Python (no eval), 317 bytes

from sys import*
def f(u,c,k):
 while(c[1]>=k)*u:
 j,u='[]<>+-,.'.find(u[0]),u[1:];b=(j>=0)*(1-j%2*2);c[1]+=b*(j<2)
 while b*c[c[0]]and j<1:f(u,c,k+1);c[1]+=1
 b*=c[1]==k;c[[0,c[0],2][j/2-1]]+=b
 if(j==6)*b:c[c[0]]=ord(stdin.read(1))
 if(j>6)*b:stdout.write(chr(c[c[0]]))
f(open(argv[1]).read(),[-1]+[0]*30003,0)

16 bit 8086 machine code: 168 bytes

Here's the base64 encoded version, convert and save as 'bf.com' and run from Windows command prompt: 'bf progname'

gMYQUoDGEFKzgI1XAgIfiEcBtD3NIR8HcmOL2LQ/i88z0s0hcleL2DPA86sz/zP2/sU783NHrL0I
AGgyAU14DTqGmAF194qOoAH/4UfDJv4Fwyb+DcO0AiaKFc0hw7QBzSGqT8MmODV1+jPtO/NzDaw8
W3UBRTxddfJNee/DJjg1dPoz7U509YpE/zxddQFFPFt18U157sM+PCstLixbXUxjTlJWXmV+

EDIT

Here's some assembler (A86 style) to create the executable (I had to reverse engineer this as I'd misplaced the original source!)

 add dh,10h 
 push dx 
 add dh,10h 
 push dx 
 mov bl,80h 
 lea dx,[bx+2] 
 add bl,[bx] 
 mov [bx+1],al 
 mov ah,3dh 
 int 21h 
 pop ds 
 pop es 
 jb ret 
 mov bx,ax 
 mov ah,3fh 
 mov cx,di 
 xor dx,dx 
 int 21h 
 jb ret 
 mov bx,ax 
 xor ax,ax 
 repz stosw 
 xor di,di 
 xor si,si 
 inc ch 
program_loop:
 cmp si,bx 
 jnb ret 
 lodsb 
 mov bp,8 
 push program_loop
symbol_search: 
 dec bp 
 js ret
 cmp al,[bp+symbols]
 jnz symbol_search
 mov cl,[bp+instructions]
 jmp cx 
forward:
 inc di 
 ret 
increment:
 inc b es:[di] 
 ret 
decrement:
 dec b es:[di] 
 ret 
output:
 mov ah,2 
 mov dl,es:[di] 
 int 21h 
 ret 
input:
 mov ah,1 
 int 21h 
 stosb 
backward:
 dec di 
 ret 
jumpforwardifzero:
 cmp es:[di],dh 
 jnz ret 
 xor bp,bp
l1: cmp si,bx 
 jnb ret
 lodsb 
 cmp al,'[' 
 jnz l2
 inc bp
l2: cmp al,']' 
 jnz l1
 dec bp 
 jns l1
 ret 
jumpbackwardifnotzero:
 cmp es:[di],dh 
 jz ret
 xor bp,bp
l3: dec si 
 jz ret
 mov al,[si-1] 
 cmp al,']'
 jnz l4
 inc bp 
l4: cmp al,'[' 
 jnz l3
 dec bp 
 jns l3
 ret 
symbols:
 db '><+-.,[]'
instructions:
 db forward and 255
 db backward and 255
 db increment and 255
 db decrement and 255
 db output and 255
 db input and 255
 db jumpforwardifzero and 255
 db jumpbackwardifnotzero and 255

brainfuck, 843 691 bytes

Edit: decided to revisit this and found a surprising number of ways to golf off bytes

>>>,[>++++[-<-------->]<-[>+<<]>[----------[>]>[<+<+>>>>]<<<-[>]>[<+<+>>>>]<<<-[>]>[<+<+>>>>]<<<-[>]>[<-<+++>>>>]<<<--------------[>]>[<++<+>>>>]<<<--[>]>[<-<+++++++>>+>>]<<++++[-<------>]+<+[>]>[<++<+>>>>]<<<--[>]>[<<+>>>>]<<-<[+]<[>]>,>]<]<-[<]>[-[<<]>[<+[>]>>[<+[<<[<]<<-[>>]<[>>>>[>]>+<<[<]<]<-[>>]<[>>>>[>]>-<<[<]<]<++[->>+<<]>>[>]>]]<<<[<]>-<]>-[<<]>[<++[>]>+>[<-]<[<<[<]>[-<<+>>]>--[<<]>[[>]>+<<[<]<]>+[<<]>[[>]>-<<[<]<]>+[>]>]<<[<]>--<]>-[<<]>[[>]>>.<<<[<]<]>-[<<]>[[>]>>-<<<[<]<]>-[<<]>[[>]>>,<<<[<]<]>-[<<]>[[>]>>+<<<[<]<]>-[<<]>[[>]>>>[>>]>[<<<[<<]<+>>>[>>]>-]>[-]<<+[<[->>+<<]<]<[->>+<<]<[<]<]>-[<<]>[[>]>-[+>[-<<+>>]>]+<<[-]+[-<<]<[->>>[>>]>+<<<[<<]<]<[<]<]<++++++++>>[+<<->>]>]

This takes input in the form code!input where the !input is optional. It also simulates negative cells without using negative cells itself and can store up to (30000-(length of code+6))/2 cells.

Try it online!

Ruby 1.8.7, 188 185 149 147 characters

eval"a=[i=0]*3e4;"+$<.bytes.mapb*";"

Somewhat readable version:

code = "a = [0] * 3e4; i = 0;"
more_code ARGF.bytes.map .join(";")
eval code+more_code

As you see I shamelessly stole your idea of translating to the host language and then using eval to run it.

