Can you post a short example of real, overdone spaghetti code, possibly saying what it does? Can you show me a little debugger's nightmare?
I don't mean IOCCC code, that is science fiction. I mean real life examples that happened to you...
The focus has changed from "post some spaghetti code" to "what is exactly spaghetti code?". From a historical perspective, the current choices seem to be:
Real spaghetti-code requires a multitude of non-local gotos. Sadly this is not possible using most modern languages.
Edit: Some suggest exceptions and longjmp as substitutes for GOTO. But these are far to limited and structured, since they only allow you to return up the callstack. Real GOTO allows you to jump to any line anywhere in the program, which is necessary to create real spaghetti.
You've asked for it, you'll get it:
This is the source of a DOS .com file that plays the Blue Danube waltz. The executable is just 176 bytes in size. Code is re-used as data and vice versa.
.286
.model tiny
g4 equ 55-48 ; removed note-decoding !
a4 equ 57-48 ; now: storing midi-notes for octaves 0..2 and convert
h4 equ 59-48 ; to 4..6 with a simple add 48.
c5 equ 60-48
d5 equ 62-48
e5 equ 64-48
g5 equ 67-48
h5 equ 71-48
c6 equ 72-48
d6 equ 74-48
e6 equ 76-48
g6 equ 79-48 ; = 00011111b
pp equ 0 ; c4 is not used in the walz, using it as play-pause.
EOM equ 1 ; c#4 is also available... End Of Music
; warning: experts only beyond this point !
pau1 equ 00100000b ; bitfield definitions for note-compression
pau2 equ 01000000b ; you can or a pau to each note!
pau3 equ 01100000b
;rep1 equ 01000000b ; rep1 is history (only used once).
;rep3 equ 11000000b ; rep3 was never used.
rep2 equ 10000000b ; or a rep2 to a note to play it 3 times.
drumsize equ 5
.code
org 100h
start:
mov ah,9
mov dx,offset msg
int 21h ; print our headerstring
mov dx,0330h ; gus midi megaem -port
mov si,offset music_code ; start of music data
mainloop:
; get new note (melody)
xor bp,bp ; bp= repeat-counter
lodsb ; get a new note
cmp al, EOM ; check for end
jne continue
ret
continue:
jns no_rep2 ; check for rep2-Bit
inc bp
inc bp ; "build" repeat-counter
no_rep2:
push ax ; save the note for pause
; "convert" to midi-note
and al,00011111b
jz skip_pp ; check pp, keep it 0
add al,48 ; fix-up oktave
skip_pp:
xchg ax,bx ; bl= midi-note
play_again:
mov cl,3
push cx ; patch program (3= piano)
push 0c8h ; program change, channel 9
; wait (cx:dx) times
mov ah,86h ; wait a little bit
int 15h
; prepare drums
dec di ; get the current drum
jns no_drum_underflow
mov di,drumsize
no_drum_underflow:
; play drum
push dx ; volume drum
push [word ptr drumtrk+di] ; note drum
mov al,99h
push ax ; play channel 10
; play melody
push dx ; volume melody
push bx ; note melody
dec ax ; replaces dec al :)
push ax ; play channel 9
; send data to midi-port
mov cl,8 ; we have to send 8 bytes
play_loop:
pop ax ; get the midi event
out dx,al ; and send it
loop play_loop
; repeat "bp" times
dec bp ; repeat the note
jns play_again
; check and "play" pause
xor bx,bx ; clear the note, so we can hear
; a pause
; decode pause value
pop ax
test al,01100000b
jz mainloop ; no pause, get next note
; decrement pause value and save on stack
sub al,20h
push ax
jmp play_again ; and play next drum
; don't change the order of the following data, it is heavily crosslinked !
