"Kurt D. Starsinic" <[EMAIL PROTECTED]> writes:
> On Wed, Nov 01, 2000 at 06:53:02PM -0800, Ask Bjoern Hansen wrote:
> > On Wed, 1 Nov 2000, Kurt D. Starsinic wrote:
> >
> > > I'm working on a patch to LWP::UserAgent to detect failures other than
> > > those that can be assigned HTTP status codes -- particularly, network
> > > failures.
> >
> > That sounds absurdly useful.
> >
> > [...]
> > > Before I do any more work on the right patch, does anybody have a strong
> > > view that LWP::UserAgent and HTTP::Status should _not_ be cognizant of
> > > errors that aren't HTTP errors?
> >
> > IMHO, everything should be fine they're just marked clearly as such
> > in the documentation (and hence are something that might change if
> > the status codes gets official values).
>
> Should a connection error be is_client_error() or is_server_error()?
> My purist view is that it should be neither, but that it should be
> is_error(), and that its numeric value should be undef. Thoughts?
I consider the LWP core to be an implicit local proxy server that you
can always connect too. A failed connection to the final destination
means that this local proxy could not relay the request. With this
viewpoint I consider it correct to report the failure with some 5xx
code.
Codes that are undef does not seem very attractive to me.
Regards,
Gisle
