Pan Verification Options



Verification Options


pan - Spin generated source for a model-specific verifier.

Overview of options that are available with the verifiers generated by Spin with Spin's run-time option -a. There are two groups of options: those that are available after the verifier source in pan.c has been compiled, and those that are available at compilation time. The reason for the split is that knowledge of some options at compile time can be used to produce a more efficient verification system. Attached is also a brief explanation of the numbers that are printed by the verifiers at the end of a run.

A Run-Time Options for Pan

  • -a
    find acceptance cycles (available if compiled without -DNP)
  • -cN
    stop at Nth error (defaults to first error if N is absent)
  • -d
    print state tables and stop
  • -e
    create trails for all errors encountered (default is first one only)
  • -f
    add weak fairness (to -a or -l)
  • -hN
    choose another hash-function, with N: 1..32 (defaults to 1)
  • -i
    search for shortest path to error (causes an increase of complexity)
  • -I
    like -i, but approximate and faster
  • -J reverse the evaluation order of nested unless statements (to conform to the one used in Java)
  • -l
    find non-progress cycles (requires compilation with -DNP)
  • -mN
    set max search depth to N steps (default N=10000)
  • -n
    no listing of unreached states at the end of the run
  • -q
    require empty channels in valid endstates
  • -s
    use 1-bit hashing (default is 2-bit hashing, assumes compilation -DBITSTATE)
  • -V
    print Spin version number and stop
  • -wN
    use ahashtable of 2^N entries(defaults to -w18)

B Compile Time Options for Spin and Pan

The directives are grouped in eight sets, depending on their main purpose. There are four directives that can be used for compiling the Spin sources itself. These should never be needed by Spin users, only (once) by someone recompiling and installing Spin from its sources.

Directives for Compiling Spin itself

NXT if defined, the NEXT operator X can be used in LTL formulae; risky, not compatible with partial order reductions
PC required when compiling Spin on a PC
PRINTF if defined, printf statements in the model are enabled during the verification process (not recommended)
SOLARIS required when compiling Spin on a Solaris system

The next tables give optional directives for compiling the verifiers that are generated by Spin. Traditionally these are stored in a file named pan.c (with a number of dependent files). Usage of the directives below is always optional, and typically of the form:

    $ spin -a spec
    $ cc -o pan -DNOBOUNDCHECK pan.c
Each directive modifies the default behavior of the verifier to achieve a specific effect noted in the tables below.

Directives Supported by Xspin

BITSTATE use supertrace/bitstate instead of exhaustive exploration
MEMCNT=N set upperbound to the amount of memory that can be allocated usage, e.g.: -DMEMCNT=20 for a maximum of 2^20 bytes
MEMLIM=N set upperbound to the true number of Megabytes that can be allocated; usage, e.g.: -DMEMLIM=200 for a maximum of 200 Megabytes (meant to be a simple alternative to MEMCNT)
NOCLAIM exclude the never claim from the verification, if present
NOFAIR disable the code for weak-fairness (is faster)
NOREDUCE disables the partial order reduction algorithm
NP enable non-progress cycle detection (option -l), replacing option -a for acceptance cycle detection
PEG add complexity profiling (transition counts)
SAFETY optimize for the case where no cycle detection is needed (faster, uses less memory, disables both -l and -a)
VAR_RANGES compute the effective value range of variables (restricted to the interval 0..255)
CHECK generate debugging information (see also DEBUG)

Directives Related to Partial Order Reduction

CTL allow only those reductions that are consistent with branching time logics like CTL (i.e., the persistent set contains either one or all transitions)
GLOB_ALPHA consider process death a global action (for compatibility with versions of Spin between 2.8.5 and 2.9.7)
NIBIS apply a small optimization of partial order reduction (sometimes faster, sometimes not...)
NOREDUCE disables the partial order reduction algorithm
XUSAFE disable validity checks of x[rs] assertions (faster, and sometimes useful if the check is too strict, e.g. when channels are passed around as process parameters)

