array size vs index of last element (was Re: setarray[unsetkey] doesn't trigger nounset in arithmetic context)

[Martin D Kealey]

On Tue, 23 May 2023 at 23:32, Chet Ramey <chet.ramey@case.edu> wrote:

> On 5/22/23 10:56 PM, Martin D Kealey wrote:
>
> > For example, if one is filling an array in random order, rather than
> > progressively adding to the end, then it is useful and makes sense to be
> > able to ask the array for its highest index.
>
> You can parse the output of ${!array[@]} to get that.
>

True, but stupidly inefficient for a large array. (If you're inclined to
respond "but this is Shell so nobody cares about performance", please read
my rebuttal in the PS below.)

How about making ${!array[@]:s:n} work in the obvious manner, including
when s=-1 n=1?

-Marin

PS: Everyone assumes that "the speed of external executables is the
limiting factor", but it turns out that's not true. It's possible to do
MUCH worse, just using built-in features:

27$ echo $BASH_VERSION
> 5.1.4(47)-release
>
> 28$ grep -m1 'model name' /proc/cpuinfo
> model name      : Intel(R) Core(TM)2 Duo CPU     T7100  @ 1.80GHz
>
> 29$ time A=({1..20000})
>
> real    0m0.093s
> user    0m0.076s
> sys     0m0.008s
> 30$ time B=( "${A[@]}" )
>
> real    0m0.135s
> user    0m0.120s
> sys     0m0.004s
> 31$ time I=( "${!A[@]}" )
>
> real    0m0.135s
> user    0m0.124s
> sys     0m0.000s
> 32$ time ls -ld .
> drwxrwxr-x 7 martin martin 300 2023-05-28 00:50:48.683430004 +1000 ./
>
> real    0m0.091s
> user    0m0.008s
> sys     0m0.068s
>

Now imagine doing that in a loop, and a script that takes HOURS instead of
sub-second is entirely plausible.

Or imagine running it on an older version of the shell:

16$ echo $BASH_VERSION
> 4.0.0(58)-release
> 17$ grep -m1 'model name' /proc/cpuinfo
> model name      : Intel(R) Core(TM)2 Duo CPU     T7100  @ 1.80GHz
> 18$ time A=({1..20000})
>
> real    0m2.215s
> user    0m2.172s
> sys     0m0.008s
> You have mail in /var/mail/martin
> 19$ time B=( "${A[@]}" )
>
> real    0m0.390s
> user    0m0.344s
> sys     0m0.016s
> 20$ time I=( "${!A[@]}" )
>
> real    0m2.754s
> user    0m2.724s
> sys     0m0.004s
> 21$ time ls -ld .
> drwxrwxr-x 7 martin martin 300 May 28 00:50 .
>
> real    0m0.069s
> user    0m0.000s
> sys     0m0.004s
> 22$
>

Yes, that's 40 times slower.

-Martin