1 (* Title: Tools/quickcheck.ML
2 Author: Stefan Berghofer, Florian Haftmann, Lukas Bulwahn, TU Muenchen
4 Generic counterexample search engine.
9 val setup: theory -> theory
11 val auto: bool Unsynchronized.ref
12 val timing : bool Unsynchronized.ref
13 datatype report = Report of
14 { iterations : int, raised_match_errors : int,
15 satisfied_assms : int list, positive_concl_tests : int }
16 datatype expectation = No_Expectation | No_Counterexample | Counterexample;
17 datatype test_params = Test_Params of
18 { size: int, iterations: int, default_type: typ list, no_assms: bool,
19 expect : expectation, report: bool, quiet : bool, timeout : real};
20 val test_params_of : Proof.context -> test_params
21 val report : Proof.context -> bool
23 ((int * int) * ((typ list * bool) * ((expectation * bool) * (bool * real)))
24 -> (int * int) * ((typ list * bool) * ((expectation * bool) * (bool * real))))
25 -> Context.generic -> Context.generic
27 string * (Proof.context -> term -> int -> term list option * (bool list * bool))
28 -> Context.generic -> Context.generic
29 (* testing terms and proof states *)
30 val test_term_small: Proof.context -> term ->
31 (string * term) list option * ((string * int) list * ((int * report list) list) option)
32 val test_term: Proof.context -> string option -> term ->
33 (string * term) list option * ((string * int) list * ((int * report list) list) option)
34 val quickcheck: (string * string list) list -> int -> Proof.state -> (string * term) list option
37 structure Quickcheck : QUICKCHECK =
42 val auto = Unsynchronized.ref false;
44 val timing = Unsynchronized.ref false;
47 ProofGeneralPgip.add_preference Preferences.category_tracing
48 (Unsynchronized.setmp auto true (fn () =>
49 Preferences.bool_pref auto
51 "Run Quickcheck automatically.") ());
53 (* quickcheck report *)
55 datatype single_report = Run of bool list * bool | MatchExc
57 datatype report = Report of
58 { iterations : int, raised_match_errors : int,
59 satisfied_assms : int list, positive_concl_tests : int }
61 fun collect_single_report single_report
62 (Report {iterations = iterations, raised_match_errors = raised_match_errors,
63 satisfied_assms = satisfied_assms, positive_concl_tests = positive_concl_tests}) =
66 Report {iterations = iterations + 1, raised_match_errors = raised_match_errors + 1,
67 satisfied_assms = satisfied_assms, positive_concl_tests = positive_concl_tests}
68 | Run (assms, concl) =>
69 Report {iterations = iterations + 1, raised_match_errors = raised_match_errors,
71 map2 (fn b => fn s => if b then s + 1 else s) assms
72 (if null satisfied_assms then replicate (length assms) 0 else satisfied_assms),
73 positive_concl_tests = if concl then positive_concl_tests + 1 else positive_concl_tests}
77 datatype expectation = No_Expectation | No_Counterexample | Counterexample;
79 fun merge_expectation (expect1, expect2) =
80 if expect1 = expect2 then expect1 else No_Expectation
82 (* quickcheck configuration -- default parameters, test generators *)
84 datatype test_params = Test_Params of
85 { size: int, iterations: int, default_type: typ list, no_assms: bool,
86 expect : expectation, report: bool, quiet : bool, timeout : real};
89 (Test_Params { size, iterations, default_type, no_assms, expect, report, quiet, timeout }) =
90 ((size, iterations), ((default_type, no_assms), ((expect, report), (quiet, timeout))));
93 ((size, iterations), ((default_type, no_assms), ((expect, report), (quiet, timeout)))) =
94 Test_Params { size = size, iterations = iterations, default_type = default_type,
95 no_assms = no_assms, expect = expect, report = report, quiet = quiet, timeout = timeout };
97 fun map_test_params' f
98 (Test_Params { size, iterations, default_type, no_assms, expect, report, quiet, timeout }) =
100 (f ((size, iterations), ((default_type, no_assms), ((expect, report), (quiet, timeout)))));