Directives to Increase Speed

NOBOUNDCHECK don't check array bound violations (faster)
NOCOMP don't compress states with fullstate storage (faster, but not compatible with liveness unless -DBITSTATE)
NOFAIR disable the code for weak-fairness (is faster)
NOSTUTTER disable stuttering rules (warning: changes semantics) stuttering rules are the standard way to extend a finite execution sequence into and infinite one, to allow for a consistent interpretation of B\(u"chi acceptance rules
SAFETY optimize for the case where no cycle detection is needed (faster, uses less memory, disables both -l and -a)

Directives to Reduce Memory Use

BITSTATE use supertrace/bitstate instead of exhaustive exploration
HC a state vector compression mode; collapses state vector sizes down to 32+16 bits and stores them in conventional hash-table (a version of Wolper's hash-compact method -- new in version 3.2.2.) Variations: HC0, HC1, HC2, HC3 for 32, 40, 48, or 56 bits respectively. The default is equivalent to HC2.
COLLAPSE a state vector compression mode; collapses state vector sizes by up to 80% to 90% (see Spin97 workshop paper) variations: add -DSEPQS or -DJOINPROCS (off by default)
MA=N use a minimized DFA encoding for the state space, similar to a BDD, assuming a maximum of N bytes in the state-vector (this can be combined with -DCOLLAPSE for greater effect in cases when the original state vector is long)
MEMCNT=N set upperbound to the amount of memory that can be allocated usage, e.g.: -DMEMCNT=20 for a maximum of 2^20 bytes
MEMLIM=N set upperbound to the true number of Megabytes that can be allocated; usage, e.g.: -DMEMLIM=200 for a maximum of 200 Megabytes (meant to be a simple alternative to MEMCNT)
SC enables stack cycling. this will swap parts of a very long search stack to a diskfile during verifications. the runtime flag -m for setting the size of the search stack still remains, but now sets the size of the part of the stack that remains in core. it is meant for rare applications where the search stack is many millions of states deep and eats up the majority of the memory requirements.

Directives Reserved for Use When Prompted by Pan

NFAIR=N allocates memory for enforcing weak fairness usage, e.g.: -DNFAIR=3 (default is 2)
VECTORSZ=N allocates memory (in bytes) for state vector usage, e.g.: -DVECTORSZ=2048 (default is 1024)

Directives For Debugging Pan Verifiers

VERBOSE adds elaborate debugging printouts
CHECK more frugal debugging printouts
SVDUMP if defined, adds an option -pN to the runtime verifiers to produce a file sv_dump at the end of the run, with a binary representation of all states, using a fixed size of N bytes per state. (see also SDUMP below)
SDUMP if used in addition to CHECK: adds ascii dumps of state vectors to verbose output (i.e., an ascii version of SVDUMP)

Directives For Experimental Use

BCOMP when in BITSTATE mode, this computes hash functions over the compressed state-vector (compressed with byte-masking) in some cases, this can improve the coverage
COVEST compiles in extra code for giving a coverage estimate at the end of BITSTATE runs, defined by Uli Stern. must compile pan.c with -lm in this case
HYBRID_HASH can shave a word (4 bytes) from every state vector, in 25% of the cases... when the state vector is 1 byte longer than a multiple of 4 the memory allocator ends up adding 3 dummy bytes to secure memory alignment. to avoid this, the HYBRID_HASH mode will take away 1 byte from the state vector, and use it as part of the hash -- this causes more collisions, but preserves correct exhaustive exploration - and saves 4 bytes per stored state in memory...
LC to be used in combination with BITSTATE hashing only. it is automatically enabled when -DSC is used in BITSTATE mode. LC forces the use of hashcompact compression for stackstates (instead of the dedault which is full-state storage for states while they are on the search stack, even in bitstate mode). it slows down the search, but can save memory. it uses 4 bytes per state (giving very low probability of collision).
NOVSZ risky - removes 4 bytes from state vector - its length field. in most cases this is redundant - so when memory is tight in fullstate storage, try this mode. if the number of states stored changes when -DNOVSZ is used, the information wasn't redundant... (safety checks will still be valid, but liveness checks may then fail) NOVSZ cannot be combined with COLLAPSE
PRINTF enables printfs during verification runs (Version 2.8 and later -- earlier versions always left these enabled)
RANDSTORE when in BITSTATE mode, use for instance -DRANDSTORE=33 to reduce the probability of storing the bits in the hasharray to 33%. the value assigned must be between 0 and 99 low values increase the amount of work done (time complexity) and increase the effective coverage for large state spaces. most useful in sequential bitstate hashing runs to improve the accumulative coverage of all runs significantly
REACH guarantee absence of errors within the -m depth-limit (described in more detail in Newsletter 4 and in the V2.Updates notes for Version 2.2.)
W_XPT=N in combination with MA, write checkpoint files every multiple of N states stored
R_XPT in combination with MA, restart a verification run from the last checkpoint file written, can be combined with W_XPT