music_code db pp or rep2
db g4 or rep2 or pau1
db h4 or pau1, d5 or pau1, d5 or pau3
db d6 or pau1, d6 or pau3, h5 or pau1, h5 or pau3
db g4 or rep2 or pau1
db h4 or pau1, d5 or pau1, d5 or pau3
db d6 or pau1, d6 or pau3, c6 or pau1, c6 or pau3
db a4 or rep2 or pau1
db c5 or pau1, e5 or pau1, e5 or pau3
db e6 or pau1, e6 or pau3, c6 or pau1, c6 or pau3
db a4 or rep2 or pau1
db c5 or pau1, e5 or pau1, e5 or pau3
db e6 or pau1, e6 or pau3, h5 or pau1, h5 or pau3
db g4 or rep2 or pau1
db h4 or pau1, g5 or pau1, g5 or pau3
db g6 or pau1, g6 or pau3, d6 or pau1, d6 or pau3
db g4 or rep2 or pau1
db h4 or pau1, g5 or pau1, g5 or pau3
db g6 or pau1, g6 or pau3, e6 or pau1, e6 or pau3
db a4 or rep2 or pau1
db c5 or pau1, e5 or pau1, e5 or pau3, pp or pau3
db c5 or pau1, e5 or pau1, h5 or pau3, pp or pau3, d5 or pau1
db h4 or pau1, h4 or pau3
db a4 or pau1, e5 or pau3
db d5 or pau1, g4 or pau2
; db g4 or rep1 or pau1
; replace this last "rep1"-note with two (equal-sounding) notes
db g4
db g4 or pau1
msg db EOM, 'Docking Station',10,'doj&sub'
drumtrk db 36, 42, 38, 42, 38, 59 ; reversed order to save some bytes !
end start
From a Linux SCSI driver (which shall remain nameless to protect the guilty):
wait_nomsg:
if ((inb(tmport) & 0x04) != 0) {
goto wait_nomsg;
}
outb(1, 0x80);
udelay(100);
for (n = 0; n < 0x30000; n++) {
if ((inb(tmport) & 0x80) != 0) { /* bsy ? */
goto wait_io;
}
}
goto TCM_SYNC;
wait_io:
for (n = 0; n < 0x30000; n++) {
if ((inb(tmport) & 0x81) == 0x0081) {
goto wait_io1;
}
}
goto TCM_SYNC;
wait_io1:
inb(0x80);
val |= 0x8003; /* io,cd,db7 */
outw(val, tmport);
inb(0x80);
val &= 0x00bf; /* no sel */
outw(val, tmport);
outb(2, 0x80);
TCM_SYNC:
/* ... */
small_id:
m = 1;
m <<= k;
if ((m & assignid_map) == 0) {
goto G2Q_QUIN;
}
if (k > 0) {
k--;
goto small_id;
}
G2Q5: /* srch from max acceptable ID# */
k = i; /* max acceptable ID# */
G2Q_LP:
m = 1;
m <<= k;
if ((m & assignid_map) == 0) {
goto G2Q_QUIN;
}
if (k > 0) {
k--;
goto G2Q_LP;
}
G2Q_QUIN: /* k=binID#, */
How did I locate this gem?
find /usr/src/linux -type f -name \*.c |
while read f
do
echo -n "$f "
sed -n 's/^.*goto *\([^;]*\);.*/\1/p' $f | sort -u | wc -l
done |
sort +1rn |
head
The output is a series of lines listing files ordered by the number of gotos to distinct labels, like the following:
kernel/fork.c 31
fs/namei.c 35
drivers/infiniband/hw/mthca/mthca_main.c 36
fs/cifs/cifssmb.c 45
fs/ntfs/super.c 47
Real spaghetti code was done in COBOL and used the ALTER statement.
Here's an example. Here's another, while listed a "humor", I've seen this kind of thing. Almost got fired once for noting that any program with an Alter statement was obviously in a state of sin. I refused to "maintain" that program, it was quicker to replace it than understand it.
Spaghetti code: Originating in the early 60's in Italy as an alternate recipe for certain pasta dishes, spaghetti code was cooked up by one restaurant entrepreneur who attempted to automate the creation of a fool-proof entree. Pressed by the lack of time to complete the design the engineer/chef cut corners which introduced problems in the recipe early on. In a frantic attempt to remedy a good idea gone bad, various spices were quickly added to the concoction as the recipe grew out of control. The result was a stringy, twisty, yet potentially tasty pile of text that would later grow to be a practice cherished by developers world-wide.
This is from a MIDI parser I wrote some time ago. It was a quick and dirty proof of concept, but nevertheless, I shall take the blame for its ugliness: 4 levels of nested conditionals plus the dreaded multiple returns. This code was meant to compare 2 MIDI events in order to sort them by priority when writing to a file. Ugly as it was, it did the job decently, though.
