Modeling Process. Rich Hickey

Transcription

1 Modeling Process Rich Hickey

2 Which are more fundamental? Messages, classes, encapsulation, inheritance, dispatch... Time, value, identity, state, persistence, transience, place, perception, visibility, memory, process...

3 Coming to Terms Value An immutable magnitude, quantity, number... or immutable composite thereof Identity A putative entity we associate with a series of causally related values (states) over time State Value of an identity at a Time moment in time Relative before/after ordering of causal values

4 A Real Problem

5 Place open space Relative Include time coordinate, and process results happen in new places

6 What Would a Program Do? Not this!

7 Are places in charge?

8 What do we see?

9 Our Problem Reality Perception Memory Logic

10 Perception Perceive - take entirely Sensory systems only ever perceive the past Discretizing, snapshots Most useful when coupled with memory idelity matters Visible == can be perceived not merely reachable via reference

11 Memory mindful, remembering If our mental memory behaved the way we use computer memory, we d be ill In the mind we talk about forming memories New memories about the same identities don t replace the old idelity matters Stability matters (persistence)

12 Program Memory Sometimes we use computer memory like brain memory Sometimes like perception Sometimes (commonly, most of the time) like places

13 Using the same memory for everything Destroys the past Corrupts remembering Interferes with perception We must use memory for all three things, but not necessarily the same memory

14 Process

15 Process go forward, advance They re not called food calculators Potentially richer than this Manipulate contents of place(s) May involve multiple forces

16 Process across multiple places

17 Process with multiple forces/participants

18 Philosophy Things don't change in place Becomes obvious once you incorporate time as a dimension Place includes time The future is a (multi-force) function of the past Co-located entities can observe each other without cooperation Coordination is desirable in local context

19 Epochal Time Model Process events (pure functions*) Identity (succession of states) States (immutable values) Observers/perception/memory

20 Persistent lasting or enduring tenaciously Root: to stand firm permanently When applied to data structures A) safe on the disk (not today s topic) B) immutable++ Great fit for perceptions and memories

21 Identity Constructs as Gatekeepers of Time Responsible for coherent successive states Multiple semantics possible And providing proper values to observers Support coordination (multiple places) and process functions supplied from multiple threads of control (multiple participants)

22 unctional Model n + args v2 v3

23 CAS as Time Construct vn vn+1 vns v2 v3 v4 AtomicReference 23 (swap! an-atom f args) (f vn args) becomes vn+1 - can automate spin 1:1 timeline/identity Point-in-time value Atomic state succession perception

24 Agents as Time Construct vn vn+1 (send aref f args) returns immediately vns queue enforces serialization 24 (f vn args) becomes vn+1 happens asynchronously in thread pool thread 1:1 timeline/identity Atomic state succession Point-in-time value perception

25 STM as Time Construct v1 v2 v3 v4 v1 v2 v3 v4 v1 v2 v3 v4 v1 v2 v3 v4 Transactions 25

26 But... What if my logical unit of work involves a million steps? Creating a million interim values via pure function invocation is a waste I m going back to my cubbyholes!

27 Transient not lasting, enduring, or permanent; transitory Root: go across When applied to data structures: Not persistent! Each operation returns the next transient Can t presume modify-in-place Doesn t preclude it either No identity

28 Transient-Based Model Proc + args Proc + args Make Transient (in new memory) can be O(1) Make Persistent (immutable) can be O(1) v2 atransient atransient atransient v3

29 What about those Procs? Might modify their arguments Isn t this just icky mutable side-effecting coding again? hard to test difficult to reason about No!

30 Proc unction of transient to transient Like pure function, can t effect the world nor be effected by it Only used in a context where transient cannot leak Can always be sandwiched in value->transient and transient->value functions and become pure Pure unction value V->T Proc T->V value

31 The sweet, creamy, efficient middle of pure functions

32 Transient/Proc Model Prototype implementation - pods Can support multiple participants, in multiple threads and coordination of multiple identities/places even ad hoc grouping But not arbitrary composition/nesting Same limitation as locks, but detectable Agents could support as well

33 Summary We need to talk about these things Better, more precise language and terms Language and library support Examine high-level abstractions and constructs in terms of these fundamental issues

34 Thanks for listening!