Re: build.sh on DragonFly, sh detection for HOST_SH, and configure cannot guess build type

[Alan Barrett <apb%cequrux.com@localhost>]

On Thu, 01 Jan 2009, Jeremy C. Reed wrote:
> build.sh (on 5.0_BETA) has:
> 
>  [ -z "${HOST_SH}" ] && HOST_SH="$(
>         (ps -p $$ -o comm | sed -ne '2s/[ \t]*$//p') 2>/dev/null )"
> 
> "comm" is unknown on DragonFly. It does have "command" keyword.
>
> This results in HOST_SH being set to the unrelated ps output.

"comm" is standardised, see
<http://www.opengroup.org/onlinepubs/009695399/utilities/ps.html>.

If it's not supported, I'd expect the ps command to produce an error
message on stderr, and nothing on stdout, with the result that HOST_SH
remains set to the empty string.

> Also the \t is not a TAB in that version of sed/regex. It is a literal 
> letter "t".

That's a bug in build.sh; I'll change it to use [[:space:]], or maybe
avoid using sed at that point.

> So I used:
> 
>         [ -z "${HOST_SH}" ] && HOST_SH="$(
> -               (ps -p $$ -o comm | sed -ne '2s/[ \t]*$//p') 2>/dev/null )"
> +               (ps -p $$ -o command 2>/dev/null | sed -ne '2s/[        
> ].*$//p') 2>/dev/null )"

That will truncate the command name after the first space.  I worry
about people keeping their shell in a directory whose name contains a
space (especially on Windows).

> Second issue was:
> 
> checking build system type... 
> /build/reed/nb/src/tools/gmake/../../gnu/dist/gmake/config/config.guess: 
> unable to guess system type

I can't help you there (but see Alistair's reply).

--apb (Alan Barrett)