Modified:
   doc/trunk/design/syn/S06.pod

Log:
More old use of multiple invocant terminology changed to longnames.
Added mechanism for both short and long switch names.

Modified: doc/trunk/design/syn/S06.pod
=========================================================================== ===
--- doc/trunk/design/syn/S06.pod        (original)
+++ doc/trunk/design/syn/S06.pod        Thu Aug 17 16:39:38 2006
@@ -13,9 +13,9 @@

   Maintainer: Larry Wall <la...@wall.org>
   Date: 21 Mar 2003
-  Last Modified: 14 Aug 2006
+  Last Modified: 17 Aug 2006
   Number: 6
-  Version: 49
+  Version: 50

 This document summarizes Apocalypse 6, which covers subroutines and the
@@ -481,21 +481,10 @@
     print $obj.get_name();
     $obj.set_name("Sam");

-Multimethod and multisub invocants are specified at the start of the parameter
-list, with a colon terminating the list of invocants:
-
-    multi sub handle_event ($window, $event: $mode) {...}   # two invocants
-    multi method set_name ($self, $name: $nick) {...}       # two invocants
-
-If the parameter list for a C<multi> contains no colon to delimit
-the list of invocant parameters, then all positional parameters are
-considered invocants.  If it's a C<multi method> or C<multi submethod>,
-an additional implicit unnamed C<self> invocant is prepended to the
-signature list.
-
 For the purpose of matching positional arguments against invocant parameters,
 the invocant argument passed via the method call syntax is considered the
-first positional argument:
+first positional argument when failover happens from single dispatch to
+multiple dispatch:

     handle_event($w, $e, $m);   # calls the multi sub
     $w.handle_event($e, $m);    # ditto, but only if there is no
@@ -509,14 +498,28 @@
                             # fall-back to set_name($obj, "Sam")
     $obj.set_name("Sam");   # same as the above

-Passing too many or too few invocants is a fatal error if no matching
-definition can be found.
-
 An invocant is the topic of the corresponding method or multi if that
 formal parameter is declared with the name C<$_>.  A method's first
 invocant always has the alias C<self>.  Other styles of self can be
 declared with the C<self> pragma.

+=head2 Longname parameters
+
+Much like ordinary methods give preference to the invocant,
+multimethods and multisubs can give preference to earlier parameters.
+These are called I<longnames>; see S12 for more about the semantics
+of multiple dispatch.  Syntactically, longnames are declared by
+terminating the list of important parameters with a semicolon:
+
+    multi sub handle_event ($window, $event; $mode) {...}
+    multi method set_name ($self: $name; $nick) {...}
+
+If the parameter list for a C<multi> contains no semicolon to delimit
+the list of invocant parameters, then all positional parameters are
+considered invocants.  If it's a C<multi method> or C<multi submethod>,
+an additional implicit unnamed C<self> invocant is prepended to the
+signature list unless the first parameter is explicitly marked with a colon.
+

 =head2 Required parameters

@@ -2534,3 +2537,14 @@
 parameters, but still give you access to nested matches through those
 parameters, just as any C<Match> object would.  Of course, in this example,
 there's no particular reason the sub has to be named C<MAIN>.
+
+To give both a long and a short switch name, you may use the pair
+notation.  The key will be considered the short switch name, while
+the variable name will be considered the long switch name.  So if
+the previous declaration had been:
+
+    sub MAIN (:f($frompart), :t($topart), *@rest)
+
+then you could invoke the program with either C<-f> or C<--frompart>
+to specify the first parameter.  Likewise you could use either C<-t>
+or C<--topart> for the second parameter.

--
Markus Laire

Hmm, that opens up a world of hurt.  Either you have to distinguish a
--abc from -abc, or you have to have some kind of fallback heuristic,
and it doesn't work terribly well with arguments in any case except
for the final one.  Should probably make it possible, just because the
external interface is one of the places where Perl has always tried
to be accommodating to existing culture rather than revisionist.
We can probably work something out here, along the lines of:

    if there's only one -
    if single character aliases are defined
    if the word matches that alphabet
    if the word doesn't match any longer names

At first I was inclined to say that if there's a *% then all the
unrecognized go in there and you can parse the -abc yourself, but
that doesn't tell you how to treat the next argument unless we look
at the definition of -c anyway.  We can't just say that -c's arg
must use the -c=arg form, since even Perl 5 violates that with -e.  :/

Larry

Still I was thinking of behaviour where C<-abc> would allways mean
C<-a -b -c> regardless of what 1-char aliases or longer names have
been defined. This would make --abc and -abc mean completely different
things.

And in this proposal only the last switch would be able to get an
argument, e.g. with C<-abc=99> or C<-abc 99> or something like that.

If this can't be the default behaviour, then it would be nice to be
able to easily switch to this kind of behaviour.

ps. Then there's the perl5-behaviour of "perl -n0e unlink" where also
the intervening switches can get arguments. This could be expanded so
that all chars for which there's no 1-char alias defined, are
parameters. So C<-aHellobWorld> would mean C<-a=Hello -b=World> if
there are 1-char aliases only for a & b. ;)

--
Markus Laire

I think that safely falls into the category of completely psychotic.  @L@

Larry