The FSM boilerplate based on callbacks
Finitomata
provides a boilerplate for FSM implementation, allowing to concentrate on the business logic rather than on the process management and transitions/events consistency tweaking.
It reads a description of the FSM from a string in PlantUML, Mermaid, or even custom format.
Mermaid
state diagram format is literally the same asPlantUML
, so if you want to use it, specifysyntax: :state_diagram
and if you want to use mermaid graph, specifysyntax: :flowchart
. The latter is the default.
Basically, it looks more or less like this
[*] --> s1 : to_s1
s1 --> s2 : to_s2
s1 --> s3 : to_s3
s2 --> [*] : ok
s3 --> [*] : ok
s1 --> |to_s2| s2
s1 --> |to_s3| s3
Mermaid
does not allow to explicitly specify transitions (and hence event names) from the starting state and to the end state(s), these states names are implicitly set to:*
and events to:__start__
and:__end__
respectively.
Finitomata
validates the FSM is consistent, namely it has a single initial state, one or more final states, and no orphan states. If everything is OK, it generates a GenServer
that could be used both alone, and with provided supervision tree. This GenServer
requires to implement six callbacks
on_transition/4
— mandatoryon_failure/3
— optionalon_enter/2
— optionalon_exit/2
— optionalon_terminate/1
— optionalon_timer/2
— optional
All the callbacks do have a default implementation, that would perfectly handle transitions having a single to state and not requiring any additional business logic attached.
Upon start, it moves to the next to initial state and sits there awaiting for the transition request. Then it would call an on_transition/4
callback and move to the next state, or remain in the current one, according to the response.
Upon reaching a final state, it would terminate itself. The process keeps all the history of states it went through, and might have a payload in its state.
If the event name is ended with a bang (e. g. idle --> |start!| started
) and
this event is the only one allowed from this state (there might be several transitions though,)
it’d be considered as determined and FSM will be transitioned into the new state instantly.
If the event name is ended with a question mark (e. g. idle --> |start?| started
,)
the transition is considered as expected to fail; no on_failure/2
callback would
be called on failure and no log warning will be printed.
If timer: non_neg_integer()
option is passed to use Finitomata
,
then c:Finitomata.on_timer/2
callback will be executed recurrently.
This might be helpful if FSM needs to update its state from the outside
world on regular basis.
If auto_terminate: true() | state() | [state()]
option is passed to use Finitomata
,
the special __end__
event to transition to the end state will be called automatically
under the hood, if the current state is either listed explicitly, or if the value of
the parameter is true
.
If ensure_entry: true() | [state()]
option is passed to use Finitomata
, the transition
attempt will be retried with {:continue, {:transition, {event(), event_payload()}}}
message
until succeeded. Neither on_failure/2
callback is called nor warning message is logged.
The payload would be updated to hold __retries__: pos_integer()
key. If the payload was not a map,
it will be converted to a map %{payload: payload}
.
See examples directory for
real-life examples of Finitomata
usage.
Let’s define the FSM instance
defmodule MyFSM do
@fsm """
s1 --> |to_s2| s2
s1 --> |to_s3| s3
"""
use Finitomata, fsm: @fsm, syntax: :flowchart
## or uncomment lines below for `:state_diagram` syntax
# @fsm """
# [*] --> s1 : to_s1
# s1 --> s2 : to_s2
# s1 --> s3 : to_s3
# s2 --> [*] : __end__
# s3 --> [*] : __end__
# """
# use Finitomata, fsm: @fsm, syntax: :state_diagram
@impl Finitomata
def on_transition(:s1, :to_s2, _event_payload, state_payload),
do: {:ok, :s2, state_payload}
end
Now we can play with it a bit.
