This is a brief introduction to Python for Lisp programmers. (Although it wasn't my intent, Python programers have told me this page has helped them learn Lisp.) Basically, Python can be seen as a dialect of Lisp with "traditional" syntax (what Lisp people call "infix" or "m-lisp" syntax). One message on comp.lang.python said "I never understood why LISP was a good idea until I started playing with python." Python supports all of Lisp's essential features except macros, and you don't miss macros all that much because it does have eval, and operator overloading, and regular expression parsing, so some--but not all--of the use cases for macros are covered.
I looked into Python because I was considering translating the Lisp code for the Russell & Norvig AI textbook into Java. Some instructors and students wanted Java because
- That's the language they're most familiar with from other courses.
- They want to have graphical applications.
- A minority want applets in browsers.
- Some just couldn't get used to Lisp syntax in the limited amount of class time they had to devote to it.
However, our first attempt at writing Java versions was largely unsuccesful. Java was too verbose, and the differences between the pseudocode in the book and the Java code was too large. I looked around for a language that was closer to the pseudocode in the book, and discovered Python was the closest. Furthermore, with and Jython, I could target the Java JVM.
My conclusion
Python is an excellent language for my intended use. It is easy to use (interactive with no compile-link-load-run cycle), which is important for my pedagogical purposes. While Python doesn't satisfy the prerequisite of being spelled J-A-V-A, Jython is close. Python seems to be easier to read than Lisp for someone with no experience in either language. The Python code I developed looks much more like the (independently developed) pseudo-code in the book than does the Lisp code. This is important, because some students were complaining that they had a hard time seeing how the pseudo-code in the book mapped into the online Lisp code (even though it seemed obvious to Lisp programmers).
The two main drawbacks of Python from my point of view are (1) there is very little compile-time error analysis and type declaration, even less than Lisp, and (2) execution time is much slower than Lisp, often by a factor of 10 (sometimes by 100 and sometimes by 1). Qualitatively, Python feels about the same speed as interpreted Lisp, but very noticably slower than compiled Lisp. For this reason I wouldn't recommend Python for applications that are (or are likely to become over time) compute intensive (unless you are willing to move the speed bottlenecks into C). But my purpose is oriented towards pedagogy, not production, so this is less of an issue.
Introducing Python
Python can be seen as either a practical (better libraries) version of Scheme, or as a cleaned-up (no $@&% characters) version of Perl. While Perl's philosophy is TIMTOWTDI (there's more than one way to do it), Python tries to provide a minimal subset that people will tend to use in the same way (maybe TOOWTDI for there's only one way to do it, but of course there's always more than one way if you try hard). One of Python's controversial features, using indentation level rather than begin/end or braces, was driven by this philosophy: since there are no braces, there are no style wars over where to put the braces. Interestingly, Lisp has exactly the same philosphy on this point: everyone uses emacs to indent their code, so they don't argue over the indentation. Take a Lisp program, indent it properly, and delete the opening parens at the start of lines and their matching close parens, and you end up with something that looks rather like a Python program.
Python has the philosophy of making sensible compromises that make the easy things very easy, and don't preclude too many hard things. In my opinion it does a very good job. The easy things are easy, the harder things are progressively harder, and you tend not to notice the inconsistencies. Lisp has the philosophy of making fewer compromises: of providing a very powerful and totally consistent core. This can make Lisp harder to learn because you operate at a higher level of abstraction right from the start and because you need to understand what you're doing, rather than just relying on what feels or looks nice. But it also means that in Lisp it is easier to add levels of abstraction and complexity; Lisp is optimized to make the very hard things not too hard, while Python is optimized to make medium hard things easier.
Here I've taken a blurb from Python.org and created two vesions of it: one for Python in blue italics and one for Lisp in green bold. The bulk of the blurb, common to both languages, is in black.
