The best way to extend a given function is generally to add args to the end of the arg list. This should not break old code but there is always the possibility that it will. So if you have lots of args you want to pass then passing an arg hash is a sound approach....BUT it looks like the new() function in Net::SMTP might cause you grief.

If you trace through what occurs when you call send('smtp',@args) in Mime::Lite you effectively end up calling send_by_smtp(@args) which ends up with this:

my $smtp = MIME::Lite::SMTP->new(@args)
[download]

Given that this just calls the Mime::Lite::SMTP package which is a subclass of Net::SMTP.....this is where that call FINALLY ends up in Net::SMTP:

sub new
{
 my $self = shift;
 my $type = ref($self) || $self;
 my $host = shift if @_ % 2;
 my %arg  = @_; 
 my $hosts = defined $host ? [ $host ] : $NetConfig{smtp_hosts};
 my $obj;
 my $h;
 foreach $h (@{$hosts})
  {
   $obj = $type->SUPER::new(PeerAddr => ($host = $h), 
                PeerPort => $arg{Port} || 'smtp(25)',
                LocalAddr => $arg{LocalAddr},
                LocalPort => $arg{LocalPort},
                Proto    => 'tcp',
                Timeout  => defined $arg{Timeout}
                        ? $arg{Timeout}
                        : 120
               ) and last;
  }
[download]

It seems to me that there is an assumption there that you are going to break. First the first arg to new() for Net::SMTP needs to be ($server/$host) and worse the %2 means it only recognises the $host arg if an ODD number of args is passed to new..... It then goes on to assume that all following args are a hash presented as a flat list. Adding your single scalar (hash ref) will cause it to explode with an odd number of elements in hash assignment error. Also passing {Debug=>1} as suggested will cause Net::SMTP to try to connect to a HASH REF which just aint gonna work. Presumably you plan to override this function in Mime::Lite - looks like you will have to to me.

Personally a standard coding practice around here is NEVER to write a function like func($scalar, @ary) - we always do func($scalar, \@ary) so we can easily add $forgot_this, $forgot_that to the end of the arg list func( $scalar, \@ary, $forgot_this, $forgot_that ). As soon as you code a trailing array into a function call you are screwed if you need to extend it without causing major dramas (ie breaking just about everything that calls that func)......

I'm thinking of intercepting the hashref before calling new. So that

    require Net::SMTP;
    my $smtp = MIME::Lite::SMTP->new(@args)
        or Carp::croak("Failed to connect to mail server: $!\n");
[download]

becomes something like...

    require Net::SMTP;
    my $smtp;
    if (ref $args[0]) {
        $opts = shift @args;
        # handle special cases, and finally resulting in a usable valu
+e
        # in $smtp.
    } else {
       $smtp = MIME::Lite::SMTP->new(@args)
         or Carp::croak("Failed to connect to mail server: $!\n");
    }
[download]

So before MIME::Lite::SMTP even gets called @args has been precleaned as appropriate. This would allow special cases like having a list of SMTP hosts that are tried until one lets us in, or handling authentication, or turning Debug on in Net::SMTP without having to do the ugly Net::SMTP->new(undef,Debug=>1) in the user code (inside MIME::Lite its ok to do this, as in there its unlikely to cause trouble to a newbie).

Also I agree quite a bit with your points about positional arguments. The only quibble I have is from Perls flexibility with arguments and weakish typing. You can often extend positional arguments by getting creative with types. For instance the trick above works out because its extremely unlikely that somebody is using a blessed reference as a hostname. In fact I could get a lot more picky and check that the ref is a HASH and that if it is a hash that stringification is not overloaded. This would result in total backwards compatibility while still allowing extension.

Cheers,


---
demerphq



First they ignore you, then they laugh at you, then they fight you, then you win.
-- Gandhi

The else doesn't need to be there. The $smtp is still going to need to be created in the same way (or at least it seems so to me). The authentication action occurs after the Net::SMTP object has been created, which was part of my initial concern when adding support to MIME::Lite. I am still working through why Net::SMTP doesn't allow for the authentication arguments in the object creation. It would seem that Net::SMTP should perhaps be the one that bends and allows for the authenication to be passed into it on object creation.

If we moved this logic into the Net::SMTP module, what would happen if we did this right before we return the Net::STMP object:
if (defined $arg{User}) {
       $obj->auth($arg{User},$arg{Password});
}
[download]
inside of the Net::SMTP new? Then when we create a new Net::SMTP call we do it as:
        MIME::Lite->send('smtp', $out_smtp ,
                          Port =>$smtp_port ,
                          Timeout=>60 ,
                          Debug => $debug ,
                          Hello => 'domain.com',
                          User  => 'trs80',
                          Password  => 'blah' );
[download]
My simple test indicates that it works, but I need to carefully review the side effects of performing the auth in the object creation.

Placing the change inside of Net::SMTP makes more sense in that it keeps the logic in the right problem space and thereby making it benefical to a larger user base.

This reply is to address the change if the modification takes place in MIME::Lite vs. Net::SMTP as my previous post suggests. I think the change in behavior should occur at the send method rather then the send_by_smtp. This would allow future modifications of send methods to accounted for in the same manner. So here is my proposed revised sub send:

# add new global that the top
my $SenderOpts = '';


sub send {
    my $self = shift;

    if (ref($self)) {              ### instance method:
        my ($method, @args);
        if (@_) {                            ### args; use them just t
+his once
            $method = 'send_by_' . shift;
            @args   = @_;
        }
        else {                               ### no args; use defaults
            $method = "send_by_$Sender";
            @args   = @{$SenderArgs{$Sender} || []};
        }
        $self->verify_data if $AUTO_VERIFY;  ### prevents missing part
+s!
        return $self->$method(@args);
    }
    else {                         ### class method:
        if (@_) {

            my @old = ($Sender, @{$SenderArgs{$Sender}});
            $Sender = shift;

            if ( ref($_[0]) eq 'HASH') {
                $SenderOpts = shift;
            }
            $SenderArgs{$Sender} = [@_];    ### remaining args
            return @old;
        }
        else {
          Carp::croak "class method send must have HOW... arguments\n"
+;
        }
    }
}
[download]
Then below in our send_by_smtp method we add:
    if ($SenderOpts) {
         $smtp->auth( $SenderOpts->{auth_username}, $SenderOpts->{auth
+_password} );
    }
[download]
directly after our $smtp object is created.