Binary Lambda Calculus 112

The program shown in the hex dump below

00000000 44 51 a1 01 84 55 d5 02 b7 70 30 22 ff 32 f0 00 |DQ...U...p0".2..|
00000010 bf f9 85 7f 5e e1 6f 95 7f 7d ee c0 e5 54 68 00 |....^.o..}...Th.|
00000020 58 55 fd fb e0 45 57 fd eb fb f0 b6 f0 2f d6 07 |XU...EW....../..|
00000030 e1 6f 73 d7 f1 14 bc c0 0b ff 2e 1f a1 6f 66 17 |.os..........of.|
00000040 e8 5b ef 2f cf ff 13 ff e1 ca 34 20 0a c8 d0 0b |.[./......4 ....|
00000050 99 ee 1f e5 ff 7f 5a 6a 1f ff 0f ff 87 9d 04 d0 |......Zj........|
00000060 ab 00 05 db 23 40 b7 3b 28 cc c0 b0 6c 0e 74 10 |....#@.;(...l.t.|
00000070

expects its input to consist of a Brainfuck program
(looking only at bits 0,1,4 to distinguish among ,-.+<>][ )
followed by a ], followed by the input for the Brainfuck program.

Save the above hex dump with xxd -r > bf.Blc

Grab a blc interpreter from https://tromp.github.io/cl/cl.html

cc -O2 -DM=0x100000 -m32 -std=c99 uni.c -o uni
echo -n "++++++++++[>+++++++>++++++++++>+++>+<<<<-]>++.>+.+++++++..+++.>++.<<+++++++++++++++.>.+++.------.--------.>+.>.]" > hw.bf
cat bf.Blc hw.bf | ./uni

Hello World!

Retina 0.8.2, 386 391 386 bytes

Code contains unprintable NUL (0x00) characters. It's also not super golfed yet, because it's already really slow, and if I golf it more, I don't know how long it'd take to finish. Appears to time out on the prime-finding sample.

There may be bugs in the online interpreter or in my program (leading new lines don't show in the output?).

Takes input like <code>│<input>. No, that is not a pipe (|). It's the Unicode character U+2502. The code also uses the Unicode characters ÿ▶◀├║. Unicode characters are used in order to support input of all ASCII characters. Therefore, these characters need to be separated from the code by a non-ASCII character.

Try it online

s`^.*
▶$0├║▶
s`(▶>.*║.*)▶(.)(.?)
$1$2▶$3
▶$
▶
║▶
║▶
(▶<.*║.*)(.)▶
$1▶$2
T`ÿ-`o`(?<=▶+.*║.*▶).
^+

T`-ÿ`ÿo`(?<=▶-.*║.*▶).
^-

(▶..*├.*)(║.*▶)(.)
$1$3$2$3
(▶,.*│)(.?)(.*├.*▶).
$1$3$2
▶[(.*║.*▶)
[▶▶$1
`(▶▶+)([^[]]*)[
$2[$1▶
`▶(▶+)([^[]]*)]
$2]$1
r`([[]]*)▶](.*║.*▶[^])
$1◀◀]$2
r`[([^[]]*)(◀+)◀
$2[$1
`]([^[]]*)(◀◀+)
$2◀]$1
◀
▶
`▶([^│])(.*║)
$1▶$2
s`.*├|║.*

Note there is a trailing newline there.

Brief Explanation:

Zeros 0x00 are used for the tape, which is infinite. The first replacement sets up the interpreter in the form ▶<code>│<input>├<output>║▶<tape>, where the first ▶ is the pointer for the code, and the second one is the pointer for the tape.

ÿ is 0xFF (255), which is used for Transliteration (used to implement + and -) to wrap the cells back around to zero.

◀ is only used for readability (in case the program is stopped in the middle or you want to see the program mid-execution). Otherwise, you couldn't tell which way the pointer was moving.

Commented Code:

s`^.* # Initialize
▶$0├║▶
s`(▶>.*║.*)▶(.)(.?) # >
$1$2▶$3
▶$
▶
║▶ # <
║▶
(▶<.*║.*)(.)▶
$1▶$2
T`ÿ-`o`(?<=▶+.*║.*▶). # +
^+

T`-ÿ`ÿo`(?<=▶-.*║.*▶). # -
^-

(▶..*├.*)(║.*▶)(.) # .
$1$3$2$3
(▶,.*│)(.?)(.*├.*▶). # ,
$1$3$2
▶[(.*║.*▶) # [
[▶▶$1
`(▶▶+)([^[]]*)[
$2[$1▶
`▶(▶+)([^[]]*)]
$2]$1
r`([[]]*)▶](.*║.*▶[^]) # ]
$1◀◀]$2
r`[([^[]]*)(◀+)◀
$2[$1
`]([^[]]*)(◀◀+)
$2◀]$1
◀
▶
`▶([^│])(.*║) # next instruction
$1▶$2
s`.*├|║.* # print output

Click here for the code with zeros in place of null bytes. Any occurrences of $0 should not be replaced with nulls.

Edit: Now supports empty input and suppresses trailing newline.

Infinite output is now supported. (403 bytes)

TI-BASIC, 264 bytes

Because of limitations in TI-BASIC, this actually doesn't qualify for this challenge since it breaks rule 2; the calculators' RAM is very limited, and doing something like 30000->dim(L1 (I use L1 for the stack/array) will force it to throw an ERR:MEMORY. As such, the stack/array starts at a size of 1 and grows if the pointer is pointing to an element past the end of it. It also breaks rule 3, because it's already breaking rule 2 so I may as well not bother with a cell-size limit.

Could probably still be golfed, by the way... I've made one or two edits to that end since first submitting, but if the version below doesn't work then head back to the edit from May 6 '15 and use that code instead. Also, as there's really no ASCII in TI-BASIC, this takes numbers of any size (and anything that returns a number, like a variable or expression) as input, and outputs numbers in turn.