Python/Lisp is an interpreted and compiled, object-oriented, high-level programming language with dynamic semantics. Its high-level built in data structures, combined with dynamic typing and dynamic binding, make it very attractive for Rapid Application Development, as well as for use as a scripting or glue language to connect existing components together. Python/Lisp's simple, easy to learn syntax emphasizes readability and therefore reduces the cost of program maintenance. Python/Lisp supports modules and packages, which encourages program modularity and code reuse. The Python/Lispinterpreter and the extensive standard library are available in source or binary form without charge for all major platforms, and can be freely distributed. Often, programmers fall in love with Python/Lispbecause of the increased productivity it provides. Since there is no separate compilation step, the edit-test-debug cycle is incredibly fast. Debugging Python/Lisp programs is easy: a bug or bad input will never cause a segmentation fault. Instead, when the interpreter discovers an error, it raises an exception. When the program doesn't catch the exception, the interpreter prints a stack trace. A source level debugger allows inspection of local and global variables, evaluation of arbitrary expressions, setting breakpoints, stepping through the code a line at a time, and so on. The debugger is written inPython/Lisp itself, testifying to Python/Lisp's introspective power. On the other hand, often the quickest way to debug a program is to add a few print statements to the source: the fast edit-test-debug cycle makes this simple approach very effective.
To which I can only add:
Although some people have initial resistance to the indentation as block structure/parentheses, most come to like/deeply appreciate them.
To learn more about Python, if you are an experienced programmer, I recommend going to the download page at Python.org and getting the documentation package, and paying particular attention to the Python Reference Manual and the Python Library Reference. There are all sorts of tutorials and published books, but these references are what you really need.
The following table serves as a Lisp/Python translation guide. Entries in red mark places where one language is noticibly worse, in my opinion. Entries in bold mark places where the languages are noticibly different, but neither approach is clearly better. Entries in regular font mean the languages are similar; the syntax might be slightly different, but the concepts are the same or very close. The table is followed by a list of gotchas and some sample programs in Python.
| Key Features | Lisp Features | Python Features |
| Everything is an object | Yes | Yes |
| Objects have type, variables don't | Yes | Yes |
| Support heterogeneous lists | Yes (linked list and array/vector) | Yes (array) |
| Multi-paradigm language | Yes: Functional, Imperative, OO, Generic | Yes: Functional, Imperative, OO |
| Storage management | Automatic garbage collection | Automatic garbage collection |
| Packages/Modules | Harder to use | Easy to use |
| Introspection of objects, classes | Strong | Strong |
| Macros for metaprogramming | Powerful macros | No macros |
| Interactive Read-eval-print loop | > (string-append "hello" " " "world") "hello world" | >>> ' '.join(['hello', 'world']) 'hello world' |
| Concise expressive language | (defun transpose (m) (apply #'mapcar #'list m)) > (transpose '((1 2 3) (4 5 6))) ((1 4) (2 5) (3 6)) | def transpose (m): return zip(*m) >>> transpose([[1,2,3], [4,5,6]]) [(1, 4), (2, 5), (3, 6)] |
| Cross-platform portability | Windows, Mac, Unix, Gnu/Linux | Windows, Mac, Unix, Gnu/Linux |
| Number of implementations | Many | One main, plus branches (e.g. Jython, Stackless) |
| Development Model | Proprietary and open source | Open source |