102 fun merge_test_params
103 (Test_Params { size = size1, iterations = iterations1, default_type = default_type1,
104 no_assms = no_assms1, expect = expect1, report = report1, quiet = quiet1, timeout = timeout1 },
105 Test_Params { size = size2, iterations = iterations2, default_type = default_type2,
106 no_assms = no_assms2, expect = expect2, report = report2, quiet = quiet2, timeout = timeout2}) =
107 make_test_params ((Int.max (size1, size2), Int.max (iterations1, iterations2)),
108 ((merge (op =) (default_type1, default_type2), no_assms1 orelse no_assms2),
109 ((merge_expectation (expect1, expect2), report1 orelse report2),
110 (quiet1 orelse quiet2, Real.min (timeout1, timeout2)))));
112 structure Data = Generic_Data
115 (string * (Proof.context -> term -> int -> term list option * (bool list * bool))) list
117 val empty = ([], Test_Params
118 { size = 10, iterations = 100, default_type = [], no_assms = false,
119 expect = No_Expectation, report = false, quiet = false, timeout = 30.0});
121 fun merge ((generators1, params1), (generators2, params2)) : T =
122 (AList.merge (op =) (K true) (generators1, generators2),
123 merge_test_params (params1, params2));
126 val test_params_of = snd o Data.get o Context.Proof;
128 val size = fst o fst o dest_test_params o test_params_of
130 val iterations = snd o fst o dest_test_params o test_params_of
132 val default_type = fst o fst o snd o dest_test_params o test_params_of
134 val no_assms = snd o fst o snd o dest_test_params o test_params_of
136 val expect = fst o fst o snd o snd o dest_test_params o test_params_of
138 val report = snd o fst o snd o snd o dest_test_params o test_params_of
140 val quiet = fst o snd o snd o snd o dest_test_params o test_params_of
142 val timeout = snd o snd o snd o snd o dest_test_params o test_params_of
145 (Test_Params { size, iterations, default_type, no_assms, expect, report, quiet, timeout }) =
147 ((size, iterations), ((default_type, no_assms), ((expect, f report), (quiet, timeout))));
150 (Test_Params { size, iterations, default_type, no_assms, expect, report, quiet, timeout }) =
152 ((size, iterations), ((default_type, no_assms), ((expect, report), (f quiet, timeout))));
154 fun set_report report = Data.map (apsnd (map_report (K report)))
156 fun set_quiet quiet = Data.map (apsnd (map_quiet (K quiet)))
158 val map_test_params = Data.map o apsnd o map_test_params'
160 val add_generator = Data.map o apfst o AList.update (op =);
162 (* generating tests *)
164 fun mk_tester_select name ctxt =
165 case AList.lookup (op =) ((fst o Data.get o Context.Proof) ctxt) name
166 of NONE => error ("No such quickcheck generator: " ^ name)
167 | SOME generator => generator ctxt;
169 fun mk_testers ctxt t =
170 (map snd o fst o Data.get o Context.Proof) ctxt
171 |> map_filter (fn generator => try (generator ctxt) t);
173 fun mk_testers_strict ctxt t =
175 val generators = ((map snd o fst o Data.get o Context.Proof) ctxt)
176 val testers = map (fn generator => Exn.interruptible_capture (generator ctxt) t) generators;
178 if forall (is_none o Exn.get_result) testers
179 then [(Exn.release o snd o split_last) testers]
180 else map_filter Exn.get_result testers
184 (* testing propositions *)
186 fun prep_test_term t =
188 val _ = (null (Term.add_tvars t []) andalso null (Term.add_tfrees t [])) orelse
189 error "Term to be tested contains type variables";
190 val _ = null (Term.add_vars t []) orelse
191 error "Term to be tested contains schematic variables";
192 val frees = Term.add_frees t [];
193 in (map fst frees, list_abs_free (frees, t)) end
195 fun cpu_time description f =
197 val start = start_timing ()
198 val result = Exn.capture f ()
199 val time = Time.toMilliseconds (#cpu (end_timing start))
200 in (Exn.release result, (description, time)) end
202 fun test_term_small ctxt t =
204 val (names, t') = prep_test_term t;
