fond
Model Checking Contest 2024
14th edition, Geneva, Switzerland, June 25, 2024
Execution of r030-smll-171620172000412
Last Updated
July 7, 2024

About the Execution of Tapaal for BART-COL-030

Execution Summary
Max Memory
Used (MB)
Time wait (ms) CPU Usage (ms) I/O Wait (ms) Computed Result Execution
Status
2913.840 3592708.00 4956058.00 9777.60 TFFFFFF?FFFTTFFF normal

Execution Chart

We display below the execution chart for this examination (boot time has been removed).

Trace from the execution

Formatting '/data/fkordon/mcc2024-input.r030-smll-171620172000412.qcow2', fmt=qcow2 size=4294967296 backing_file=/data/fkordon/mcc2024-input.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off refcount_bits=16
Waiting for the VM to be ready (probing ssh)
..........................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................
=====================================================================
Generated by BenchKit 2-5568
Executing tool tapaal
Input is BART-COL-030, examination is LTLFireability
Time confinement is 3600 seconds
Memory confinement is 16384 MBytes
Number of cores is 4
Run identifier is r030-smll-171620172000412
=====================================================================

--------------------
preparation of the directory to be used:
/home/mcc/execution
total 708K
-rw-r--r-- 1 mcc users 7.1K Apr 12 23:29 CTLCardinality.txt
-rw-r--r-- 1 mcc users 65K Apr 12 23:29 CTLCardinality.xml
-rw-r--r-- 1 mcc users 6.2K Apr 12 21:42 CTLFireability.txt
-rw-r--r-- 1 mcc users 54K Apr 12 21:42 CTLFireability.xml
-rw-r--r-- 1 mcc users 4.2K May 18 16:42 GenericPropertiesDefinition.xml
-rw-r--r-- 1 mcc users 6.4K May 18 16:42 GenericPropertiesVerdict.xml
-rw-r--r-- 1 mcc users 4.1K Apr 22 14:29 LTLCardinality.txt
-rw-r--r-- 1 mcc users 26K Apr 22 14:29 LTLCardinality.xml
-rw-r--r-- 1 mcc users 2.3K Apr 22 14:29 LTLFireability.txt
-rw-r--r-- 1 mcc users 16K Apr 22 14:29 LTLFireability.xml
-rw-r--r-- 1 mcc users 17K Apr 13 03:10 ReachabilityCardinality.txt
-rw-r--r-- 1 mcc users 162K Apr 13 03:10 ReachabilityCardinality.xml
-rw-r--r-- 1 mcc users 9.7K Apr 13 00:35 ReachabilityFireability.txt
-rw-r--r-- 1 mcc users 75K Apr 13 00:35 ReachabilityFireability.xml
-rw-r--r-- 1 mcc users 1.7K Apr 22 14:29 UpperBounds.txt
-rw-r--r-- 1 mcc users 3.7K Apr 22 14:29 UpperBounds.xml
-rw-r--r-- 1 mcc users 5 May 18 16:42 equiv_pt
-rw-r--r-- 1 mcc users 4 May 18 16:42 instance
-rw-r--r-- 1 mcc users 5 May 18 16:42 iscolored
-rw-r--r-- 1 mcc users 201K May 18 16:42 model.pnml

--------------------
content from stdout:

=== Data for post analysis generated by BenchKit (invocation template)

The expected result is a vector of booleans
BOOL_VECTOR

here is the order used to build the result vector(from text file)
FORMULA_NAME BART-COL-030-LTLFireability-00
FORMULA_NAME BART-COL-030-LTLFireability-01
FORMULA_NAME BART-COL-030-LTLFireability-02
FORMULA_NAME BART-COL-030-LTLFireability-03
FORMULA_NAME BART-COL-030-LTLFireability-04
FORMULA_NAME BART-COL-030-LTLFireability-05
FORMULA_NAME BART-COL-030-LTLFireability-06
FORMULA_NAME BART-COL-030-LTLFireability-07
FORMULA_NAME BART-COL-030-LTLFireability-08
FORMULA_NAME BART-COL-030-LTLFireability-09
FORMULA_NAME BART-COL-030-LTLFireability-10
FORMULA_NAME BART-COL-030-LTLFireability-11
FORMULA_NAME BART-COL-030-LTLFireability-12
FORMULA_NAME BART-COL-030-LTLFireability-13
FORMULA_NAME BART-COL-030-LTLFireability-14
FORMULA_NAME BART-COL-030-LTLFireability-15