| Efficiency | About 1 to 2 times slower than C++ | About 2 to 100 times slower than C++ |
| GUI, Web, etc. librariees | Not standard | GUI, Web libraries standard |
| Methods Method dispatch | Dynamic, (meth obj arg) syntax runtime-type, multi-methods | Dynamic, obj.meth(arg) syntax runtime-type, single class-based |
| Data Types | Lisp Data Types | Python Data Types |
| Integer Bignum Float Complex String Symbol Hashtable/Dictionary Function Class Instance Stream Boolean Empty Sequence Missing Value Lisp List (linked) Python List (adjustable array) Others | 42 100000000000000000 12.34 #C(1, 2) "hello" hello (make-hash-table) (lambda (x) (+ x x)) (defclass stack ...) (make 'stack) (open "file") t, nil (), #() linked list, array nil (1 2.0 "three") (make-arrary 3 :adjustable t :initial-contents '(1 2 3)) Many (in core language) | 42 100000000000000000 12.34 1 + 2J "hello" or 'hello' ## immutable 'hello' {} lambda x: x + x class Stack: ... Stack() open("file") True, False (), [] tuple, array None (1, (2.0, ("three", None))) [1, 2.0, "three"] Many (in libraries) |
| Control Structures | Lisp Control Structures | Python Control Structures |
| Statements and expressions | Everything is an expression | Distinguish statements from expressions |
| False values | nil is only false value | False, None, 0, '', [ ], {} are all false |
| Function call | (func x y z) | func(x,y,z) |
| Conditional test | (if x y z) | if x: y else: z |
| Conditional expression | (if x y z) | y if x else z |
| While loop | (loop while (test) do (f)) | while test(): f() |
| Other loops | (dotimes (i n) (f i)) (loop for x in s do (f x)) (loop for (name addr salary) in db do ...) | for i in range(n): f(i) for x in s: f(x) ## works on any sequence for (name, addr, salary) in db: ... |
| Assignment | (setq x y) (psetq x 1 y 2) (rotatef x y) (setf (slot x) y) (values 1 2 3) on stack (multiple-value-setq (x y) (values 1 2)) | x = y x, y = 1, 2 x, y = y, x x.slot = y (1, 2, 3) uses memory in heap x, y = 1, 2 |
| Exceptions | (assert (/= denom 0)) (unwind-protect (attempt) (recovery)) (catch 'ball ... (throw 'ball)) | assert denom != 0, "denom != 0" try: attempt() finally: recovery() try: ...; raise 'ball' except 'ball': ... |
| Other control structures | case, etypecase, cond, with-open-file, etc. | Extensible with statement No other control structures |
| Lexical Structure | Lisp Lexical Structure | Python Lexical Structure |
| Comments | ;; semicolon to end of line | ## hash mark to end of line |
| Delimiters | Parentheses to delimit expressions: (defun fact (n) (if (<= n 1) 1 (* n (fact (- n 1))))) | Indentation to delimit statements: def fact (n): if n <= 1: return 1 else: return n * fact(n — 1) |
| Higher-Order Functions | Lisp Higher-Order Functions | Python Higher-Order Functions |
| Function application evaluate an expression execute a statement load a file | (apply fn args) (eval '(+ 2 2)) => 4 (eval '(dolist (x list) (f x))) (load "file.lisp") or (require 'file) | apply(fn, args) or fn(*args) eval("2+2") => 4 exec("for x in list: f(x)") execfile("file.py") or import file |
| Sequence functions | (mapcar length '("one" (2 3))) => (3 2) (reduce #'+ numbers) (every #'oddp '(1 3 5)) => T (some #'oddp '(1 2 3)) => 1 (remove-if-not #'evenp numbers) (reduce #'min numbers) | map(len, ["one", [2, 3]]) => [3, 2] or [len(x) for x in ["one", [2, 3]]] reduce(operator.add, numbers) all(x%2 for x in [1,3,5]) => True any(x%2 for x in [1,2,3]) => True filter(lambda x: x%2 == 0, numbers) or [x for x in numbers if x%2 == 0] min(numbers) |
| Other higher-order functions | count-if, etc. :test, :key, etc keywords | No other higher-order functions built-in No keywords on map/reduce/filter |
| Close over read-only var Close over writable var | (lambda (x) (f x y)) (lambda (x) (incf y x)) | lambda x: f(x, y) Can't be done; use objects |
| Parameter Lists | Lisp Parameter Lists | Python Parameter Lists |