Use SourceCoder to build it into a .8xp file then send it to your calculator with TI-Connect or TILP or something, and run it by including your brainfuck program in double quotes followed by a colon and whatever you named the TI-BASIC program. For instance, if you named it BRAINF, you'd run a program like this: "brainfuck goes here":prgmBRAINF.
If you have a shell on your calc that intercepts other commands when it detects the prgm token, though, do this: "brainfuck goes here" -> press ENTER -> prgmBRAINF.

seq(inString("<>-+.,[]",sub(Ans,S,1)),S,1,length(Ans->L2
cumSum((Ans=7)-(Ans=8->L3
seq(Ans(X),X,dim(Ans),1,~1->L4
1->P:DelVar L11->dim(L1 //this is the same as DelVar L1:1->dim(L1 as DelVar does not require a colon or newline after its argument
For(S,1,dim(L2
L2(S->T
P-(T=1)+(T=2->P
dim(L1
Ans+(P-Ans)(P>Ans->dim(L1
L1(P)-(T=3)+(T=4->L1(P
If T=5
Disp Ans
If T=6:Then
Input V
V->L1(P
End
If T=7 and not(L1(P
S+2+sum(not(cumSum(L3(S)-1=seq(L3(X),X,S+1,dim(L3->S
1-S+dim(L3
If T=8 and L1(P
S-sum(not(cumSum(L4(Ans)=seq(L4(X),X,Ans+1,dim(L4->S
End

If you don't have a way of connecting your calculator to your computer and want to type this out on-calc instead (I can't imagine why you'd want to, but I digress) note that -> is the STO> button above the ON key, ~ is the negative symbol next to ENTER, and to replace all instances of L<number> with the corresponding list token found on 2ND -> <number on keypad>

Thanks to thomas-kwa (at least, I think that's his Stack username) for helping me optimize this, especially with the [ and ] instructions.

Python 275 248 255

I decided to give it a try.

import sys
i=0
b=[0]*30000
t=''
for e in open(sys.argv[1]).read():
 t+=' '*i+['i+=1','i-=1','b[i]+=1','b[i]-=1','sys.stdout.write(chr(b[i]))','b[i]=ord(sys.stdin.read(1))','while b[i]:','pass','']['><+-.,['.find(e)]+'n'
 i+=(92-ord(e))*(e in'][')
exec t 

Haskell, 457 413 characters

import IO
import System
z=return
'>'#(c,(l,d:r))=z(d,(c:l,r))
'<'#(d,(c:l,r))=z(c,(l,d:r))
'+'#(c,m)=z(succ c,m)
'-'#(c,m)=z(pred c,m)
'.'#t@(c,_)=putChar c>>hFlush stdout>>z t
','#(_,m)=getChar>>=(c->z(c,m))
_#t=z t
_%t@('',_)=z t
i%t=i t>>=(i%)
b('[':r)=k$b r
b(']':r)=(z,r)
b(c:r)=f(c#)$b r
b[]=(z,[])
f j(i,r)=(t->j t>>=i,r)
k(i,r)=f(i%)$b r
main=getArgs>>=readFile.head>>=($('',("",repeat ''))).fst.b

This code "compiles" the BF program into an IO action of the form State -> IO State the state is a zipper on an infinite string.

Sad that I had to expend 29 characters to turn buffering off. Without those, it works, but you don't see the prompts before you have to type input. The compiler itself (b, f, and k) is just 99 characters, the runtime (# and %) is 216. The driver w/initial state another 32.

>ghc -O3 --make BF.hs 
[1 of 1] Compiling Main ( BF.hs, BF.o )
Linking BF ...

>./BF HELLO.BF 
Hello World!

>./BF PRIME.BF 
Primes up to: 100
2 3 5 7 11 13 17 19 23 29 31 37 41 43 47 53 59 61 67 71 73 79 83 89 97 

update 2011-02-15: Incorporated J B's suggestions, did a little renaming, and tightened up main

Conveyor, 953

This might be the most beautiful code you will ever see:

0

:I1@p
>#====)
^#====<
PP0
P<=======================<
00t:)01t1 a:P:P:P:P:P:P:^
>===========">">2>">2>">"^
^ +^-^5^ ^5^]^.^
^ "^"^*^"^*^"^"^
^ -^-^6^-^6^-^-^
^ #^#^*^#^*^#^#^
^ P P -^P )^P P
^ P P #^P )^P P
^t1)t0:))t01 P -^ 1
^===========< P #^ 0
^ t1(t0:))t01 P t
^=============< P )
^ t11(t01 0 0 )
^===============<. t P 10
^ FT#T#=<
^=================< P 
^ t11)t01 
^===================< 10t))0tP00t:(01t(1a:P:
^ >=====#=>==========">"
^ ^ ]^[
^ P ^ "^"
^===========================<=^#=====< -^-
 ^==< ^ PP#^#=
 ^===PTPT<
 ^ )P P
 ^=<=< (
 ^===<

C 284 362 (From a file)

#include <stdio.h>
char b[30000],z[9999],*p=b,c,*a,i;f(char*r,int s)while(c=*a++)if(!s)(c-62)?(c-60)?(c-43)?(c-45)?(c-46)?(c-44)?0:(*p=getchar()):putchar(*p):--*p:++*p:--p:++p;if(c==91)f(a,!*p);else if(c==93)if(!*p)return;else a=r;elseif(c==93)--s;if(!*p&&!s)return;else if(c==91)s++;main(int c,char**v)fread(z,1,9999,fopen(*++v,"r"));a=z;f(0,0);

Primes:

Primes up to: 100
2 3 5 7 11 13 17 19 23 29 31 37 41 43 47 53 59 61 67 71 73 79 83 89 97
Press any key to continue . . .

Compiled and ran successfully VS2008

Original solution failed to recognize loops that were initially set to zero. Still some room to golf. But finally solves the Prime Number program.

