RE: Re: Re: time command with shell builtins

[zeurkous@xxxxxxxxxxxxxxx]

On Mon, 23 Jan 2023 15:23:57 +0100, Roman Perepelitsa <roman.perepelitsa@xxxxxxxxx> wrote:
> On Mon, Jan 23, 2023 at 3:20 PM <zeurkous@xxxxxxxxxxxxxxx> wrote:
>>
>> On Mon, 23 Jan 2023 14:42:05 +0100, Roman Perepelitsa <roman.perepelitsa@xxxxxxxxx> wrote:
>> > On Mon, Jan 23, 2023 at 2:40 PM Dominik Vogt <dominik.vogt@xxxxxx> wrote:
>> >>
>> >> Is it possible to get timing statistics of shell builtins too?
>> >> Timing "echo" isn't very interesting, but timing loop constructs
>> >> would be:
>> >>
>> >>   $ time while foo; do bar done
>> >
>> > This:
>> >
>> >     % time ( while foo; do bar; done )
>>
>> Doesn't that imply a fork, which will also be timed...?
>
> Yes, it does.
>
>> (Could lead to misleading results, in that case.)
>
> Any tool that measures how long a piece of code takes to run gives you
> an upper bound, so there is nothing new here.

A fork(2) can be particularly expensive, though, so there's an
opportunity to make things better here.

> The standard solution is
> to ensure that the code under benchmark takes a non-trivial amount of
> time to execute. Here's the standard pattern I use:
>
>     time ( repeat 1000 code-under-benchmark )
>
> I tune the loop constant so that the benchmark takes around a second.

Me's of course aware of such workarounds (they aren't solutions, though
me agrees that a complete solution would be far beyond our scope).

         --zeurkous.

> Roman.

-- 
Friggin' Machines!