A fast optimizing Brainfuck interpreter in pure python


Keywords
brainfuck, brainfuck-interpreter, brainfuck-language, esoteric-interpreter, esoteric-language, esoteric-languages, esoteric-programming-language, pure-python, python
License
Apache-2.0
Install
pip install bfi==1.0

Documentation

Fast Brainfuck interpreter in pure python

This is a pure python interpreter for the Brainfuck esoteric programming language. bfi is quite fast without requiring any special python implementations or compiled extension modules. bfi Supports Python 2x and 3x.

bfi achieves a significant speedup in the execution of brainfuck programs by first compiling brainfuck source code into an intermediate form. This intermediate form takes advantage of common brainfuck programming constructs to execute much faster than if we were to interpret & execute the brainfuck source directly.

Take moving the cell pointer, as a relativey simple example; to execute <<<<<<<<<<, we could iterate over each < character, and perform 10 separate "cell pointer decrement" operations. This would be the slow option. Alternatively, we could collapse those 10 instructions into a single instruction to decrement the cell pointer by 10 in a single operation. This is generally how the opcodes for the intermediate form work. All runs of cell pointer increment/decrements are collapsed like this, as well as several other similar optimizations.

Speed benchmark

Here is a quick comparison between bfi and two other popular pure-python brainfuck interpreters on github. The time shown is the time that each interpreter took to complete the "Towers of Hanoi" program (hanoi.b, available in the examples directory):

Interpreter name Time to complete hanoi.b
bfi 1 minute, 9 seconds
pocmo's interpreter 28 minutes, 51 seconds
alexprengere's intrepreter 1 hour, 7 minutes, 54 seconds

(I should note here that alexprengere's interpreter can actually go much faster than this, but not without using the alternative PyPy interpreter, or compiling some stuff. Speeds here are shown without such modifications. All tests were done using the standard CPython 2.7.14 interpreter)

Implementation details

  • No change on EOF
  • Tape size is configurable, default is 30,000 cells
  • Cells are one byte, valid values between 0-255. Overflow/underflow wraps around

Installing

Use pip to install:

pip install bfi

Using the interpreter from the command-line

Once installed, the brainfuck interpreter can be invoked from the command line using the bfi command. Just run bfi and pass a brainfuck source file. Several sample Brainfuck programs are provided in the examples directory within the installed package (in your system's python2.7/dist-packages directory- on linux-based systems, for example, the full path might be /usr/local/lib/python2.7/dist-packages/bfi/examples).

In the sample commands below, we will run "Lost Kingdom", a text-based adventure game written in Brainfuck:

$> cd <dist-packages-directory>/bfi/examples
$> bfi LostKingdom.b

Using the interpreter in your own code

Here is how you use the bfi module to execute some Brainfuck code normally (reading data directly from stdin and writing directly to stdout):

>>> import bfi
>>> with open('samples/hello_world.b', 'r') as fh:
...     brainfuck_code = fh.read()
...
>>> Brainfuck.interpret(brainfuck_code)

Hello World!

Here is how you use the bfi module to execute some Brainfuck code without reading/writing the user's terminal; input is passed a parameter to interpret(), and any output is returned as a string.

>>> input_data = "test input"
>>> ret = bfi.interpret(brainfuck_code, stdin=input_data, buffer_stdout=True)
>>> print ret

Hello World!

Reference

Documentation for the python API is here: https://bfi.readthedocs.io

Gratuitous unnecessary extras

In order to make Brainfuck code execute more efficiently, it is compiled into an intermediate form that takes advantage of common brainfuck idioms and constructs. This intermediate form consists of 11 opcodes, 8 of which are similar to the original 8 brainfuck instructions. The following table describes the opcodes:

Opcode Description
move <off> <num> Moves the cell pointer by <num> cells. <off> is unused
sub <off> <num> Moves the cell pointer by <off>, and decrements value of current cell by <num> cells
add <off> <num> Moves the cell pointer by <off>, and increments value of current cell by <num> cells
open <off> <location> <location> is an index into the list of program opcodes. If the value of current cell is zero, jump to <location>. Otherwise, continue execution normally (Same functionality as brainfuck "[" instruction, except jump location is stored with opcode). <off> is unused
close <off> <location> <location> is an index into the list of program opcodes. If the value of current cell is zero, continue execution normally. Otherwise, jump to <location> (Same functionality as brainfuck "]" instruction, except jump location is stored with opcode). In all cases the cell pointer will be moved by <off>
input <off> Moves the cell pointer by <off>, then reads one character of input and writes to current cell
output <off> Moves the cell pointer by <off>, then prints value of current cell as an ASCII character
clear <off> Moves the cell pointer by <off>, then sets the value of current cell to zero
copy <off> {<o>:<m>,... } Moves the cell pointer by <off>, then for each key/value pair, sets the value of the cell at (current cell + <o>) to be (value of current cell * <m>)
scanl <off> Moves the cell pointer by <off>, then decrements the cell pointer until it points at a cell containing 0
scanr <off> Moves the cell pointer by <off>, then increments the cell pointer until it points at a cell containing 0

If you really want to, you can actually view a brainfuck program in this intermediate form, by using the bfi.parse method and printing the resulting opcodes:

>>> with open('bfi/examples/mandel.b', 'r') as fh:
...     program = fh.read()
...
>>> opcodes = bfi.parse(program)
>>> for c in opcodes: print c
...

add 0 13
copy 0 {1: 2, 4: 5, 5: 2, 6: 1}
add 5 6
sub 1 3
add 10 15
open 0 12
open 0 7
close 9 6
add 0 1
open 0 10

... (long output, truncated ...)

And of course, you can execute the compiled opcodes as many times as you like using bfi.execute.

Example Brainfuck programs

I have included several random Brainfuck programs that I've found in various places. I didn't write any of these programs, I just copied them as-is from other public sources. Descriptive comments (and author's name, in some cases) can be seen in the Brainfuck source files themselves.

A description of the example Brainfuck programs included with this package follows:

  • bfcl.bf: A Brainfuck-to-ELF translator, in Brainfuck. Reads in Brainfuck source from stdin and writes a Linux ELF file to stdout
  • bitwidth.bf Assorted tests for Brainfuck interpreter/compiler correctness
  • collatz.b A demonstration of the Collatz problem in Brainfuck
  • eoftest.b Tests EOF behaviour of brainfuck interpreters/compilers
  • fib.b Prints a neverending fibonacci sequence
  • gameoflife.b Conway's Game of Life in Brainfuck
  • hanoi.b Towers of Hanoi in Brainfuck
  • hello_world.b Classic "hello, world!" in Brainfuck
  • LostKingdom.b A text-based adventure game in Brainfuck
  • mandel.b An ASCII mandelbrot fractal set viewer in Brainfuck
  • numwarp.b Prints an enlarged ASCII representation of numbers entered by the user
  • primes.bf Prints prime numbers
  • rot13.b Prints the ROT13 encoding of the string entered by the user
  • sierpinksi.b Displays the Sierpinksi triangle
  • TheBrainfuckedLoneWolf.b ASCII asteroids-inspired top-down shooter game in Brainfuck