205 val current_size = Unsynchronized.ref 0
207 "Quickcheck " ^ s ^ " while testing at size " ^ string_of_int (!current_size)
208 val (tester, comp_time) = cpu_time "compilation"
209 (fn () => the (AList.lookup (op =) ((fst o Data.get o Context.Proof) ctxt) "small") ctxt t');
210 val empty_report = Report { iterations = 0, raised_match_errors = 0,
211 satisfied_assms = [], positive_concl_tests = 0 }
212 fun with_size k timings =
213 if k > size ctxt then (NONE, timings)
216 val _ = if quiet ctxt then () else Output.urgent_message ("Test data size: " ^ string_of_int k);
217 val _ = current_size := k
218 val (result, timing) = cpu_time ("size " ^ string_of_int k) (fn () => (fst (tester k)) handle Match => (if quiet ctxt then ()
219 else warning "Exception Match raised during quickcheck"; NONE))
222 NONE => with_size (k + 1) (timing :: timings)
223 | SOME q => (SOME q, (timing :: timings))
225 val result = with_size 1 [comp_time]
227 apsnd (rpair NONE) (apfst (Option.map (curry (op ~~) names)) result)
230 fun test_term ctxt generator_name t =
232 val (names, t') = prep_test_term t;
233 val current_size = Unsynchronized.ref 0
235 "Quickcheck " ^ s ^ " while testing at size " ^ string_of_int (!current_size)
236 val (testers, comp_time) = cpu_time "quickcheck compilation"
237 (fn () => (case generator_name
238 of NONE => if quiet ctxt then mk_testers ctxt t' else mk_testers_strict ctxt t'
239 | SOME name => [mk_tester_select name ctxt t']));
240 fun iterate f 0 report = (NONE, report)
241 | iterate f j report =
243 val (test_result, single_report) = apsnd Run (f ()) handle Match => (if quiet ctxt then ()
244 else warning "Exception Match raised during quickcheck"; (NONE, MatchExc))
245 val report = collect_single_report single_report report
247 case test_result of NONE => iterate f (j - 1) report | SOME q => (SOME q, report)
249 val empty_report = Report { iterations = 0, raised_match_errors = 0,
250 satisfied_assms = [], positive_concl_tests = 0 }
251 fun with_testers k [] = (NONE, [])
252 | with_testers k (tester :: testers) =
253 case iterate (fn () => tester (k - 1)) (iterations ctxt) empty_report
254 of (NONE, report) => apsnd (cons report) (with_testers k testers)
255 | (SOME q, report) => (SOME q, [report]);
256 fun with_size k reports =
257 if k > size ctxt then (NONE, reports)
259 (if quiet ctxt then () else Output.urgent_message ("Test data size: " ^ string_of_int k);
261 val _ = current_size := k
262 val (result, new_report) = with_testers k testers
263 val reports = ((k, new_report) :: reports)
264 in case result of NONE => with_size (k + 1) reports | SOME q => (SOME q, reports) end);
265 val ((result, reports), exec_time) =
266 TimeLimit.timeLimit (seconds (timeout ctxt)) (fn () => cpu_time "quickcheck execution"
268 (fn result => case result of NONE => NONE
269 | SOME ts => SOME (names ~~ ts)) (with_size 1 []))) ()
270 handle TimeLimit.TimeOut =>
271 error (excipit "ran out of time")
272 | Exn.Interrupt => error (excipit "was interrupted") (* FIXME violates Isabelle/ML exception model *)
274 (result, ([exec_time, comp_time], if report ctxt then SOME reports else NONE))
277 exception WELLSORTED of string
279 fun monomorphic_term thy insts default_T =
281 fun subst (T as TFree (v, S)) =
283 val T' = AList.lookup (op =) insts v
284 |> the_default default_T
285 in if Sign.of_sort thy (T', S) then T'
286 else raise (WELLSORTED ("For instantiation with default_type " ^ Syntax.string_of_typ_global thy default_T ^
287 ":\n" ^ Syntax.string_of_typ_global thy T' ^
288 " to be substituted for variable " ^
289 Syntax.string_of_typ_global thy T ^ " does not have sort " ^
290 Syntax.string_of_sort_global thy S))
293 in (map_types o map_atyps) subst end;
295 datatype wellsorted_error = Wellsorted_Error of string | Term of term
297 fun test_goal generator_name insts i state =
299 val lthy = Proof.context_of state;