| Optional arg Variable-length arg Unspecified keyword args Calling convention | (defun f (&optional (arg val) ...) (defun f (&rest arg) ...) (defun f (&allow-other-keys &rest arg) ...) Call with keywords only when declared: (defun f (&key x y) ...) (f :y 1 :x 2) | def f (arg=val): ... def f (*arg): ... def f (**arg): ... Call any function with keywords: def f (x,y): ... f(y=1, x=2) |
| Efficiency | Lisp Efficiency Issues | Python Efficiency Issues |
| Compilation Function reference resolution Declarations | Compiles to native code Most function/method lookups are fast Declarations can be made for efficiency | Compiles to bytecode only Most function/method lookups are slow No declarations |
| Features | Lisp Features and Functions | Python Features and Functions |
| Quotation | Quote whole list structure: 'hello '(this is a test) '(hello world (+ 2 2)) | Quote individual strings or .split(): 'hello' 'this is a test'.split() ['hello', 'world', [2, "+", 2]] |
| Introspectible doc strings | (defun f (x) "compute f value" ...) > (documentation 'f 'function) "compute f value" | def f(x): "compute f value" ... >>> f.__doc__ "compute f value" |
| List access | Via functions: (first list) (setf (elt list n) val) (first (last list)) (subseq list start end) (subseq list start) | Via syntax: list[0] list[n] = val list[-1] list[start:end] list[start:] |
| Hashtable access | Via functions: (setq h (make-hash-table)) (setf (gethash "one" h) 1.0) (gethash "one" h) (let ((h (make-hash-table))) (setf (gethash "one" h) 1) (setf (gethash "two" h) 2) h) | Via syntax: h = {} h["one"] = 1.0 h["one"] or h.get("one") h = {"one": 1, "two": 2} |
| Operations on lists | (cons x y) (car x) (cdr x) (equal x y) (eq x y) nil (length seq) (vector 1 2 3) | [x] + y but O(n); also y.append(x) x[0] x[1:] but O(n) x == y x is y () or [ ] len(seq) (1, 2, 3) |
| Operations on arrays | (make-array 10 :initial-element 42) (aref x i) (incf (aref x i)) (setf (aref x i) 0) (length x) #(10 20 30) if size unchanging | 10 * [42] x[i] x[i] += 1 x[i] = 0 len(x) [10, 20, 30] |
An important point for many people is the speed of Python and Lisp versus other languages. Its hard to get benchmark data that is relevent to your set of applications, but this may be useful:
| Test | Lisp | Java | Python | Perl | C++ | ||
|---|---|---|---|---|---|---|---|
| hash access | 1.06 | 3.23 | 4.01 | 1.85 | 1.00 | ||
| exception handling | 0.01 | 0.90 | 1.54 | 1.73 | 1.00 | Legend | |
| sum numbers from file | 7.54 | 2.63 | 8.34 | 2.49 | 1.00 | > 100 x C++ | |
| reverse lines | 1.61 | 1.22 | 1.38 | 1.25 | 1.00 | 50-100 x C++ | |
| matrix multiplication | 3.30 | 8.90 | 278.00 | 226.00 | 1.00 | 10-50 x C++ | |
| heapsort | 1.67 | 7.00 | 84.42 | 75.67 | 1.00 | 5-10 x C++ | |
| array access | 1.75 | 6.83 | 141.08 | 127.25 | 1.00 | 2-5 x C++ | |
| list processing | 0.93 | 20.47 | 20.33 | 11.27 | 1.00 | 1-2 x C++ | |
| object instantiation | 1.32 | 2.39 | 49.11 | 89.21 | 1.00 | < 1 x C++ | |
| word count | 0.73 | 4.61 | 2.57 | 1.64 | 1.00 | ||
| Median | 1.67 | 4.61 | 20.33 | 11.27 | 1.00 | ||
| 25% to 75% | 0.93 to 1.67 | 2.63 to 7.00 | 2.57 to 84.42 | 1.73 to 89.21 | 1.00 to 1.00 | ||
| Range | 0.01 to 7.54 | 0.90 to 20.47 | 1.38 to 278 | 1.25 to 226 | 1.00 to 1.00 |
Speeds are normalized so the g++ compiler for C++ is 1.00, so 2.00 means twice as slow; 0.01 means 100 times faster. For Lisp, the CMUCL compiler was used. Background colors are coded according to legend on right. The last three lines give the mean score, 25% to 75% quartile scores (throwing out the bottom two and top two scores for each language), and overall range. Comparing Lisp and Python and throwing out the top and bottom two, we find Python is 3 to 85 times slower than Lisp -- about the same as Perl, but much slower than Java or Lisp. Lisp is about twice as fast as Java.
