 3 4 5 6 7 8))
nil
user=> (use 'clojure.contrib.pprint)
nil
user=> (pprint (for [x (range 10)] (range x)))
(()
(0)
(0 1)
(0 1 2)
(0 1 2 3)
(0 1 2 3 4)
(0 1 2 3 4 5)
(0 1 2 3 4 5 6)
(0 1 2 3 4 5 6 7)
(0 1 2 3 4 5 6 7 8))
nil
user=>
The pretty printer supports two modes: code which has special formatting for special forms and core macros and simple (the default) which formats the various Clojure data structures as appropriate for raw data. In the future, the pretty printer will be highly customizable, but right now it is pretty simple.
The Common Lisp-compatible format function is a 100% comptible implementation of format from Common Lisp (the one incompatibility is that it's called cl-format in Clojure because format existed and meant something different).
All the functions and variables described here are in the clojure.contrib.pprint namespace. Using them is as simple as adding cl-format.jar to your classpath and adding a (:use clojure.contrib.pprint) to your namespace declarations.
cl-format is being developed by Tom Faulhaber (to mail me you can use my first name at my domain which is infolace.com).
cl-format has been developed completely from scratch and is licensed under the Eclipse Public License 1.0 (like Clojure itself) which can be found in the file epl-v10.html at the root of this distribution. This means it can go anywhere Clojure can go.
cl-format is hosted at github at http://github.com/tomfaulhaber/cl-format.
You can just grab the jar file here: http://github.com/tomfaulhaber/cl-format/raw/master/release/cl-format.jar
Future development is guided by those using it, so send feedback about what's working and not working for you and what you'd like to see in cl-format.
Pretty printing is primarily implemented with the function pprint. pprint takes a single argument and formats it according to the settings of several special variables.
Generally, the defaults are fine for pretty printing and you can simply use:
(pprint obj)
to print your object. If you wish to write to another stream besides *out*, you can use:
(write obj :pretty true :stream foo)
where foo is the stream to which you wish to write. (The write function has a lot more options which are not yet documented. Stay tuned.)
When at the REPL, the pp macro pretty prints the last output value. This is useful when you get something too complex to read comfortably. Just type:
user=> (pp)
and you'll get a pretty printed version of the last thing output (the magic variable *1).
The behavior of the pretty printer can be finely controlled through the use of dispatch tables that contain descriptions for how different structures should be formatted. The exact design of the dispatch table is still evolving and is, therefore, not yet documented.
The pretty printer comes with two pre-defined dispatch tables to cover the most common situations:
*simple-dispatch* - supports basic representation of data in various Clojure structures: seqs, maps, vectors, etc. in a fairly statndard way. When structures need to be broken across lines, following lines are indented to line up with the first element. *simple-dispatch* is the default and is good from showing the output of most operations.
*code-dispatch* - has special representation for various structures found in code: defn, condp, binding vectors, anonymous functions, etc. This dispatch indents following lines of a list one more space as appropriate for a function/argument type of list.
An example formatted with code dispatch:
user=> (def code '(defn cl-format
"An implementation of a Common Lisp compatible format function"
[stream format-in & args] (let [compiled-format (if (string? format-in)
(compile-format format-in) format-in) navigator (init-navigator args)]
(execute-format stream compiled-format navigator))))
#'user/code
user=> (with-pprint-dispatch *code-dispatch* (pprint code))
(defn cl-format
"An implementation of a Common Lisp compatible format function"
[stream format-in & args]
(let [compiled-format (if (string? format-in)
(compile-format format-in)
format-in)
navigator (init-navigator args)]
(execute-format stream compiled-format navigator)))
nil
user=>
There are two ways to set the current dispatch: set it to a specific table permanantly with set-pprint-dispatch or bind it with with-pprint-dispatch (as shown in the example above).
The operation of pretty printing is also controlled by a set of variables that control general parameters of how the pretty printer makes decisions. The current list is as follows:
*print-pretty*: Default: true
Bind to true if you want write to use pretty printing. (pprint and pp automatically bind this to true.)
*print-right-margin*: Default: 72
Pretty printing will try to avoid anything going beyond this column.
*print-miser-width*: Default: 40
The column at which to enter miser style. Depending on the dispatch table, miser style add newlines in more places to try to keep lines short allowing for further levels of nesting. For example, in the code dispatch table, the pretty printer will insert a newline between the "if" and its condition when in miser style.
