Re: [tlaplus] Temporal logic

OK thanks. That confirms I am looking in the right direction then.

I realised last night I could partly simplify the proof of pc \in ValidPC if I prove the invariant stack = << >> \/ Head(stack).pc \in Valid PC, which should be straightforward. But then I still need to be able to prove stack # << >> in states that return, which is where I think I will need more complex temporal logic. Essentially I need to be able to prove that state X came from A, B or C and in A, B, C stack' # << >> therefore state X stack # << >>.

I have attached my work, it's in a bit of a mess but <4>2 is where I am working.

Regards,
Chris.

On Wednesday, November 12, 2014 11:51:28 PM UTC-8, Stephan Merz wrote:

Hi Chris,

reasoning about ENABLED is still not supported by TLAPS, unfortunately. For the moment you'll have to assert statements like step <1>2 in your proof of z1_liveness. Moreover, temporal reasoning beyond propositional temporal logic is not yet possible, and you will typically need more complex temporal reasoning, such as combining leads-to and well-founded orderings, for realistic liveness properties. Full support for temporal reasoning, and then for ENABLED, is on the top of our to-do list.

Concerning your first question, proving type correctness for your MergeSort algorithm should not be difficult in principle. Of course, you'll need type assertions for the complete state space – for a recursive function, this includes a suitable type correctness predicate for the call stack (which contains the return address from a recursive call). I'm afraid I cannot be more specific without seeing the spec.

Best,
Stephan

On 13 Nov 2014, at 02:49, chri...@xxxxxxxxx wrote:

Hi, two questions:

Firstly I implemented a model for MergeSort, based on the BubbleSort example. It is a recursive function so I am using PlusCal procedure and calling it recursively. It model checks OK with TLC.

I am having trouble proving my TypeOK properties I assume this is because following the data flow into the function call requires some proof using temporal properties. For example, in states which return, I cannot prove pc \in ValidPC. Is this expected to be a hard problem? I can't find any examples that use recursive function calls, and very few that involve this type of temporal proof.

Secondly, I am finding it harder than expected to prove temporal properties. I have traced it down and it seems that I sometimes need to assert ENABLED without proof. I notice a comment in one of the Paxos proofs that TLAPS cannot reason about ENABLED, is this currently true?

My test example follows, note z1_liveness <1>2:

VARIABLE z1

z1_init == z1 = 1
z1_next == z1' = z1 + 1
z1_spec == z1_init /\ [][z1_next]_z1 /\ WF_z1(z1_next)

z1_inv == z1 \in Int

THEOREM z1_safety == z1_spec => []z1_inv
<1>1. z1_init => z1_inv
    BY SMT DEF z1_spec, z1_init, z1_inv
<1>2. z1_inv /\ [z1_next]_z1 => z1_inv'
    BY SMT DEF z1_inv, z1_next
<1> QED
    BY <1>1, <1>2, PTL DEF z1_spec

THEOREM z1_liveness == z1_spec => (<> <<z1_next>>_z1)
<1> SUFFICES ASSUME z1_spec PROVE (<> <<z1_next>>_z1)
    BY PTL
<1>1. []z1_inv
    BY z1_safety, PTL
<1>2. [](ENABLED <<z1_next>>_z1)
    (**** BY SMT DEF z1_spec, z1_init, z1_next ****) PROOF OMITTED
<1>3. []([]ENABLED <<z1_next>>_z1 => (<> <<z1_next>>_z1))
    BY PTL DEF z1_spec
<1> QED
    BY <1>1, <1>2, <1>3, PTL DEF z1_spec, z1_inv, z1_next, z1_init

--
You received this message because you are subscribed to the Google Groups "tlaplus" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tlaplu...@xxxxxxxxxxxxxxxx.
To post to this group, send email to tla...@xxxxxxxxxxxxxxxx.
Visit this group at http://groups.google.com/group/tlaplus.
For more options, visit https://groups.google.com/d/optout.

--
Stephan Merz