Gotchas for Lisp Programmers in Python
Here I list conceptual problems for me as a Lisp programmer coming to Python:
- Lists are not Conses. Python lists are actually like adjustable arrays in Lisp or Vectors in Java. That means that list access is O(1), but that the equivalent of both cons and cdr generate O(n) new storage. You really want to use map or for e in x: rather than car/cdr recursion. Note that there are multiple empty lists, not just one. This fixes a common bug in Lisp, where users do (nconc old new) and expect oldto be modified, but it is not modified when old is nil. In Python, old.extend(new) works even when old is [ ]. But it does mean that you have to test against [] with ==, not is, and it means that if you set a default argument equal to [] you better not modify the value.
- Python is less functional. Partially because lists are not conses, Python uses more destructive functions than Lisp, and to emphasize that they are destructive, they tend to return None. You might expect to be able to do for x in list.reverse(), but Python's reverse is like nreverse but returns None. You need to do it in several statements, or write your own reverse function. Besides reverse, this is also true for remove and sort, among others.
- Python classes are more functional. In Lisp (CLOS), when you redefine a class C, the object that represents C gets modified. Existing instances and subclasses that refer to C are thus redirected to the new class. This can sometimes cause problems, but in interactive debugging this is usually what you want. In Python, when you redefine a class you get a new class object, but the old instances and subclasses still refer to the old class. This means that most of the time you have to reload your subclasses and rebuild your data structures every time you redefine a class. If you forget, you can get confused.
- Python is more dynamic, does less error-checking. In Python you won't get any warnings for undefined functions or fields, or wrong number of arguments passed to a function, or most anything else at load time; you have to wait until run time. The commercial Lisp implementations will flag many of these as warnings; simpler implementations like clisp do not. The one place where Python is demonstrably more dangerous is when you do self.feild = 0 when you meant to type self.field = 0; the former will dynamically create a new field. The equivalent in Lisp, (setf (feild self) 0) will give you an error. On the other hand, accessing an undefined field will give you an error in both languages.
- Don't forget self. This is more for Java programmers than for Lisp programmers: within a method, make sure you do self.field, not field. There is no implicit scope. Most of the time this gives you a run-time error. It is annoying, but I suppose one learns not to do it after a while.
- Don't forget return. Writing def twice(x): x+x is tempting and doesn't signal a warning or exception, but you probably meant to have a return in there. This is particularly irksome because in a lambdayou are prohibited from writing return, but the semantics is to do the return.
- Watch out for singleton tuples. A tuple is just an immutable list, and is formed with parens rather than square braces. () is the empty tuple, and (1, 2) is a two-element tuple, but (1) is just 1. Use (1,)instead. Yuck. Damian Morton pointed out to me that it makes sense if you understand that tuples are printed with parens, but that they are formed by commas; the parens are just there to disambiguate the grouping. Under this interpretation, 1, 2 is a two element tuple, and 1, is a one-element tuple, and the parens are sometimes necessary, depending on where the tuple appears. For example, 2, + 2, is a legal expression, but it would probably be clearer to use (2,) + (2,) or (2, 2).
- Watch out for certain exceptions. Be careful: dict[key] raises KeyError when key is missing; Lisp hashtable users expect nil. You need to catch the exception or test with key in dict.