*print-suppress-namespaces*: Default: false
Don't print namespaces with symbols. This is particularly useful when pretty printing the results of macro expansions
*print-level*: Default: nil
As with the regular Clojure print function, this variable controls the depth of structure that is printed. The argument itself is level 0, the first level of a collection is level 1, etc. When the structure gets deeper than the specified *print-level*, a hash sign (#) is printed.
For example:
user=> (binding [*print-level* 2] (pprint '(a b (c d) ((e) ((f d) g)))))
(a b (c d) (# #))
nil
user=>
*print-length*: Default: nil
As with the regular Clojure print function, this variable controls the number of items that are printed at each layer of structure. When a layer has too many items, elipses (...) are displayed.
For example:
user=> (defn foo [x] (for [i (range x) ] (range 1 (- x (dec i)))))
#'user/foo
user=> (binding [*print-length* 6] (pprint (foo 10)))
((1 2 3 4 5 6 ...)
(1 2 3 4 5 6 ...)
(1 2 3 4 5 6 ...)
(1 2 3 4 5 6 ...)
(1 2 3 4 5 6)
(1 2 3 4 5)
...)
nil
user=>
This is an early version release of the pretty printer and there is plenty that is yet to come.
Here are some examples:
- Support all the types and forms in Clojure (most of the way there now).
- Customized pretty printing functions and dispatch tables. (They are there under the hood, but the implementation is still evolving.)
- Support for limiting pretty printing based on line counts.
- Support for circular and shared substructure detection.
- Finishing the integration with the format function (support for ~/ and tabular pretty printing).
- Performance! (Not much thought has been made to making this go fast, but there are a bunch of pretty obvious speedups to be had.)
- Handle Java objects intelligently
Please let me know about anything that's not working right, anything that should work differently, or the feature you think should be at the top of my list.
cl-format is an implementation of the incredibly baroque Common Lisp format function as specified in Common Lisp the Language, 2nd edition, Chapter 22.
Format gives you an easy and powerful way to format text and data for output. It supports rich formatting of strings and numbers, loops, conditionals, embedded formats, etc. It is really a domain-specific language for formatting.
This implementation for clojure has the following goals:
- Support the full feature set of the Common Lisp format function (including the X3J13 extensions) with the only exception being concepts that make no sense or are differently interpreted in Clojure.
- Make porting code from Common Lisp easier.
- Provide a more native feeling solution for Clojure programmers than the Java format method and its relatives.
- Be fast. This includes the ability to precompile formats that are going to be used reptitively.
- Include useful error handling and comprehensive documentation.
For some people the answer to this question is that they are used to Common Lisp and, therefore, they already know the syntax of format strings and all the directives.
A more interesting answer is that cl-format provides a way of rendering strings that is much more suited to Lisp and its data structures.
Because iteration and conditionals are built into the directive structure of cl-format, it is possible to render sequences and other complex data structures directly without having to loop over the data structure.
For example, to print the elements of a sequence separated by commas, you simply say:
(cl-format true "~{~a~^, ~}" aseq)
(This example is taken from Practical Common Lisp by Peter Seibel.)
The corresponding output using Clojure's Java-based format function would involve a nasty loop/recur with some code to figure out about the commas. Yuck!
cl-format is 100% compatible with the Common Lisp standard as specified in CLtLv2. This includes all of the functionality of Common Lisp's format function including iteration, conditionals, text justification and rich options for displaying real and integer values. It also includes the directives to support pretty printing structured output.
If you find a bug in a directive, drop me a line with a chunk of code that exhibits the bug and the version of cl-format you found it in and I'll try to get it fixed.
I also intend to have good built-in documentation for the directives, but I haven't built that yet.
The following directives are not yet supported: ~:T and ~@:T (but all other forms of ~T work) and extensions with ~/.
The pretty printer interface is similar, but not identical to the interface in Common Lisp.
The custom dispatch table functionality is not fully fleshed out yet.
Next up:
- Support for ~/
- Documentation of the pretty printer.
- Restructure unit tests into modular chunks.
- Import tests from CLISP and SBCL.
- Unit tests for exception conditions.