Ungolfed:

#include <stdio.h>
char b[30000],z[9999],*p=b,c,*a,i;
f(char*r,int s)

 while(c=*a++)
 
 if(!s)
 
 (c-62)?(c-60)?(c-43)?(c-45)?(c-46)?(c-44)?0:(*p=getchar()):putchar(*p):--*p:++*p:--p:++p;
 if(c==91)f(a,!*p);
 else if(c==93)if(!*p)return;else a=r;
 
 else
 
 if(c==93)
 
 --s;
 if(!*p&&!s)return;
 
 else if(c==91)
 
 s++;
 
 
 


main(int c,char**v)
 fread(z,1,9999,fopen(*++v,"r"));
 a=z;
 f(0,0);

Tests:

Hello World

Rot13

PHP 5.4, 296 294 273 263 261 209 191 183 178 166 characters:

I gave it a shot without using eval, but I eventually had to use it

<?$b=0;eval(strtr(`cat $argv[1]`,["]"=>'}',"["=>'while($$b)',"."=>'echo chr($$b);',","=>'$$b=fgetc(STDIN);',"+"=>'$$b++;',"-"=>'$$b--;',">"=>'$b++;',"<"=>'$b--;']));

All commands are working. This heavily abuses variable variables, and spews warnings. However, if one changes their php.ini to squelch warnings (or pipes stderr to /dev/null), this works great.

Verification (It's the "Hello World!" example from Wikipedia): http://codepad.viper-7.com/O9lYjl

Ungolfed, 367 365 335 296 267 characters:

<?php
$a[] = $b = 0;
$p = implode("",file($argv[1])); // Shorter than file_get_contents by one char
$m = array("]" => '', "[" => 'while($a[$b]){',"." => 'echo chr($a[$b]);', "," => '$a[$b]=fgetc(STDIN);', "+" => '$a[$b]++;', "-" => '$a[$b]--;', ">" => '$b++;', "<" => '$b--;');
$p = strtr($p,$m);
@eval($p);

This should be run via the command line: php bf.php hello.bf

Windows PowerShell, 204

'$c=,0*3e4;'+@62='$i++
';60='$i--
';43='$c[$i]++
';45='$c[$i]--
';44='$c[$i]=+[console]::ReadKey().keychar
';46='write-host -n([char]$c[$i])
';91='for(;$c[$i])';93=''[[int[]][char[]]"$(gc $args)"]|iex

Fairly straightforward conversion of the instructions and then Invoke-Expression.

History:

• 2011-02-13 22:24 (220) First attempt.


• 2011-02-13 22:25 (218) 3e4 is shorter than 30000.


• 2011-02-13 22:28 (216) Unnecessary line breaks. Matching on integers instead of characters is shorter.


• 2011-02-13 22:34 (207) Used indexes into a hash table instead of the switch.


• 2011-02-13 22:40 (205) Better cast to string removes two parentheses.


• 2011-02-13 22:42 (204) No need for a space after the argument to Write-Host.

C, 333 characters

This is my first BF interpreter and the first golf I actually had to debug.

This runs the prime number generator on Mac OS X/GCC, but an additional #include<string.h> may be necessary at a cost of 19 more characters if the implicit definition of strchr doesn't happen to work on another platform. Also, it assumes O_RDONLY == 0. Aside from that, leaving int out of the declaration of M saves 3 characters but that doesn't seem to be C99 compliant. Same with the third * in b().

This depends on the particulars of ASCII encoding. The Brainfuck operators are all complementary pairs separated by a distance of 2 in the ASCII code space. Each function in this program implements a pair of operators.

#include<unistd.h>
char C[30000],*c=C,o,P[9000],*p=P,*S[9999],**s=S,*O="=,-\",*t;
m()c+=o;
i()*c-=o;
w()o<0?*c=getchar():putchar(*c);
b()if(o>0)*c?p=*s:*--s;else if(*c)*++s=p;else while(*p++!=93)*p==91&&b();
int(*M[])()=m,i,w,b;
main(int N,char**V)
read(open(V[1],0),P,9e3);
while(o=*p++)
if(t=strchr(O,++o&~2))
o-=*t+1,
M[t-O]();

CJam, 75 bytes

lq3e4Vc*@"-<[],.+>"#"T1$T=(t T(:T; _T=g +(@T@t _T=o "_'(')er+S/=%s~@

Try it online: string reverser, Hello World.

Explanation

Takes code on the first line of STDIN and input on all lines below it.

l Read a line from STDIN (the program) and push it.
 q Read the rest of STDIN (the input) and push it.
 3e4Vc* Push a list of 30000 '' characters.
 @ Rotate the stack so the program is on top.

 % Apply this to each character in prog:
 "-<[],.+>"# Map '-' to 0, '<' to 1, ... and everything else to -1.
 ...= Push a magical list and index from it.

s~ Concatenate the results and evaluate the resulting string as CJam code.
 @ Rotate the top three elements again -- but there are only two, so the
 program terminates.

What about that magical list?

"T1$T=(t T(:T; _T=g +(@T@t _T=o " Space-separated CJam snippets.
 (Note the final space! We want an empty
 string at the end of the list.)
_'(')er+ Duplicate, change (s to )s, append.
 S/ Split over spaces.

The resulting list is as follows:

T1$T=(t (-)
T(:T; (<)
 ([)
_T=g (])
+(@T@t (,)
_T=o (.)
T1$T=)t (+)
T):T; (>)
 (unused)
_T=g (unused)
+(@T@t (unused)
_T=o (unused)
 (all other characters)

We generate the snippets for + and > from those for - and <, simply by changing left parens (CJam’s “decrement”) into right parens (CJam’s “increment”).

F#: 489 chars

The following program doesn't jump at '[' / ']' instructions, but scans the source code for the next matching token. This of course makes it kind of slow, but it can still find the primes under 100. F# integer types don't overflow but wrap.