=== Now, execution of the tool begins

BK_START 1716499474455

tapaal
Got BK_BIN_PATH=/home/mcc/BenchKit/bin/
---> tapaal --- TAPAAL - v2.2024
Setting MODEL_PATH=.
Setting VERIFYPN=/home/mcc/BenchKit/bin/verifypn
Got BK_TIME_CONFINEMENT=3600
Setting TEMPDIR=/home/mcc/BenchKit/bin/tmp
Got BK_MEMORY_CONFINEMENT=16384
Limiting to 16265216 kB
Total timeout: 3590
Time left: 3590

*************************************
* TAPAAL verifying LTLFireability *
*************************************
TEMPDIR=/home/mcc/BenchKit/bin/tmp
QF=/home/mcc/BenchKit/bin/tmp/tmp.gfOH3Fiagh
MF=/home/mcc/BenchKit/bin/tmp/tmp.GhBUyQlYGx
Time left: 3590
---------------------------------------------------
Step -1: Stripping Colors
---------------------------------------------------
Verifying stripped models (16 in total)

Time left: 3590
---------------------------------------------------
Step 0: Parallel Simplification
---------------------------------------------------
Doing parallel simplification (16 in total)
Total simplification timout is 718 -- reduction timeout is 299
timeout 1017 /home/mcc/BenchKit/bin/verifypn -n -ltl -q 718 -l 29 -d 299 -z 4 -s OverApprox --binary-query-io 2 --write-simplified /home/mcc/BenchKit/bin/tmp/tmp.gfOH3Fiagh --write-reduced /home/mcc/BenchKit/bin/tmp/tmp.GhBUyQlYGx -x 1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16 ./model.pnml ./LTLFireability.xml

FORMULA BART-COL-030-LTLFireability-00 TRUE TECHNIQUES COLLATERAL_PROCESSING STRUCTURAL_REDUCTION QUERY_REDUCTION SAT_SMT LP_APPROX UNFOLDING_TO_PT
Query index 0 was solved

Query is satisfied.

FORMULA BART-COL-030-LTLFireability-03 FALSE TECHNIQUES COLLATERAL_PROCESSING STRUCTURAL_REDUCTION QUERY_REDUCTION SAT_SMT LP_APPROX UNFOLDING_TO_PT
Query index 3 was solved

Query is NOT satisfied.

FORMULA BART-COL-030-LTLFireability-04 FALSE TECHNIQUES COLLATERAL_PROCESSING STRUCTURAL_REDUCTION QUERY_REDUCTION SAT_SMT LP_APPROX UNFOLDING_TO_PT
Query index 4 was solved

Query is NOT satisfied.
Solution found by parallel simplification (step 0)
Solution found by parallel simplification (step 0)
Solution found by parallel simplification (step 0)
Time left: 3435

---------------------------------------------------
Step 1: Parallel processing
---------------------------------------------------
Doing parallel verification of individual queries (13 in total)
Each query is verified by 4 parallel strategies for 299 seconds
------------------- QUERY 1 ----------------------
Solution found by parallel processing (step 1)

Query index 0 was solved
Query is NOT satisfied.
Spent 0.090074 on verification
@@@0.40,157856@@@
parallel: This job succeeded:
eval /usr/bin/time -f "@@@%e,%M@@@" /home/mcc/BenchKit/bin/verifypn -n '-ltl tarjan --ltl-heur dist -q 0 -l 0 -d 119' /home/mcc/BenchKit/bin/tmp/tmp.GhBUyQlYGx /home/mcc/BenchKit/bin/tmp/tmp.gfOH3Fiagh --binary-query-io 1 -x 1 -n

FORMULA BART-COL-030-LTLFireability-08 FALSE TECHNIQUES EXPLICIT TARJAN WEAK_SKIP STUBBORN AUT_STUB HEURISTIC DIST_HEUR OPTIM-1
Time left: 3434
------------------- QUERY 2 ----------------------
Solution found by parallel processing (step 1)