C Pan's Output Format

A typical printout of a verification run is as follows: This is what each line in this listing means:
	(Spin Version 3.0.alpha -- 21 July 1997)
Identifies the version of Spin that generated the pan.c source from which this verifier was compiled.
		+ Partial Order Reduction
The plus sign means that the default partial order reduction algorithm was used. A minus sign would indicate compilation for exhaustive, non-reduced, verification with option -DNOREDUCE .
	Full statespace search for:
Indicates the type of search. The default is a full statespace search. Large models can also be verified with a Bitstate search, which is approximate.
		never-claim         	- (none specified)
The minus sign indicates that no never claim, or LTL fomrula was used for this run. If a never claim was part of the model, it could have been suppressed with the compiler directive -DNOCLAIM .
		assertion violations	+
The plus indicates that the search checked for violations of user specified assertions, which is the default.
		acceptance   cycles 	- (not selected)
The minus indicates that the search did not check for the presence of acceptance or non-progress cycles. To do so would require a run-time option -a or compilation with -DNP combined with the run-time option -l.
		invalid endstates	+
The plus indicates that a check for invalid endstates was done (i.e., for absence of deadlocks).
	State-vector 32 byte, depth reached 13, errors: 0
The complete description of a global system state required 32 bytes of memory (per state). The longest depth-first search path contained 13 transitions from the root of the tree (i.e., from the initial system state). No errors were found in this search.
	      74 states, stored
A total of 74 unique global system states were stored in the statespace (each represented effectively by a vector of 32 bytes).
	      30 states, matched
In 30 cases did the search return to a previously visited state in the search tree.
	     104 transitions (= stored+matched)
A total of 104 transitions were explored in the search, which can serve as a statistic for the amount of work that has been performed to complete the verification.
	       1 atomic steps
One of the transitions was part of an atomic sequence, all others were outside atomic sequences.
	hash conflicts: 2 (resolved)
In 2 cases the default hashing scheme (a weaker version than what is used in bitstate hashing) encountered a collision, and had to place the states into a linked list in the hash-table.
	(max size 2^18 states)
The (perhaps default) argument that was specified for the size of the hash-table was 2^18; equivalent to a run-time option -w18.
	1.533 	memory usage (Mbyte)
Total memory usage was 1.533 Megabytes, including the stack, the hashtable, and all related data structures. Memory use would go down for smaller than the default choices for run-time options -m and -w, which could in this case, with only 74 reachable states of 32 bytes each, considerably reduce memory use.
	unreached in proctype ProcA
		line 7, state 8, "Gaap = 4"
		(1 of 13 states)
	unreached in proctype :init:
		line 21, state 14, "Gaap = 3"
		(1 of 19 states)
A listing of the state numbers and approximate line numbers for the basic statements in the specification that were not reached. Since this is a full statespace search that ran to completion this means that these transitions are effectively unreachable (dead code).
Spin Online References
Promela Manual Index
Spin HomePage
(Page Updated: 14 August 2000)