internal class EventContainerComparer : IComparer {
int IComparer.Compare(object a, object b) {
MIDIEventContainer evt1 = (MIDIEventContainer) a;
MIDIEventContainer evt2 = (MIDIEventContainer) b;
ChannelEvent chanEvt1;
ChannelEvent chanEvt2;
if (evt1.AbsoluteTime < evt2.AbsoluteTime) {
return -1;
} else if (evt1.AbsoluteTime > evt2.AbsoluteTime) {
return 1;
} else {
// a iguar valor de AbsoluteTime, los channelEvent tienen prioridad
if(evt1.MidiEvent is ChannelEvent && evt2.MidiEvent is MetaEvent) {
return -1;
} else if(evt1.MidiEvent is MetaEvent && evt2.MidiEvent is ChannelEvent){
return 1;
// si ambos son channelEvent, dar prioridad a NoteOn == 0 sobre NoteOn > 0
} else if(evt1.MidiEvent is ChannelEvent && evt2.MidiEvent is ChannelEvent) {
chanEvt1 = (ChannelEvent) evt1.MidiEvent;
chanEvt2 = (ChannelEvent) evt2.MidiEvent;
// si ambos son NoteOn
if( chanEvt1.EventType == ChannelEventType.NoteOn
&& chanEvt2.EventType == ChannelEventType.NoteOn){
// chanEvt1 en NoteOn(0) y el 2 es NoteOn(>0)
if(chanEvt1.Arg1 == 0 && chanEvt2.Arg1 > 0) {
return -1;
// chanEvt1 en NoteOn(0) y el 2 es NoteOn(>0)
} else if(chanEvt2.Arg1 == 0 && chanEvt1.Arg1 > 0) {
return 1;
} else {
return 0;
}
// son 2 ChannelEvent, pero no son los 2 NoteOn, el orden es indistinto
} else {
return 0;
}
// son 2 MetaEvent, el orden es indistinto
} else {
return 0;
}
}
}
}
Here is the Duff's Device, from Matt's answer to this question:
int n = (count + 7) / 8;
switch (count % 8) {
case 0: do { *to = *from++;
case 7: *to = *from++;
case 6: *to = *from++;
case 5: *to = *from++;
case 4: *to = *from++;
case 3: *to = *from++;
case 2: *to = *from++;
case 1: *to = *from++;
} while (--n > 0);
}
To me, a more modern example of spaghetti code is when you have 20 dlls and every DLL references each other in one way or another. Your dependency graph looks like a huge blob, and your code hops all over the place with no real order. Everything is inter-dependent.
There's also Ravioli Code, which is the opposite. Nice little chunks of functionality, a clean interface neatly wrapped around meaty goodness, all sat in a nice sauce of framework.
Don't forget to mention Object-oriented spaghetti. This is when you try to use all the design patterns in the book, even when they don't make sense. This leads to spaghetti code at conceptual level, which is far more detrimental to quality than classical goto-based spaghetti code.
Have you ever looked at code generated by Flex/Bison scanner and generator? A plethora of labels and preprocessor directives.
It's absolutely impossible to understand what's inside.. and absolutely impossible to follow flow of the program.
That's definetely spaghetti code.
I'm not pulling this out of my head. This is what I have had to work with, albeit simplified. Let's say that basically you have a program that needs an enum:
enum {
a, b, c;
} myenum;
But instead what we have is
HashTable t;
t["a"] = 0;
t["b"] = 1;
t["c"] = 2;
But of course, no implementation of a hash table is good enough so there is a local implementation of hash tables, which contains about 10 times more code than an average open source implementation with half the features and double the number of bugs. The HashTable is actually defined virtual, and there's a factory HashTableFactory to create instances of HashTables, but true to the pattern HashTableFactory is also virtual. To prevent an infite cascade of virtual classes there's a function
HashTableFactory *makeHashTableFactor();
Thus everywhere where the code needs myenum's it carries a reference to the instance of a HashTable and HashTableFactory, in case you want to make more HashTable's. But wait, that's not all! This is not how the hash table is initialized, but it's done by writing a code that reads XML:
<enum>
<item name="a" value="0"/>
<item name="b" value="1"/>
<item name="c" value="2"/>
</enum>
and inserts into a hash table. But the code is "optimised" so that it doesn't read an ascii file myenum.xml, but instead there's a compile time script which generates:
const char* myenumXML = [13, 32, 53 ....];
from myenum.xml and the hash table is initialized by a function:
void xmlToHashTable(char *xml, HashTable *h, HashTableFactory *f);
which is called:
HashTableFactory *factory = makeHashTableFactory();
HashTable *t = facotry.make();
xmlToHashTable(myenumXML, t, f);
Ok, so we have a lot of code to get an enum structure. It's basically used in a function:
void printStuff(int c) {
switch (c) {
case a: print("a");
case b: print("b");
case c: print("c");
}
}
and this is called in a context where:
void stuff(char* str) {
int c = charToEnum(str);
printStuff(c);
}
So what we actually have is instead of
void stuff(char *str) {
printf(str);
}
we have manged to generate thousands of lines of code (private new, buggy, complex, implementation of hashtables, and xml readers, and writer) in place of the above 3.