2010-01-11, 14:26(+00), Stephane CHAZELAS:
> 2010-01-9, 06:23(-06), den...@netstrata.com:
> [...]
>> This produces the correct distribution of dice values for two six-sided dice:
>>
>> $ unset dice; for i in {1..10000}; do ((dice[$RANDOM%6+1 +
>> $RANDOM%6+1]++)); done; echo "${di...@]}"
>> 290 582 837 1130 1375 1635 1315 1126 845 574 291
>>
>> The total is correct:
>>
>> $ unset t; for i in ${di...@]}; do ((t+=i)); done; echo $t
>> 10000
>>
>> This creates an even distribution which is incorrect:
>>
>> $ unset dice; for i in {1..10000}; do ((dice[RANDOM%6+1 + RANDOM%6+1]++));
>> done; echo "${di...@]}"
>> 886 884 887 882 885 879 886 887 881 879 883
>>
>> And the total is incorrect (may be larger or smaller):
>>
>> $ unset t; for i in ${di...@]}; do ((t+=i)); done; echo $t
>> 10047
> [...]
>
> I've been scratching my head for some time, but can't figure out
> what's going on. I get the exact same behavior with ksh93 and
> zsh. Again there, replacing RANDOM with $RANDOM fixes the
> problem. Strange that all 3 shells would have the exact same
> bug. Are we missing the obvious here?
Oh I get it
dice[RANDOM%6+1 + RANDOM%6+1]++
is processed as
dice[RANDOM%6+1 + RANDOM%6+1] = dice[RANDOM%6+1 + RANDOM%6+1] + 1
with RANDOM expanded to different values 4 times.
which explains both the evenness and incorrect sum
we get similar problems with:
$ bash -c 'x=0; ((a[++x]++)); echo $x'
2
The +=, -=, *=... are also affected.
--
Stéphane