- Interactive documentation
Once cl-format is in your path, adding it to your code is easy:
(ns your-namespace-here
(:use clojure.contrib.pprint))
If you want to refer to the cl-format function as "format" (rather than using the clojure function of that name), you can use this idiom:
(ns your-namespace-here
(:refer-clojure :exclude [format])
(:use clojure.contrib.pprint))
(def format cl-format)
You might want to do this in code that you've ported from Common Lisp, for instance, or maybe just because old habits die hard.
From the REPL, you can grab it using (require) and (refer):
(require 'clojure.contrib.pprint)
(refer 'clojure.contrib.pprint)
cl-format is a standard clojure function that takes a variable number of arguments. You call it like this:
(cl-format stream format args...)
stream can be any Java Writer (that is java.io.Writer) or the values true, false, or nil. The argument true is identical to using out while false or nil indicate that cl-format should return its result as a string rather than writing it to a stream.
format is either a format string or a compiled format (see below). The format string controls the output that's written in a way that's similar to (but much more powerful than) the standard Clojure API format function (which is based on Java's java.lang.String.Format).
Format strings consist of characters that are to be written to the output stream plus directives (which are marked by ~) as in "The answer is ~,2f". Format strings are documented in detail in Common Lisp the Language, 2nd edition, Chapter 22.
args is a set of arguments whose use is defined by the format.
When you use a format string many times (for example, when you're outputting in a loop), you can improve your performance by compiling the format with compile-format. The result of compile format can be passed to cl-format just like a format string but it doesn't need to be parsed.
For example:
(def log-format (compile-format "~2,'0D/~2,'0D/~D ~2D:~2,'0D ~:[PM,AM]: ~A~%"))
(defn log [msg]
(let [[m d y h min am?] (some-date-decomposition-fn)]
(cl-format log-format m d y h min am? msg)))
Writers in Java have no real idea of current column or device page width, so the format directives that want to work relative to the current position on the page have nothing to work with. To deal with this, cl-format contains an extension to writer called PrettyWriter. PrettyWriter watches the output and keeps track of what column the current output is going to.
When you call format and your format includes a directive that cares
about what column it's in (~T, ~&, <...>), cl-format will
automatically wrap the Writer you passed in with a PrettyWriter. This
means that by default all cl-format statements act like they begin on
a fresh line and have a page width of 72.
For many applications, these assumptions are fine and you need to do nothing more. But sometimes you want to use multiple cl-format calls that output partial lines. You may also want to mix cl-format calls with the native clojure calls like print. If you want stay column-aware while doingg this you need to create a PrettyWriter of your own (and possibly bind it to out).
As an example of this, this function takes a nested list and prints it as a table (returning the result as a string):
(defn list-to-table [aseq column-width]
(let [stream (PrettyWriter. (java.io.StringWriter.))]
(binding [*out* stream]
(doseq [row aseq]
(doseq [col row]
(cl-format true "~4D~7,vT" col column-width))
(prn)))
(.toString (.getWriter stream))))
(In reality, you'd probably do this as a single call to cl-format.)
The constructor to PrettyWriter takes the Writer it's wrapping and
(optionally) the page width (in columns) for use with <...>.
The following function uses cl-format to dump a columnized table of the Java system properties:
(defn show-props [stream]
(let [p (mapcat
#(vector (key %) (val %))
(sort-by key (System/getProperties)))]
(cl-format stream "~30A~A~%~{~20,,,'-A~10A~}~%~{~30A~S~%~}"
"Property" "Value" ["" "" "" ""] p)))
There are some more examples in the clojure.contrib.pprint.examples package:
- hexdump - a program that uses cl-format to create a standard formatted hexdump of the requested stream.
- multiply - a function to show a formatted multipication table in a very "first-order" way.
- props - the show-props example shown above.
- show_doc - some utilities for showing documentation from various name spaces.
The floating point directives that show exponents (~E, ~G) show E for the exponent character in all cases (unless overridden with an exponentchar). Clojure does not distinguish between floats and doubles in its printed representation and neither does cl-format.
The ~A and ~S directives accept the colon prefix, but ignore it since () and nil are not equivalent in Clojure.
Clojure has 3 different reader syntaxes for characters. The ~@c directive to cl-format has an argument extension to let you choose:
- ~@c (with no argument) prints "\c" (backslash followed by the printed representation of the character or \newline, \space, \tab, \backspace, \return)
- ~'o@c prints "\oDDD" where DDD are the octal digits representing the character.
- ~'u@c prints "\uXXXX" prints the hex Unicode representation of the character.