Here's the short version:

[<EntryPoint>]
let M a=
 let A,B,i,p,w=Array.create 30000 0uy,[|yield!System.IO.File.ReadAllText a.[0]|],ref 0,ref 0,char>>printf"%c"
 let rec g n c f a b=if c then f i;if B.[!i]=a then g(n+1)c f a b elif B.[!i]=b then(if n>0 then g(n-1)c f a b)else g n c f a b
 while !i<B.Length do(let x=A.[!p]in match B.[!i]with|'>'->incr p|'<'->decr p|'+'->A.[!p]<-x+1uy|'-'->A.[!p]<-x-1uy|'.'->w x|','->A.[!p]<-byte<|stdin.Read()|'['->g 0(x=0uy)incr '['']'|']'->g 0(x>0uy)decr ']''['|_->());incr i
 0

A nasty gotcha was that the primes.bf program chokes on windows newlines. In order to run it I had to save the input number to a UNIX formatted text document and feed it to the program with a pipe:

interpret.exe prime.bf < number.txt

Edit: entering Alt+010 followed by Enter also works in Windows cmd.exe

Here's the longer version:

[<EntryPoint>]
let Main args =
 let memory = Array.create 30000 0uy
 let source = [| yield! System.IO.File.ReadAllText args.[0] |]
 let memoryPointer = ref 0
 let sourcePointer = ref 0
 let outputByte b = printf "%c" (char b)
 let rec scan numBraces mustScan adjustFunc pushToken popToken =
 if mustScan then
 adjustFunc sourcePointer
 if source.[!sourcePointer] = pushToken then
 scan (numBraces + 1) mustScan adjustFunc pushToken popToken
 elif source.[!sourcePointer] = popToken then
 if numBraces > 0 then scan (numBraces - 1) mustScan adjustFunc pushToken popToken
 else
 scan numBraces mustScan adjustFunc pushToken popToken 

 while !sourcePointer < source.Length do
 let currentValue = memory.[!memoryPointer]
 match source.[!sourcePointer] with
 | '>' -> incr memoryPointer
 | '<' -> decr memoryPointer
 | '+' -> memory.[!memoryPointer] <- currentValue + 1uy
 | '-' -> memory.[!memoryPointer] <- currentValue - 1uy
 | '.' -> outputByte currentValue
 | ',' -> memory.[!memoryPointer] <- byte <| stdin.Read()
 | '[' -> scan 0 (currentValue = 0uy) incr '[' ']'
 | ']' -> scan 0 (currentValue > 0uy) decr ']' '['
 | _ -> ()
 incr sourcePointer
 0 

Delphi, 397 382 378 371 366 364 328 characters

Eat this Delphi!

328 var p,d:PByte;f:File;z:Word=30000;x:Int8;begin p:=AllocMem(z+z);d:=p+z;Assign(F,ParamStr(1));Reset(F,1);BlockRead(F,p^,z);repeat z:=1;x:=p^;case x-43of 1:Read(PChar(d)^);3:Write(Char(d^));0,2:d^:=d^+44-x;17,19:d:=d+x-61;48,50:if(d^=0)=(x=91)then repeat p:=p+92-x;z:=z+Ord(p^=x)-Ord(p^=x xor 6);until z=0;end;Inc(p)until x=0;end.

Here the same code, indented and commented :

var
 d,p:PByte;
 x:Int8;
 f:File;
 z:Word=30000;
begin
 // Allocate 30000 bytes for the program and the same amount for the data :
 p:=AllocMem(z+z);
 d:=p+z;
 // Read the file (which path must be specified on the command line) :
 Assign(F,ParamStr(1));
 Reset(F,1);
 BlockRead(F,p^,z);
 // Handle all input, terminating at #0 (better than the spec requires) :
 repeat
 // Prevent a begin+end block by preparing beforehand (values are only usable in '[' and ']' cases) :
 z:=1; // Start stack at 1
 x:=p^; // Starting at '[' or ']'
 // Choose a handler for this token (the offset saves 1 character in later use) :
 case x-43of
 1:Read(PChar(d)^); // ',' : Read 1 character from input into data-pointer
 3:Write(Char(d^)); // '.' : Write 1 character from data-pointer to output
 0,2:d^:=d^+44-x; // '+','-' : Increase or decrease data
 17,19:d:=d+x-61; // '<','>' : Increase or decrease data pointer
 48,50: // '[',']' : Start or end program block, the most complex part :
 if(d^=0)=(x=91)then // When (data = 0 and forward), or when (data <> 0 and backward)
 repeat //
 p:=p+92-x; // Step program 1 byte back or forward
 z:=z+Ord(p^=x) // Increase stack counter when at another bracket
 -Ord(p^=x xor 6); // Decrease stack counter when at the mirror char
 until z=0; // Stop when stack reaches 0
 end;
 Inc(p)
 until x=0;
end.

This one took me a few hours, as it's not the kind of code I normally write, but enjoy!

Note : The prime test works, but doesn't stop at 100, because it reads #13 (CR) before #10 (LF)... do other submissions suffer this problem too when running on CRLF OSes?

C, 260 + 23 = 283 bytes

I created a C program which can be found here.

main(int a,char*s[])int b[atoi(s[2])],*z=b,p;char*c=s[1],v,w;while(p=1,
*c)q('>',++z)q('<',--z)q('+',++*z)q('-',--*z)q('.',putchar(*z))q(',',*z
=getchar())if(*c=='['

Has to be compiled via gcc -D"q(a,b)"="*c-a||(b);" -o pmmbf pmmbf.c and can be called as follows: pmmbf ",[.-]" 30000 whereby the first argument (quoted) contains the bf-program to run, the second determines how large the tape should be.

C, 267

#define J break;case
char*p,a[40000],*q=a;w(n)for(;*q-93;q++)if(n)switch(*q)J'>':++p;J'<':--p;J'+':++*p;J'-':--*p;J'.':putchar(*p);J',':*p=getchar();if(*q==91)char*r=*p&&n?q-1:0;q++;w(r);q=r?r:q;main(int n,char**v)p=a+read(open(v[1],0),a,9999);*p++=93;w(1);

Run as ./a.out primes.bf

Ungolfed Version:

#define J break;case

char*p,a[40000],*q=a; // packed so program immediately followed by data

w(n)
 for(;*q-93;q++) // until ']'
 if(n)switch(*q) // n = flagged whether loop evaluate or skip(0)
 J'>':++p;
 J'<':--p;
 J'+':++*p;
 J'-':--*p;
 J'.':putchar(*p);
 J',':*p=getchar();
 
 if(*q==91)char*r=*p&&n?q-1:0;q++;w(r);q=r?r:q; // recurse on '[', record loop start
 


main(int n,char**v)
 p=a+read(open(v[1],0),a,9999);
 *p++=93; // mark EOF with extra ']' and set data pointer to next
 w(1); // begin as a loop evaluate

Python 2, 223

I admit that I recycled an old program of mine (but had to change it quite a bit, because the old version didn't have input, but error checking...).

