Default implementation of full message filtering.

The function invokes each filter in the filter chain and combines the results to get a final answer in the for of a SpamAction status.

It first determines the immediate spam status, calling the on_immediate_status callback with the result. Then it starts a background task to determine the asynchronous state and, if different to the immediate status, passes the result to the on_async_status callback.

Afterwards it will call AntispamState.classify to update any self-learning filters in the chain.

alias on_immediate_status
alias on_async_status
AntispamMessage message

Return Value

Type: Task

A handle to the background task is returned. This can be used to determine if the asynchronous part has finished.


import : parseJsonString;
import std.algorithm.comparison : among;

void main()
	auto config = parseJsonString(
			{"filter": "bayes"},
			{"filter": "blacklist",
				"settings": {
					"ips": ["", ""]

	auto antispam = new AntispamState;

	AntispamMessage msg;
	msg.headers["Subject"] = "8uy CH34P V14GR4!!11";
	msg.message = cast(const(ubyte)[])"Just look here: http://bestdrugdealz.c0m";
	msg.peerAddress = ["123.52.433.1", "vps12315.some.provider.n3t"];

		(status) {
			if (status.among(SpamAction.revoke, SpamAction.block))
				throw new Exception("Your message has been rejected!");
			// otherwise store message...
		(async_status) {
			if (async_status.among!(SpamAction.revoke, SpamAction.block)) {
				// Flag or delete the stored message.

			// It could also theoretically happen here that async_status is amnesty
			// or pass, so that a message that was already rejected in the first
			// phase would be accepted in retrospective. You'll have to decides on
			// a per-application basis if it makes sense to support this case, or
			// if immediate rejections always have precedence.