Re: More POSIX developments

[Bart Schaefer <schaefer@brasslantern.com>]

Finally got a chance to look at this.

On Mon, 27 Sep 2004, Peter Stephenson wrote:

> The following patch, which I won't commit until we've decided which way
> to go, tries to cover the bases by making test work (more) like POSIX
> while leaving [[ ... ]] the way it is.  The errors now return status 2
> from evalcond, but for backward compatibility [[ ... ]] turns them into
> shell errors.

I think this would be OK.

> I needed to add an extra warning function to choose between the cases
> where we have a builtin name and where the code is called internally as
> a condition.  Does anyone know if there's a reason zwarnnam handles the
> absence of a command name in a strange way?

I paged through a grep of all the calls to zwarnnam(), and there don't
seem to be any that intentionally get passed a NULL first argument, so it
probably is just defensive programming.

> In other words, could I remove zwarnnamopt and make zwarnnam behave like 
> that?

I think that'd be safe, but I wouldn't object to a second opinion.