P="";i,a=0,[0]*30000
import os,sys
for c in open(sys.argv[1]).read():x="><+-.[,]".find(c);P+=" "*i+"i+=1 i-=1 a[i]+=1 a[i]-=1 os.write(1,chr(a[i])) while+a[i]: a[i]=ord(os.read(0,1)) 0".split()[x]+"n";i+=(x>4)*(6-x)
exec P

Runs the primes calculator fine.

I see now that Alexandru has an answer that has some similarities. I'll post mny answer anyways, because I think there are some new ideas in it.

C, 374 368

Reads from a file. Passes PRIME.BF test.

Usage: ./a.out PRIME.BF

#include <stdio.h>
main(int c,char**v)int m[30000],s[99],p=0,i=0,n=0;char l[9999],d;FILE*f=fopen(v[1],"r");for(l[i]=0;i<9999&&l[i]!=EOF;l[i]=getc(f))i++;for(i=1;d=l[i];i++)if(!n)p+=d-62?0:1;p-=d-60?0:1;m[p]+=d-43?0:1;m[p]-=d-45?0:1;if(d==46)putchar(m[p]);if(d==44)m[p]=getchar();if(d==93)i=s[c]-1;c--;n++;if(d==91)if(m[p])c++;s[c]=i;elsen++;n-=d-93?0:1;

Reformatted:

#include <stdio.h>
main(int c,char**v)
 int m[3000],s[99],p=0,i=0,n=0;
 char l[9999],d;
 FILE*f=fopen(v[1],"r");
 for(l[i]=0;i<9999&&l[i]!=EOF;l[i]=getc(f))i++;
 for(i=1;d=l[i];i++)
 if(!n) // > < + - . , ] n [ ]
 p+=d-62?0:1;
 p-=d-60?0:1;
 m[p]+=d-43?0:1;
 m[p]-=d-45?0:1;
 if(d==46)putchar(m[p]);
 if(d==44)m[p]=getchar();
 if(d==93)i=s[c]-1;c--;n++;
 
 if(d==91)if(m[p])c++;s[c]=i;elsen++;
 n-=d-93?0:1;
 

Lua, 285

loadstring("m,p=0,1 "..io.open(arg[1]):read"*a":gsub("[^.,<>[%]+-]",""):gsub(".",["."]="io.write(string.char(@)) ",[","]="@=io.read(1):byte() ",["<"]="p=p-1 ",[">"]="p=p+1 @=@or 0 ",["["]="while @~=0 do ",["]"]="end ",["+"]="@=(@+1)%256 ",["-"]="@=(@-1)%256 "):gsub("@","m[p]"))()

Somewhat readable version:

loadstring( --execute
 "m,p=0,1 ".. --initialize memory and pointer
 io.open(arg[1]) --open file
 :read"*a" --read all
 :gsub("[^.,<>[%]+-]","") --strip non-brainfuck
 :gsub(".", --for each character left
 ["."]="io.write(string.char(@)) ", -- '@' is shortcut for 'm[p]', see below
 [","]="@=io.read(1):byte() ",
 ["<"]="p=p-1 ",
 [">"]="p=p+1 @=@or 0 ", --if a before unexplored memory cell, set to 0
 ["["]="while @~=0 do ",
 ["]"]="end ",
 ["+"]="@=(@+1)%256 ", --i like it overflowing
 ["-"]="@=(@-1)%256 "
 
 )
 :gsub("@","m[p]") --replace the '@' shortcut
 ) --loadstring returns a function
() --call it

Works perfectly

Lua, 478, w/o loadstring

local m,p,i,r,c=0,1,1,,io.open(arg[1]):read"*a"while i<=#c do(([43]=function()m[p]=(m[p]+1)%256 end,[45]=function()m[p]=(m[p]-1)%256 end,[62]=function()p=p+1 m[p]=m[p]or 0 end,[60]=function()p=p-1 end,[46]=function()io.write(string.char(m[p]))end,[44]=function()m[p]=io.read(1):byte()end,[91]=function()if m[p]==0 then i=select(2,c:find("%b[]",i))else r[#r+1]=i end end,[93]=function()if m[p]==0 then r[#r]=nil else i=r[#r] end end)[c:byte(i)]or function()end)()i=i+1 end

Readable version:

local m, p, i, r, c= --memory, pointer, brackets stack, code
 0, 1, 1, , io.open(arg[1]) --open file
 :read"*a" --read it
while i<=#c do --while there's code
 (
 (
 
 [43]=function() -- +
 m[p]=(m[p]+1)%256
 end,
 [45]=function() -- -
 m[p]=(m[p]-1)%256
 end,
 [62]=function() -- >
 p=p+1 m[p]=m[p]or 0 --if new memory cell, set it to 0
 end,
 [60]=function() -- <
 p=p-1
 end,
 [46]=function() -- .
 io.write(string.char(m[p]))
 end,
 [44]=function() -- ,
 m[p]=io.read(1):byte()
 end,
 [91]=function() -- [
 if m[p]==0 then
 i=select(2,c:find("%b[]",i)) --find matching ]
 else
 r[#r+1]=i --push position to the stack
 end
 end,
 [93]=function() -- ]
 if m[p]==0 then
 r[#r]=nil --pop from stack
 else
 i=r[#r] --go to position on the top of stack
 end
 end
 
 )[c:byte(i)] --transform character into code
 or function()end --do nothing on non-brainfuck
 )() --run the resulting function
 i=i+1 --go to the next opcode
end

Brainfuck, 948 bytes

Well, that took a while. I golfed a Brainfuck self-interpreter by ... not me.