Query index 0 was solved
Query is satisfied.
Spent 0.005647 on verification
@@@0.30,29908@@@
parallel: This job succeeded:
eval /usr/bin/time -f "@@@%e,%M@@@" /home/mcc/BenchKit/bin/verifypn -n '-ltl ndfs -q 0 -l 0 -d 119' /home/mcc/BenchKit/bin/tmp/tmp.GhBUyQlYGx /home/mcc/BenchKit/bin/tmp/tmp.gfOH3Fiagh --binary-query-io 1 -x 2 -n

FORMULA BART-COL-030-LTLFireability-11 TRUE TECHNIQUES EXPLICIT NDFS WEAK_SKIP HEURISTIC AUTOMATON_HEUR OPTIM-1
Time left: 3433
------------------- QUERY 3 ----------------------
Solution found by parallel processing (step 1)

Query index 0 was solved
Query is NOT satisfied.
Spent 0.063448 on verification
@@@0.71,157756@@@
parallel: This job succeeded:
eval /usr/bin/time -f "@@@%e,%M@@@" /home/mcc/BenchKit/bin/verifypn -n '-ltl tarjan --ltl-heur fire-count -q 0 -l 0 -d 119' /home/mcc/BenchKit/bin/tmp/tmp.GhBUyQlYGx /home/mcc/BenchKit/bin/tmp/tmp.gfOH3Fiagh --binary-query-io 1 -x 3 -n

FORMULA BART-COL-030-LTLFireability-02 FALSE TECHNIQUES EXPLICIT TARJAN WEAK_SKIP STUBBORN AUT_STUB HEURISTIC LOGFIRECOUNT_HEUR(5000) OPTIM-1
Time left: 3432
------------------- QUERY 4 ----------------------
Solution found by parallel processing (step 1)

Query index 0 was solved
Query is NOT satisfied.
Spent 0.062689 on verification
@@@0.66,157796@@@
parallel: This job succeeded:
eval /usr/bin/time -f "@@@%e,%M@@@" /home/mcc/BenchKit/bin/verifypn -n '-ltl tarjan --ltl-heur aut -s DFS -p -q 0 -l 0 -d 119' /home/mcc/BenchKit/bin/tmp/tmp.GhBUyQlYGx /home/mcc/BenchKit/bin/tmp/tmp.gfOH3Fiagh --binary-query-io 1 -x 4 -n

FORMULA BART-COL-030-LTLFireability-15 FALSE TECHNIQUES EXPLICIT TARJAN WEAK_SKIP OPTIM-1
Time left: 3431
------------------- QUERY 5 ----------------------
Solution found by parallel processing (step 1)

Query index 0 was solved
Query is NOT satisfied.
Spent 0.038413 on verification
@@@0.21,30016@@@
parallel: This job succeeded:
eval /usr/bin/time -f "@@@%e,%M@@@" /home/mcc/BenchKit/bin/verifypn -n '-ltl ndfs -q 0 -l 0 -d 119' /home/mcc/BenchKit/bin/tmp/tmp.GhBUyQlYGx /home/mcc/BenchKit/bin/tmp/tmp.gfOH3Fiagh --binary-query-io 1 -x 5 -n

FORMULA BART-COL-030-LTLFireability-06 FALSE TECHNIQUES EXPLICIT NDFS WEAK_SKIP HEURISTIC AUTOMATON_HEUR OPTIM-1
Time left: 3431
------------------- QUERY 6 ----------------------
Solution found by parallel processing (step 1)

Query index 0 was solved
Query is satisfied.
Spent 1.05522 on verification
@@@1.27,158080@@@
parallel: This job succeeded:
eval /usr/bin/time -f "@@@%e,%M@@@" /home/mcc/BenchKit/bin/verifypn -n '-ltl tarjan --ltl-heur aut -s DFS -p -q 0 -l 0 -d 119' /home/mcc/BenchKit/bin/tmp/tmp.GhBUyQlYGx /home/mcc/BenchKit/bin/tmp/tmp.gfOH3Fiagh --binary-query-io 1 -x 6 -n

