spot/doc/org/tut02.org
Alexandre Duret-Lutz 6c2985e753 fix Python bindings for relabeling_map, and document them
This fixes #61, and addresses one item of #14.

* src/ltlvisit/relabel.hh: Use a map rather than a unordered_map,
because the Swig binding for unordered_map do not seem functional.
* wrap/python/spot_impl.i: Adjust.
* wrap/python/tests/relabel.py: New file.
* wrap/python/tests/Makefile.am: Add it.
* doc/org/tut02.org: New file.
* doc/Makefile.am: Add it.
2015-06-07 14:22:31 +02:00

159 lines
4.3 KiB
Org Mode

#+TITLE: Relabeling Formulas
#+SETUPFILE: setup.org
#+HTML_LINK_UP: tut.html
The task is to read an LTL formula, relabel all (possibly complex)
atomic propositions, and provide =#define= statements for each of
these renamings, writing everything in Spin's syntax.
* Shell
#+BEGIN_SRC sh :results verbatim :exports both
ltlfilt -ps --relabel=pnn --define -f '"Proc@Here" U ("var > 10" | "var < 4")'
#+END_SRC
#+RESULTS:
: #define p0 ((Proc@Here))
: #define p1 ((var < 4))
: #define p2 ((var > 10))
: (p0) U ((p1) || (p2))
When is this output interesting, you may ask? It is useful for
instance if you want to call =ltl2ba= (or any other LTL-to-Büchi
translator) using a formula with complex atomic propositions it cannot
parse. Then you can pass the rewritten formula to =ltl2ba=, and
prepend all those =#define= to its output. For instance:
#+BEGIN_SRC sh :results verbatim :exports both
ltlfilt -ps --relabel=pnn --define=tmp.defs -f '"Proc@Here" U ("var > 10" | "var < 4")' >tmp.ltl
cat tmp.defs; ltl2ba -F tmp.ltl
rm tmp.defs tmp.ltl
#+END_SRC
#+RESULTS:
#+begin_example
#define p0 ((Proc@Here))
#define p1 ((var < 4))
#define p2 ((var > 10))
never { /* (p0) U ((p1) || (p2))
*/
T0_init:
if
:: (p0) -> goto T0_init
:: (p1) || (p2) -> goto accept_all
fi;
accept_all:
skip
}
#+end_example
* Python
The =spot.relabel= function takes an optional third parameter that
should be a =relabeling_map=. If supplied, this map is filled with
pairs of atomic propositions of the form (new-name, old-name).
#+BEGIN_SRC python :results output :exports both
import spot
f = spot.formula('"Proc@Here" U ("var > 10" | "var < 4")')
m = spot.relabeling_map()
g = spot.relabel(f, spot.Pnn, m)
for newname, oldname in m.items():
print("#define {} ({})".format(newname.to_str(), oldname.to_str('spin', True)))
print(g.to_str('spin', True))
#+END_SRC
#+RESULTS:
: #define p0 ((Proc@Here))
: #define p1 ((var < 4))
: #define p2 ((var > 10))
: (p0) U ((p1) || (p2))
* C++
The =spot::ltl::relabeling_map= is just a =std::map= with a custom
destructor.
#+BEGIN_SRC C++ :results verbatim :exports both
#include <string>
#include <iostream>
#include "ltlparse/public.hh"
#include "ltlvisit/print.hh"
#include "ltlvisit/relabel.hh"
int main()
{
std::string input = "\"Proc@Here\" U (\"var > 10\" | \"var < 4\")";
spot::ltl::parse_error_list pel;
const spot::ltl::formula* f = spot::ltl::parse_infix_psl(input, pel);
if (spot::ltl::format_parse_errors(std::cerr, input, pel))
{
if (f)
f->destroy();
return 1;
}
spot::ltl::relabeling_map m;
const spot::ltl::formula* g = spot::ltl::relabel(f, spot::ltl::Pnn, &m);
for (auto& i: m)
{
std::cout << "#define ";
print_psl(std::cout, i.first) << " (";
print_spin_ltl(std::cout, i.second, true) << ")\n";
}
print_spin_ltl(std::cout, g, true) << '\n';
g->destroy();
f->destroy();
return 0;
}
#+END_SRC
#+RESULTS:
: #define p0 ((Proc@Here))
: #define p1 ((var < 4))
: #define p2 ((var > 10))
: (p0) U ((p1) || (p2))
* Additional comments
** Two ways to name atomic propositions
Instead of =--relabel=pnn= (or =spot.Pnn=, or =spot::ltl::Pnn=), you can
actually use =--relabel=abc= (or =spot.Abc=, or =spot::ltl::Abc=) to have
the atomic propositions named =a=, =b=, =c=, etc.
** Relabeling Boolean sub-expressions
Instead of relabeling each atomic proposition, you could decide to
relabel each Boolean sub-expression:
#+BEGIN_SRC sh :results verbatim :exports both
ltlfilt -ps --relabel-bool=pnn --define -f '"Proc@Here" U ("var > 10" | "var < 4")'
#+END_SRC
#+RESULTS:
: #define p0 ((Proc@Here))
: #define p1 ((var < 4) || (var > 10))
: (p0) U (p1)
The relabeling routine is smart enough to not give different names
to Boolean expressions that have some sub-expression in common.
For instance =a U (a & b)= will not be relabeled into =(p0) U (p1)=
because that would hide the fact that both =p0= and =p1= check for
=a=. Instead we get this:
#+BEGIN_SRC sh :results verbatim :exports both
ltlfilt -ps --relabel-bool=pnn --define -f 'a U (a & b)'
#+END_SRC
#+RESULTS:
: #define p0 ((a))
: #define p1 ((b))
: (p0) U ((p0) && (p1))
This "Boolean sub-expression" relabeling is available in Python and
C++ via the =relabel_bse= function. The interface is identical to
=relabel=.