response

[ikiwiki.git] / doc / todo / support_for_plugins_written_in_other_languages.mdwn

diff --git a/doc/todo/support_for_plugins_written_in_other_languages.mdwn b/doc/todo/support_for_plugins_written_in_other_languages.mdwn
index 65fd84f90786b33eeeec88a4f8e501e57ed56e95..8476d1b44f8abdb5b9c38e7de7947acdd9de9511 100644 (file)
--- a/doc/todo/support_for_plugins_written_in_other_languages.mdwn
+++ b/doc/todo/support_for_plugins_written_in_other_languages.mdwn
@@ -1,5 +1,7 @@
 ikiwiki should support writing plugins in other languages
 
+> [[done]] !!
+
 While it should be possible to call ikiwiki from C, doing the callbacks in C is
 probably hard. And accessing perl at all from C is ugly. It also doesn't
 make it very easy to write plugins in an interpreted language, since that
@@ -8,40 +10,47 @@ that, never again.)
 
 Instead, I'm considering using XML RPC to let ikiwiki communicate with a
 child process that it can spawn. The child could then be any program,
-written in any language. It could talk XML RPC via stdio.
+written in any language. It could talk XML RPC via stdio. (This assumes
+that most languages allow easily serialising XML::RPC calls and responses
+to a file descriptor. Some XML RPC implementations may be hardcoded to use
+http..) For ease of implementation, each rpc request sent via stio should
+end with a newline, and begin with "<?xml ..>".
 
 Here's how it would basically look, not showing the actual XML RPC used to
 pass values.
 
-       -> init
-       <- 1
-       <- hook type => preprocess, id => foo
-       -> 1
-       <- done 1
-       -> 1
-
-       -> callback type => preprocess id => foo, page => bar
-       <- 1
-       <- getconfig url
-       -> "http://example.com", ...
-       <- debug "foo"
-       -> 1
-       <- done "my return value"
-       -> 1
+       -> call import
+       <- call hook type => preprocess, id => foo, call => plugin_preprocess
+       -> result 1
+       <- result 1
+
+       -> call plugin_preprocess page => bar
+       <- call getconfig url
+       -> result "http://example.com", ...
+       <- call debug "foo"
+       -> result 1
+       <- result done "my return value"
 
 From ikiwiki's POV:
 
 * ikiwiki always initiates each conversation with a command
 * After sending a command, ikiwiki reads commands, dispatches them, and 
-  returns the results, in a loop.
-* The loop continues until the plugin calls the "done" command, with a value
-  that is the return value for the command that initiated the conversation.
+  returns the results, in a loop, until it gets a result for the command it
+  called.
 
 From the plugin's POV:
 
 * It's probably sitting in an XML::RPC loop.
 * Get a command from ikiwiki.
-* Call the appropriate callback.
-* The callback can use XML::RPC to communicate with ikiwiki to get things
+* Dispatch the command to the appropriate function.
+* The function can use XML::RPC to communicate with ikiwiki to get things
   like config values; and to call ikiwiki functions.
-* When the callback returns, use XML::RPC to send a "done" command to ikiwiki.
+* Send the function's return value back to ikiwiki.
+
+Simple enough, really. ikiwiki would need to add accessor functions for
+all important variables, such as "getconfig" and "setconfig". It would
+probably be easiest for ikiwiki to dispatch a command by just evaling
+IkiWiki::$command.
+
+Plugin programs could be dropped into /usr/share/ikiwiki/plugins/, and
+load_plugin() would just open2 the plugin program and call import.