FORMULA BART-COL-030-LTLFireability-12 TRUE TECHNIQUES EXPLICIT TARJAN WEAK_SKIP OPTIM-1
Time left: 3429
------------------- QUERY 7 ----------------------
Solution found by parallel processing (step 1)

Query index 0 was solved
Query is NOT satisfied.
Spent 0.070616 on verification
@@@0.26,157724@@@
parallel: This job succeeded:
eval /usr/bin/time -f "@@@%e,%M@@@" /home/mcc/BenchKit/bin/verifypn -n '-ltl tarjan --ltl-heur aut -s DFS -p -q 0 -l 0 -d 119' /home/mcc/BenchKit/bin/tmp/tmp.GhBUyQlYGx /home/mcc/BenchKit/bin/tmp/tmp.gfOH3Fiagh --binary-query-io 1 -x 7 -n

FORMULA BART-COL-030-LTLFireability-09 FALSE TECHNIQUES EXPLICIT TARJAN WEAK_SKIP OPTIM-1
Time left: 3429
------------------- QUERY 8 ----------------------
Solution found by parallel processing (step 1)

Query index 0 was solved
Query is NOT satisfied.
Spent 0.062991 on verification
@@@0.25,157672@@@
parallel: This job succeeded:
eval /usr/bin/time -f "@@@%e,%M@@@" /home/mcc/BenchKit/bin/verifypn -n '-ltl tarjan --ltl-heur aut -s DFS -p -q 0 -l 0 -d 119' /home/mcc/BenchKit/bin/tmp/tmp.GhBUyQlYGx /home/mcc/BenchKit/bin/tmp/tmp.gfOH3Fiagh --binary-query-io 1 -x 8 -n

FORMULA BART-COL-030-LTLFireability-13 FALSE TECHNIQUES EXPLICIT TARJAN WEAK_SKIP OPTIM-1
Time left: 3428
------------------- QUERY 9 ----------------------
Solution found by parallel processing (step 1)

Query index 0 was solved
Query is NOT satisfied.
Spent 0.069179 on verification
@@@0.27,158172@@@
parallel: This job succeeded:
eval /usr/bin/time -f "@@@%e,%M@@@" /home/mcc/BenchKit/bin/verifypn -n '-ltl tarjan --ltl-heur aut -s DFS -p -q 0 -l 0 -d 119' /home/mcc/BenchKit/bin/tmp/tmp.GhBUyQlYGx /home/mcc/BenchKit/bin/tmp/tmp.gfOH3Fiagh --binary-query-io 1 -x 9 -n

FORMULA BART-COL-030-LTLFireability-10 FALSE TECHNIQUES EXPLICIT TARJAN WEAK_SKIP OPTIM-1
Time left: 3428
------------------- QUERY 10 ----------------------
Solution found by parallel processing (step 1)

Query index 0 was solved
Query is NOT satisfied.
Spent 0.020562 on verification
@@@0.22,30268@@@
parallel: This job succeeded:
eval /usr/bin/time -f "@@@%e,%M@@@" /home/mcc/BenchKit/bin/verifypn -n '-ltl ndfs -q 0 -l 0 -d 119' /home/mcc/BenchKit/bin/tmp/tmp.GhBUyQlYGx /home/mcc/BenchKit/bin/tmp/tmp.gfOH3Fiagh --binary-query-io 1 -x 10 -n

FORMULA BART-COL-030-LTLFireability-05 FALSE TECHNIQUES EXPLICIT NDFS WEAK_SKIP HEURISTIC AUTOMATON_HEUR OPTIM-1
Time left: 3427
------------------- QUERY 11 ----------------------
Solution found by parallel processing (step 1)

Query index 0 was solved
Query is NOT satisfied.
Spent 0.081827 on verification
@@@0.25,157980@@@
parallel: This job succeeded:
eval /usr/bin/time -f "@@@%e,%M@@@" /home/mcc/BenchKit/bin/verifypn -n '-ltl tarjan --ltl-heur aut -s DFS -p -q 0 -l 0 -d 119' /home/mcc/BenchKit/bin/tmp/tmp.GhBUyQlYGx /home/mcc/BenchKit/bin/tmp/tmp.gfOH3Fiagh --binary-query-io 1 -x 11 -n

