finished when we need to trigger the continuation immediately.
Via a fairly complex path, this fixes the problem where all hosts in
the build farm that do not have ipv6 failed a lot of the RPC
tests. This happened because the dcerpc_connect() async code used a
composite_continue() on a context which was already in an error state,
due to the socket backend saying that ipv6 was unavailable
(This used to be commit
dbf935d38b6b1fea5ed00e94c9b1a518cb14768b)
if (composite_nomem(new_ctx, ctx)) return;
new_ctx->async.fn = continuation;
new_ctx->async.private_data = private_data;
+
+ /* if we are setting up a continuation, and the context has
+ already finished, then we should run the callback with an
+ immediate event, otherwise we can be stuck forever */
+ if (new_ctx->state >= COMPOSITE_STATE_DONE && continuation) {
+ event_add_timed(new_ctx->event_ctx, new_ctx, timeval_zero(), composite_trigger, new_ctx);
+ }
}
_PUBLIC_ void composite_continue_rpc(struct composite_context *ctx,