->->>>-[,+>+<[->-]>[->]<+<-------------------------------------[+++++++++++++++++++++++++++++++++++++>-]>[->]<<[>++++++++[-<----->]<---[-[-[-[--------------[--[>+++++++[-<---->]<-[--[[+]->]<+[->++>]->]<+[->+>]->]<+[->+++++>]->]<+[->++++++>]->]<+[->+++++++>]->]<+[->++++>]->]<+[->++++++++>]->]<+[->+++>]->]+<+[->->]>[-<->]<]>>->>-<<<<<+++[<]>[-[-[-[-[-[-[-[-<<++++++++>>>[>]>>>>+[->>+]->,<<<+[-<<+]-<<<[<]<]>[<<<+++++++>>>[>]>>>>+[->>+]->.<<<+[-<<+]-<<<[<]]<]>[<<<++++++>>>[>]>>>>+[->>+]<<-<<+[-<<+]-<<<[<]]<]>[<<<+++++>>>[>]>>>>+[->>+]+>>-<<[-<<+]-<<<[<]]<]>[<<<++++>>>[>]>>>>+[->>+]->-<<<+[-<<+]-<<<[<]]<]>[<<<+++>>>[>]>>>>+[->>+]->+<<<+[-<<+]-<<<[<]]<]>[<++[>]>>>>+[->>+]->[<<<+[-<<+]-<<<[<]-[<<-[>->-[<+]]<+[->>[<]]<-[>-->+[<++]]<++[-->>[<]]<++>>[[-<+>]<<[->>+<<]]<[>]>]]<[<<+[-<<+]-<<<[<]>--<<++>]>]<]>[<<<+>>>[>]>>>>+[->>+]->[<<<+[-<<+]-<<<[<]]<[<<+[-<<+]-<<<[<]+[>-[<-<]<<[>>]>>-[<+<]<<[>>]>>++<[>[-<<+>>]<[->+<]]<[>]>]]>[[-<<+>>]<[->+<]>]]>]

Recall, 594 bytes

In short: Recall has no arithmetic operators in a classic sense, it only has bitwise operations. You can not just "add one" etc. Recall is also strictly stack-based.

DC505M22022M32032M606M42042M707M92092M4405022o032o06o042o07o092o044o1305022o06o042o092o52052q.q2305022o06o07o93093q.q5403206o07o14014q.q6403206o042o07o24024q.q74Yx34034z03MMMMMMMM034o3yY030401r3.4.101zyY040301r4.3.101zY01052gZ02Z040301052023s4.3.10zyY01023gZ02z030401023052s3.4.10zyY01093gZ02q20zyY01054gZ02u20zyY01014gZx20zyY01064gZ02X0zyY01024gZ03304302r33.43.20zyY01074gZ04303302r43.33.20zyyQ6205.8Y06208g6206208iZ08M808013izy062U7205.9Y07209g7207209iz09M909013izy072R53.63.82063MMMMMMMM053o63082013i53082KKKKKKKK82053063082S84.94.12.73.83t012073083TY083073012r83.73.12012084gzY012094gZt0zyy

Example 1: Print something

Input:

-[--->+<]>-----..-[----->+<]>.++++.+[->++++<]>.---[----->++<]>.---.------------.++++++++.++++++++.+[-->+++++<]>-.

Output:

PPCG rocks!

Example 2: Output square numbers up to 100

Input:

+[>++<-]>[<+++++>-]+<+[>[>+>+<<-]++>>[<<+>>-]>>>[-]++>[-]+>>>+[[-]++++++>>>]<<<[[<++++++++<++>>-]+<.<[>----<-]<]<<[>>>>>[>>>[-]+++++++++<[>-<-]+++++++++>[-[<->-]+[<<<]]<[>+<-]>]<<-]<<-]

Output:

0
1
4
9
16
25
36
49
64
81
100

This example might take a few minuted to execute and might cause a "this tab is frozen" message. Ignore that and wait.

OCaml(lex), 497 chars

OCamllex is part of the standard distribution of OCaml.

let a=Array.create 30000 0
let(%)f g h=f(g h)
let s v i=a.(i)<-v;i
let o d i=s(a.(i)+d)i
let p i=print_char(Char.chr a.(i));flush stdout;i
let r i=s(Char.code(input_char stdin))i
let rec w g i=if 0=a.(i)then i else w g(g i)
let n x=x
rule t f=parse
|'>'t(succ%f)lexbuf
|'<'t(pred%f)lexbuf
|'+'t((o 1)%f)lexbuf
|'-'t((o(-1))%f)lexbuf
|'.'t(p%f)lexbuf
|','t(r%f)lexbuf
|'['t((w(t n lexbuf))%f)lexbuf
|']'|eoff
|_t f lexbuf
let _=t n(Lexing.from_channel(open_in Sys.argv.(1)))0

Save as b.mll and run with

ocamllex b.mll && ocaml b.ml prime.bf

I don't like parsing by hand, so I used the provided lexer generator. From the tokens read, we compose a function for the whole brainf*ck program.