FORMULA BART-COL-030-LTLFireability-14 FALSE TECHNIQUES EXPLICIT TARJAN WEAK_SKIP OPTIM-1
Time left: 3427
------------------- QUERY 12 ----------------------
Solution found by parallel processing (step 1)

Query index 0 was solved
Query is NOT satisfied.
Spent 0.064743 on verification
@@@0.25,158076@@@
parallel: This job succeeded:
eval /usr/bin/time -f "@@@%e,%M@@@" /home/mcc/BenchKit/bin/verifypn -n '-ltl tarjan --ltl-heur aut -s DFS -p -q 0 -l 0 -d 119' /home/mcc/BenchKit/bin/tmp/tmp.GhBUyQlYGx /home/mcc/BenchKit/bin/tmp/tmp.gfOH3Fiagh --binary-query-io 1 -x 12 -n

FORMULA BART-COL-030-LTLFireability-01 FALSE TECHNIQUES EXPLICIT TARJAN WEAK_SKIP OPTIM-1
Time left: 3426
------------------- QUERY 13 ----------------------
No solution found
parallel: Warning: This job was killed because it timed out:
parallel: Warning: eval /usr/bin/time -f "@@@%e,%M@@@" /home/mcc/BenchKit/bin/verifypn -n '-ltl tarjan --ltl-heur dist -q 0 -l 0 -d 119' /home/mcc/BenchKit/bin/tmp/tmp.GhBUyQlYGx /home/mcc/BenchKit/bin/tmp/tmp.gfOH3Fiagh --binary-query-io 1 -x 13 -n
parallel: Warning: This job was killed because it timed out:
parallel: Warning: eval /usr/bin/time -f "@@@%e,%M@@@" /home/mcc/BenchKit/bin/verifypn -n '-ltl tarjan --ltl-heur fire-count -q 0 -l 0 -d 119' /home/mcc/BenchKit/bin/tmp/tmp.GhBUyQlYGx /home/mcc/BenchKit/bin/tmp/tmp.gfOH3Fiagh --binary-query-io 1 -x 13 -n
parallel: Warning: This job was killed because it timed out:
parallel: Warning: eval /usr/bin/time -f "@@@%e,%M@@@" /home/mcc/BenchKit/bin/verifypn -n '-ltl tarjan --ltl-heur aut -s DFS -p -q 0 -l 0 -d 119' /home/mcc/BenchKit/bin/tmp/tmp.GhBUyQlYGx /home/mcc/BenchKit/bin/tmp/tmp.gfOH3Fiagh --binary-query-io 1 -x 13 -n
parallel: Warning: This job was killed because it timed out:
parallel: Warning: eval /usr/bin/time -f "@@@%e,%M@@@" /home/mcc/BenchKit/bin/verifypn -n '-ltl ndfs -q 0 -l 0 -d 119' /home/mcc/BenchKit/bin/tmp/tmp.GhBUyQlYGx /home/mcc/BenchKit/bin/tmp/tmp.gfOH3Fiagh --binary-query-io 1 -x 13 -n


Time left: 3125
---------------------------------------------------
Step 2: Sequential processing
---------------------------------------------------
Remaining 1 queries are verified sequentially.
Each query is verified for a dynamic timeout (at least 598 seconds)
Time left: 3125
------------------- QUERY 13 ----------------------
Running query 13 for 3125 seconds. Remaining: 1 queries and 3125 seconds
No solution found
parallel: Warning: This job was killed because it timed out:
parallel: Warning: eval /usr/bin/time -f "@@@%e,%M@@@" /home/mcc/BenchKit/bin/verifypn -n '-ltl tarjan -q 40 -l 5 -d 119 -s BestFS --ltl-por automaton --ltl-heur aut' /home/mcc/BenchKit/bin/tmp/tmp.GhBUyQlYGx /home/mcc/BenchKit/bin/tmp/tmp.gfOH3Fiagh --binary-query-io 1 -x 13 -n


Time left: -3
Out of time, terminating!
terminated-with-cleanup

BK_STOP 1716503067163

--------------------
content from stderr:

ERROR: A query could not be translated from CTL* to CTL.

