Re: init.sh changes pushed

[Jim Meyering]

Eric Blake wrote:

> According to Jim Meyering on 2/15/2010 11:36 PM:
>> I've pushed the init.sh changes we discussed yesterday.
>
> I just realized something - for coreutils, you guarantee that init.sh will
> be called by a relatively-compliant shell.  But gnulib makes no special
> effort to avoid /bin/sh, which means init.sh will now fail on Solaris,
> over constructs like:
>
>> +      *) feb_file_=$(echo $feb_file_ | sed "s,^$feb_dir_/,,;"'s/\.exe$//')
>
> How do we go about fixing this?  Should init.sh be given some smarts (like
> autoconf-generated files) to re-exec the calling script using a saner
> shell?

I like this idea.
Encapsulating the smarts in the test framework itself
would be more maintainable/reusable than what coreutils does.

> Is this something where any test using init.sh will have to modify
> Makefile.am to guarantee that the script is called with the proper
> ${CONFIG_SHELL} or equivalent?