Re: [Qemu-devel] tests/Makefile.include non-portable use of $RANDOM

[Eric Blake <eblake@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 1486 bytes --]

On 07/13/2017 12:32 PM, Peter Maydell wrote:
> Currently our test makefile does
> 
>    MALLOC_PERTURB_=$${MALLOC_PERTURB_:-$$((RANDOM % 255 + 1))}
> 
> This works on bash, but $RANDOM is bash-specific.
> On dash, it doesn't do what we want, but it does do something:
> 
> $ echo $((RANDOM % 255 + 1))
> 1

Being non-random doesn't hurt (it's just less test coverage by default).

> 
> On NetBSD the shell complains:
> # echo $((RANDOM % 255 + 1))
> -sh: arith: syntax error: "RANDOM % 255 + 1"
> 
> and so 'make check' doesn't work.
> 
> Any suggestions for something more portable?

Yes; echo $(( ${RANDOM:-0} % 255 + 1))

should reliably give you 1, even on current BSD /bin/sh, regardless of
whether BSD adds $RANDOM support in the future.

> https://unix.stackexchange.com/questions/140750/generate-random-numbers-in-specific-range
> suggests
>   awk 'BEGIN{srand(); print int(1+rand()*256)}'
> and of course there are similar things possible with perl et al.

But why bother. MALLOC_PERTURB_ only makes sense on glibc systems, so
picking a non-random value is no worse on non-glibc systems.

> 
> Or we could just have the makefile check whether we have bash,
> and if not then set
>   RANDOM=4

Picking that is the same as using ${RANDOM:-4}, without having to
hardcode a check for whether you have bash.

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 604 bytes --]