Re: $!

At 10:11 AM 8/17/00 -0700, Larry Wall wrote:
>Peter Scott writes:
>: Do we want to come up with any modifications to the scope of $! (I'm not
>: talking about the dual string-number nature)?  I've occasionally been
>: flummoxed by looking at it when there hadn't really been an error (but it's
>: set anyway, randomly it seems); or calling some module method that failed
>: and then wondering whether $! tells me anything useful or not (did the
>: method make a failed system call?).
>:
>: I realize this is very pedestrian compared to the exception-handling stuff
>: we've been tossing around, which could largely be said to render the issue
>: moot; but I thought I'd shake the branches anyway and see what fell out.
>
>At minimum we could lexically scope $! to the current file.  I would hate
>to lose the notational convenience of $!.

Well, I wasn't thinking of doing away with it (although there are other 
proposals that would probably have that effect).  I just wanted to make it 
more useful.  There are loads of times that I call out to something and it 
fails, and I never know whether $! will tell me something or not.  Suppose 
I get a new Net::POP3 handle but I get undef instead.  Will $! tell me 
anything about why?  Maybe, if there was a bad socket call... maybe 
not.  But as far as I'm concerned, a system call failed (where 'system' is 
just some piece of underlying infrastructure).  Having to know the 
difference between those and functions from section 2 of the Unix manual is 
onerous; I keep expecting $! to be a generic sideband error notification 
thingy.

Is it worth trying to do that?

--
Peter Scott
Pacific Systems Design Technologies