Phased start - bug or misconception?

[Steve Davis]

Hi,

I've a need to use a phased start for an application, but came across
some unexpected behaviour. i.e. that a start failure in a phase leaves
things in an apparently inconsistent state.... Boiling this down to
the minimum, and given:
---
%%phased.app
{application, phased, [
	{description, "Is this behaviour expected?"},
	{vsn, "1.0"},
	{applications, [kernel, stdlib]},
	{mod, {phased, []}},
	{start_phases, [ {first, []}, {second, []} ]},
	{env, []}
]}.
---
%% phased.erl
-module(phased).

-behaviour(application).
-export([start/2, start_phase/3, config_change/3, prep_stop/1, stop/
1]).

-behaviour(supervisor).
-export([init/1]).

start(normal, []) ->
	io:format("start~n"),
	supervisor:start_link({local, phased_sup}, ?MODULE, []).

start_phase(first, normal, []) ->
	io:format("start_phase: first~n"),
	ok;

start_phase(second, normal, []) ->
	io:format("start_phase: second~n"),
	{error, intentional}.

config_change(_, _, _) ->
	ok.

prep_stop(State) ->
	State.

stop(_State) ->
	ok.

init([]) ->
	{ok, {{one_for_all, 0, 1}, []}}.
---

Noting that I'm intentionally failing phase 2 of the startup
process... here's the console session:

4> application:start(phased).
start
start_phase: first
start_phase: second

=INFO REPORT==== 23-Mar-2010::04:11:25 ===
    application: phased
    exited: {intentional,{phased,start_phase,[second,normal,[]]}}
    type: temporary
{error,{intentional,{phased,start_phase,
                            [second,normal,[]]}}}
5> application:which_applications().
[{stdlib,"ERTS  CXC 138 10","1.16.5"},
 {kernel,"ERTS  CXC 138 10","2.13.5"}]
6> application:start(phased).
start

=INFO REPORT==== 23-Mar-2010::04:11:49 ===
    application: phased
    exited: {{already_started,<0.45.0>},{phased,start,[normal,[]]}}
    type: temporary
{error,{{already_started,<0.45.0>},
        {phased,start,[normal,[]]}}}
7> application:stop(phased).
{error,{not_started,phased}}
8> registered().
[global_group,init,erl_prim_loader,user,error_logger,rex,
 standard_error_sup,kernel_sup,global_name_server,inet_db,
 file_server_2,code_server,user_drv,phased_sup,
 standard_error,application_controller,kernel_safe_sup]

Should phased_sup still be running here? Is so, how do you recover?

This looks like a bug to me but before reporting as such, I thought
I'd better check that thIs is not some misconception on my part?

TIA,
/s