# or embed into supervision tree using `Finitomata.child_spec()`
{:ok, _pid} = Finitomata.start_link()
Finitomata.start_fsm MyFSM, "My first FSM", %{foo: :bar}
Finitomata.transition "My first FSM", {:to_s2, nil}
Finitomata.state "My first FSM"
#⇒ %Finitomata.State{current: :s2, history: [:s1], payload: %{foo: :bar}}
Finitomata.allowed? "My first FSM", :* # state
#⇒ true
Finitomata.responds? "My first FSM", :to_s2 # event
#⇒ false
Finitomata.transition "My first FSM", {:__end__, nil} # to final state
#⇒ [info] [◉ ⇄] [state: %Finitomata.State{current: :s2, history: [:s1], payload: %{foo: :bar}}]
Finitomata.alive? "My first FSM"
#⇒ false
Typically, one would implement all the on_transition/4
handlers, pattern matching on the state/event.
def deps do
[
{:finitomata, "~> 0.30"}
]
end
0.30.0
— [UPD]Finitomata.Flow
, tons of tiny improvements0.28.0
— [UPD] initialtelemetria
integration0.27.0
— [UPD] optionshibernate: boolean()
andcache_state: boolean()
0.26.0
— [UPD] a lot of tiny improvements,Finitomata.Accessible
,reset_timer
message + tests, experimentalFinitomata.Cache
0.25.0
— [UPD] allow assertions of entry states inFinitomata.ExUnit
0.24.2
— [UPD/FIX] many fixes for better diagnostics inFinitomata.ExUnit
0.23.7
— [UPD] allow both:mox
and{:mox, MyApp.Listener}
as well as justMyApp.Listener
as a listener in FSM definition0.23.4
— [FIX] many fixes to aFinitomata.ExUnit
test scaffold generation0.23.0
— [UPD]mix finitomata.generate.test --module MyApp.FSM
to generate aFinitomata.ExUnit
test scaffold0.22.0
— [FIX]Infinitomata.start_fsm/4
is finally 102% sync0.21.4
— [FIX]Finitomata.Pool
initialization in cluster0.21.3
— [FIX] proper return fromInfinitomata.start_fsm/4
0.21.1
— [UPD]listener: :mox
and betterFinitomata.ExUnit
docs0.20.2
— [UPD] allow guard matches in the RHO of~>
operator inassert_transition/3
0.20.0
— [FIX] starting pool on distribution, re-synch on:badrpc
failure0.19.0
— [UPD]Finitomata.ExUnit
lighten options check (compile-time module dependencies suck in >=1.16)0.18.0
— [UPD] asynchronousFinitomata.Pool
on top ofInfinitomata
0.17.0
— [UPD] careful naming andFinitomata.Throttler
0.16.0
— [UPD]Infinitomata
as a self-contained distributed implementation leveraging:pg
0.15.0
— [UPD] support snippet formatting for modern Elixir0.14.6
— [FIX] persistency flaw when loading [credits @peaceful-james]0.14.5
— [FIX]require Logger
inHook
0.14.4
— [FIX] Docs cleanup (credits: @TwistingTwists),PlantUML
proper entry0.14.3
— [FIX] Draw diagram in docs0.14.2
— [FIX] StopEvents
process0.14.1
— [FIX] Incorrect detection of superfluous determined transitions0.14.0
—Finitomata.ExUnit
improvements0.13.0
— compile-time helpers for FSM,Finitomata.ExUnit
0.12.1
—c:Finitomata.on_start/1
callback0.11.3
— [FIX] better error message for options (credits @ray-sh)0.11.2
— [DEBT] exportedFinitomata.fqn/2
0.11.1
—Inspect
,:flowchart
/:state_diagram
as default parsers, behaviourParser
0.11.0
—{:ok, state_payload}
return fromon_timer/2
,:persistent_term
to cache state0.10.0
— support for several supervision trees withid
s, experimental support for persistence scaffold0.9.0
— [FIX] malformed callbacks had the FSM broken0.8.2
— last error is now kept in the state (credits to @egidijusz)0.8.1
— improvements to:finitomata
compiler0.8.0
—:finitomata
compiler to warn/hint about not implemented ambiguous transitions0.7.2
— [FIX]banged!
transitions must not be determined0.6.3
—soft?
events which do not callon_failure/2
and do not log errors0.6.2
—ensure_entry:
option to retry a transition0.6.1
— code cleanup +auto_terminate:
option to make:__end__
transition imminent0.6.0
—on_timer/2
and banged imminent transitions0.5.2
—state()
type on generated FSMs0.5.1
— fixed specs [credits @egidijusz]0.5.0
— all callbacks buton_transition/4
are optional, acceptimpl_for:
param touse Finitomata
0.4.0
— allow anonymous FSM instances0.3.0
—en_entry/2
andon_exit/2
optional callbacks0.2.0
—Mermaid
support