- Python is a Lisp-1. By this I mean that Python has one namespace for functions and variables, like Scheme, not two like Common Lisp. For example:
def f(list, len): return list((len, len(list))) ## bad Python
(define (f list length) (list length (length list))) ;; bad Scheme
(defun f (list length) (list length (length list))) ;; legal Common LispThis also holds for fields and methods: you can't provide an abstraction level over a field with a method of the same name:class C:
def f(self): return self.f ## bad Python
... - Python Pre-2.1 did not have lexical scopes. In Python before version 2.1 there were only two variable scopes per module: global scope and function scope. In Python 2.1, released in April 2001, if you do "from __future__ import nested_scopes", you add a third scope, block nested scope. In Python 2.2, this is the default behavior. This is what you want. Now you can create closures over read-only variables. If you do need to modify a variable that you close over in a function, you have a few options, all of them a little cumbersome. You can wrap the variable(s) in 1-element lists:
def sum(items):
total = [0.0]
def f(x): total[0] = total[0] + x
map(f, items)
return total[0]
>>> sum([1.1, 2.2, 3.3])
6.6Notice also that you could not use a lambda here, because the lambda function body must be a single expression, not a statement. The other option is to use objects instead of functions, but this is overly verbose. Still, verbosity is in the eye of the beholder. Lisp programmers think that (lambda (x) (* k x)) is about right, but Smalltalk programmers think this is way too much, they use [:x | x * k], while Java programmers put up with a ridiculously verbose inner class expression such as:new Callable() {
public Object call(Object x) {
return x.times(k)
}
} - Python strings are not quite like Lisp symbols. Python does symbol lookup by interning strings in the hash tables that exist in modules and in classes. That is, when you write obj.slot Python looks for the string "slot" in the hash table for the class of obj, at run time. Python also interns some strings in user code, for example when you say x = "str". But it does not intern strings that don't look like variables, as in x = "a str" (thanks to Brian Spilsbury for pointing this out).
- Python does not have macros. Python does have access to the abstract syntax tree of programs, but this is not for the faint of heart. On the plus side, the modules are easy to understand, and with five minutes and five lines of code I was able to get this:
>>> parse("2 + 2")
['eval_input', ['testlist', ['test', ['and_test', ['not_test', ['comparison',
['expr', ['xor_expr', ['and_expr', ['shift_expr', ['arith_expr', ['term',
['factor', ['power', ['atom', [2, '2']]]]], [14, '+'], ['term', ['factor',
['power', ['atom', [2, '2']]]]]]]]]]]]]]], [4, ''], [0, '']]This was rather a disapointment to me. The Lisp parse of the equivalent expression is (+ 2 2). It seems that only a real expert would want to manipulate Python parse trees, whereas Lisp parse trees are simple for anyone to use. It is still possible to create something similar to macros in Python by concatenating strings, but it is not integrated with the rest of the language, and so in practice is not done. In Lisp, there are two main purposes for macros: new control structures, and custom problem-specific languages. The former is just not done in Python. The later can be done for data with a problem-specific format in Python: below I define a context-free grammar in Python using a combination of the builtin syntax for dictionaries and a preprocessing step that parses strings into data structures. The combination is almost as nice as Lisp macros. But more complex tasks, such as writing a compiler for a logic programming language, are easy in Lisp but hard in Python.
Comparing Lisp and Python Programs
I took the first example program from Paradigms of Artificial Intelligence Programming, a simple random sentence generator and translated it into Python. Conclusions: conciseness is similar; Python gains becausegrammar[phrase] is simpler than (rule-rhs (assoc phrase *grammar*)), but Lisp gains because '(NP VP) beats ['NP', 'VP']. The Python program is probably less efficient, but that's not the point. Both languages seem very well suited for programs like this. Make your browser window wide to see this properly.
| Lisp Program simple.lisp | Python Program simple.py |
(defparameter *grammar* | from random import choice |
| Running the Lisp Program | Running the Python Program |
> (generate 'S) | >>> generate('S') |
I was concerned that the grammar is uglier in Python than in Lisp, so I thought about writing a Parser in Python (it turns out there are some already written and freely available) and about overloading the builtin operators. This second approach is feasible for some applications, such as my Expr class for representing and manipulating logical expressions. But for this application, a trivial ad-hoc parser for grammar rules will do: a grammar rule is a list of alternatives, separated by '|', where each alternative is a list of words, separated by ' '. That, plus rewriting the grammar program in idiomatic Python rather than a transliteration from Lisp, leads to the following program:
| Python Program simple.py (idiomatic version) |
"""Module to generate random sentences from a grammar. The grammar |