Sequence of Actions to be Executed by the VM

This is useful if one wants to reexecute the tool in the VM from the submitted image disk.

set -x
# this is for BenchKit: configuration of major elements for the test
export BK_INPUT="BART-COL-030"
export BK_EXAMINATION="LTLFireability"
export BK_TOOL="tapaal"
export BK_RESULT_DIR="/tmp/BK_RESULTS/OUTPUTS"
export BK_TIME_CONFINEMENT="3600"
export BK_MEMORY_CONFINEMENT="16384"
export BK_BIN_PATH="/home/mcc/BenchKit/bin/"

# this is specific to your benchmark or test

export BIN_DIR="$HOME/BenchKit/bin"

# remove the execution directoty if it exists (to avoid increse of .vmdk images)
if [ -d execution ] ; then
rm -rf execution
fi

# this is for BenchKit: explicit launching of the test
echo "====================================================================="
echo " Generated by BenchKit 2-5568"
echo " Executing tool tapaal"
echo " Input is BART-COL-030, examination is LTLFireability"
echo " Time confinement is $BK_TIME_CONFINEMENT seconds"
echo " Memory confinement is 16384 MBytes"
echo " Number of cores is 4"
echo " Run identifier is r030-smll-171620172000412"
echo "====================================================================="
echo
echo "--------------------"
echo "preparation of the directory to be used:"

tar xzf /home/mcc/BenchKit/INPUTS/BART-COL-030.tgz
mv BART-COL-030 execution
cd execution
if [ "LTLFireability" = "ReachabilityDeadlock" ] || [ "LTLFireability" = "UpperBounds" ] || [ "LTLFireability" = "QuasiLiveness" ] || [ "LTLFireability" = "StableMarking" ] || [ "LTLFireability" = "Liveness" ] || [ "LTLFireability" = "OneSafe" ] || [ "LTLFireability" = "StateSpace" ]; then
rm -f GenericPropertiesVerdict.xml
fi
pwd
ls -lh

echo
echo "--------------------"
echo "content from stdout:"
echo
echo "=== Data for post analysis generated by BenchKit (invocation template)"
echo
if [ "LTLFireability" = "UpperBounds" ] ; then
echo "The expected result is a vector of positive values"
echo NUM_VECTOR
elif [ "LTLFireability" != "StateSpace" ] ; then
echo "The expected result is a vector of booleans"
echo BOOL_VECTOR
else
echo "no data necessary for post analysis"
fi
echo
if [ -f "LTLFireability.txt" ] ; then
echo "here is the order used to build the result vector(from text file)"
for x in $(grep Property LTLFireability.txt | cut -d ' ' -f 2 | sort -u) ; do
echo "FORMULA_NAME $x"
done
elif [ -f "LTLFireability.xml" ] ; then # for cunf (txt files deleted;-)
echo echo "here is the order used to build the result vector(from xml file)"
for x in $(grep '' LTLFireability.xml | cut -d '>' -f 2 | cut -d '<' -f 1 | sort -u) ; do
echo "FORMULA_NAME $x"
done
elif [ "LTLFireability" = "ReachabilityDeadlock" ] || [ "LTLFireability" = "QuasiLiveness" ] || [ "LTLFireability" = "StableMarking" ] || [ "LTLFireability" = "Liveness" ] || [ "LTLFireability" = "OneSafe" ] ; then
echo "FORMULA_NAME LTLFireability"
fi
echo
echo "=== Now, execution of the tool begins"
echo
echo -n "BK_START "
date -u +%s%3N
echo
timeout -s 9 $BK_TIME_CONFINEMENT bash -c "/home/mcc/BenchKit/BenchKit_head.sh 2> STDERR ; echo ; echo -n \"BK_STOP \" ; date -u +%s%3N"
if [ $? -eq 137 ] ; then
echo
echo "BK_TIME_CONFINEMENT_REACHED"
fi
echo
echo "--------------------"
echo "content from stderr:"
echo
cat STDERR ;