C# (2861 char, ~84 lines)

This is not the prettiest solution to the problem, and probably not all that 'Golf-ish', since I wasn't as concerned with length as I probably should have been. (I didn't remove the comments or extra white space.) I just wanted to try something in a new language, to see if I could. If I did it again, I'd drop the use of the stack for returning from ']' and just look back. Run without command line arguments it runs the hello world program given in the problem description. It accepts one command line argument, the filename of the program to run.

using System;
using System.Collections.Generic;

namespace ConsoleApplication1

 class Program
 
 static void Main(string[] args)
 
 String ProgSource;
 if (args.Length > 0)
 ProgSource = System.IO.File.ReadAllText(args[0]);
 else //hello world
 ProgSource = "";

 Stack<int> stack = new Stack<int>();
 char[] bfProg = ProgSource.ToCharArray();
 char[] mem = new char[30000];
 int ptr = 0;

 for (int ip = 0; ip<bfProg.Length; ip++)
 switch (bfProg[ip])
 
 case ('>'): ptr++; break;
 case ('<'): ptr--; break;
 case ('+'): mem[ptr]++; break;
 case ('-'): mem[ptr]--; break;
 case ('.'): Console.Write(mem[ptr]); break;
 case (','): 
 char key = Console.ReadKey(false).KeyChar;
 if (key == 'r')
 
 key = (char)10;
 Console.WriteLine();
 
 mem[ptr] = key;
 break;
 case ('['):
 if (mem[ptr] == 0)
 
 int openBraces = 1;
 //find the closing brace for this expression
 for (int x = 1; x < (bfProg.Length - ip); x++)
 
 if (bfProg[ip + x] == ']') openBraces--;
 if (bfProg[ip + x] == '[') openBraces++;
 if (openBraces == 0)
 
 if (stack.Peek() == ip) stack.Pop();
 ip += x;
 break;
 
 
 
 else
 
 stack.Push(ip);
 
 break;
 case (']'):
 if (mem[ptr] == 0)
 stack.Pop();
 else
 
 ip = stack.Peek();
 
 break;
 
 

 Console.WriteLine("nnnExecution Completed Sucessfully. Press any key to continue...");
 Console.ReadKey();

 
 

Edit: Removed unused references.

C (gcc), 273 268 bytes

main(_,a)_=fopen("w.c","w");fputs("main()char a[30000],*p=a;",_);x:a=getchar();fputs(a-62?a-60?a-43?a-45?a-46?a-44?a-91?a-93?~a?"":"":"":"while(*p)":"*p=getchar();":"putchar(*p);":"--*p;":"++*p;":"--p;":"++p;",_);if(~a)goto x;fclose(_);system("cc w.c;./a.out");;

Try it online!

-5 thanks to ceilingcat

Takes input from stdin.

This relies a little bit on the environment, but is pretty consistent. This is effectively the eval solution for c. It writes an appropriate C program to the file w.c, compiles it, and runs it as the desired executable. Thus as a bonus effect this actually compiles the bf code and leaves a.out as a binary for it. Note that depending on the system you may need to modify the last string. In particular most windows c compilers call the default executable "a.exe". Luckily as far as I can tell, they all have the same length so the bytecount is the same. (though if you don't have a cc defined you may need to add a letter such as gcc to the compile command, adding 1 byte).

I am aware that this thread is a bit old, but I didn't see this style of C solution yet, so I thought I'd add it.

C (gcc) Linux x86_64, 884 621 525 487 439 383 372 bytes

*mmap();d[7500];(*p)();*j(a,g,q)char*a;char*t=a,*n,*c;for(q=0;read(g,&c,c=!q);)t=c==91?n=j(t+9,g),c=n-t-9,bcopy(&c,mempcpy(t,"x80>xfx84",5),4),n:c==93?a-=14+t,q=*t=233,mempcpy(t+1,&a,4):stpcpy(t,c-62?c-60?c-43?c-45?c-46?c-44?"":"1xc0P_xF5":"RXR_xF5":L"໾":L"۾":L"컿":L"웿");return t;main(P,g)int**g;p=mmap(0,1<<20,6,34,0,0);p(*j(p,open(g[1],0))=195,d,1);

Try it online!

This is a JIT that compiles BF code into x86_64 machine language at runtime. This performs a straight translation so commonly occurring sequences such as >>>, <<<, +++ and --- aren't coalesced into faster instructions.

Less golfed version:

// size of data area
*mmap();d[7500];(*p)();
// recursive function translates BF commands to x86_64 instructions
*j(a,g,q)char*a;
 char*t=a,*n,*c;
 for(q=0;read(g,&c,c=!q);)
 t=c==91? // [
 // cmpb $0x0,(%rsi)
 // je N
 n=j(t+9,g),c=n-t-9,bcopy(&c,mempcpy(t,"x80>xfx84",5),4),n
 :
 c==93? // ]
 // jmp a
 a-=14+t,q=*t=233,mempcpy(t+1,&a,4)
 :
 stpcpy(t,c-62? // >
 c-60? // <
 c-43? // +
 c-45? // -
 c-46? // .
 c-44? // ,
 ""
 :
 // xor %eax,%eax
 // push %rax
 // pop %rdi
 // syscall
 "1xc0P_xF5"
 :
 // push %rdx
 // pop %rax
 // push %rdx
 // pop %rdi
 // syscall
 "RXR_xF5"
 :
 // decb (%rsi)
 L"໾"
 :
 // incb (%rsi)
 L"۾"
 :
 // dec %esi
 L"컿"
 :
 // inc %esi
 L"웿");
 return t;

main(P,g)int**g;
 // allocate text (executable) memory and mark as executable
 p=mmap(0,1<<20,6,34,0,0);
 // run JIT, set %rdx=1 and call code like a function
 p(*j(p,open(g[1],0))=195,d,1);

[EDIT]

C++11, 355, reads from file:

#include<functional>
#include<stdio.h>
main()
char b[30000],g[9999],*f=g,*p=b,n[]="+-,.><[]",j;
std::function<void()>m[]=
[&p](*p)++;,
[&p](*p)--;,
[&p]*p=getchar();,
[&p]putchar(*p);,
[&p]p++;,
[&p]p--;,
[&p,&f]if(!(*p))while(*f-93)f++;,
[&f,&m]while(*f-91)f--;m[6]();
;
fread(g,1,9999,fopen(a[1],0));
for(;*f;f++)for(j=0;n[j];j++)if(n[j]==*f)m[j]();

Test

http://ideone.com/b7vO4

[OLD VERSION]

C++11, 391, to see running: http://ideone.com/yZHVv

#include<functional>
#include<stdio.h>
main(int c,char **a) c==']'))m[c]();