300 val thy = Proof.theory_of state;
301 fun strip (Const ("all", _) $ Abs (_, _, t)) = strip t
303 val {goal = st, ...} = Proof.raw_goal state;
304 val (gi, frees) = Logic.goal_params (prop_of st) i;
305 val some_locale = case (Option.map #target o Named_Target.peek) lthy
308 | SOME locale => SOME locale;
309 val assms = if no_assms lthy then [] else case some_locale
310 of NONE => Assumption.all_assms_of lthy
311 | SOME locale => Assumption.local_assms_of lthy (Locale.init locale thy);
312 val proto_goal = Logic.list_implies (map Thm.term_of assms, subst_bounds (frees, strip gi));
313 val check_goals = case some_locale
314 of NONE => [proto_goal]
315 | SOME locale => map (fn (_, phi) => Morphism.term phi proto_goal)
316 (Locale.registrations_of (Context.Theory thy) (*FIXME*) locale);
317 val inst_goals = maps (fn check_goal => map (fn T =>
318 Term ((Object_Logic.atomize_term thy o monomorphic_term thy insts T) check_goal)
319 handle WELLSORTED s => Wellsorted_Error s) (default_type lthy)) check_goals
320 val error_msg = cat_lines (map_filter (fn Term t => NONE | Wellsorted_Error s => SOME s) inst_goals)
321 val correct_inst_goals =
322 case map_filter (fn Term t => SOME t | Wellsorted_Error s => NONE) inst_goals of
323 [] => error error_msg
325 val _ = if quiet lthy then () else warning error_msg
326 fun collect_results f reports [] = (NONE, rev reports)
327 | collect_results f reports (t :: ts) =
329 (SOME res, report) => (SOME res, rev (report :: reports))
330 | (NONE, report) => collect_results f (report :: reports) ts
331 in collect_results (test_term lthy generator_name) [] correct_inst_goals end;
333 (* pretty printing *)
335 fun tool_name auto = (if auto then "Auto " else "") ^ "Quickcheck"
337 fun pretty_counterex ctxt auto NONE = Pretty.str (tool_name auto ^ " found no counterexample.")
338 | pretty_counterex ctxt auto (SOME cex) =
339 Pretty.chunks (Pretty.str (tool_name auto ^ " found a counterexample:\n") ::
341 Pretty.block [Pretty.str (s ^ " ="), Pretty.brk 1, Syntax.pretty_term ctxt t]) cex);
343 fun pretty_report (Report {iterations = iterations, raised_match_errors = raised_match_errors,
344 satisfied_assms = satisfied_assms, positive_concl_tests = positive_concl_tests}) =
346 fun pretty_stat s i = Pretty.block ([Pretty.str (s ^ ": " ^ string_of_int i)])
348 ([pretty_stat "iterations" iterations,
349 pretty_stat "match exceptions" raised_match_errors]
350 @ map_index (fn (i, n) => pretty_stat ("satisfied " ^ string_of_int (i + 1) ^ ". assumption") n)
352 @ [pretty_stat "positive conclusion tests" positive_concl_tests])
355 fun pretty_reports' [report] = [Pretty.chunks (pretty_report report)]
356 | pretty_reports' reports =
357 map_index (fn (i, report) =>
358 Pretty.chunks (Pretty.str (string_of_int (i + 1) ^ ". generator:\n") :: pretty_report report))
361 fun pretty_reports ctxt (SOME reports) =
362 Pretty.chunks (Pretty.str "Quickcheck report:" ::
363 maps (fn (size, reports) =>
364 Pretty.str ("size " ^ string_of_int size ^ ":") :: pretty_reports' reports @ [Pretty.brk 1])
366 | pretty_reports ctxt NONE = Pretty.str ""
368 fun pretty_counterex_and_reports ctxt auto (cex, timing_and_reports) =
369 Pretty.chunks (pretty_counterex ctxt auto cex ::
370 map (pretty_reports ctxt) (map snd timing_and_reports))
372 (* automatic testing *)
374 fun auto_quickcheck state =
379 val ctxt = Proof.context_of state;
382 |> Proof.map_context (Context.proof_map (set_report false #> set_quiet true))
383 |> try (test_goal NONE [] 1);
386 NONE => (false, state)
387 | SOME (NONE, report) => (false, state)
388 | SOME (cex, report) => (true, Proof.goal_message (K (Pretty.chunks [Pretty.str "",
389 Pretty.mark Markup.hilite (pretty_counterex ctxt true cex)])) state)
392 val setup = Auto_Tools.register_tool ("quickcheck", auto_quickcheck)
397 fun read_nat s = case (Library.read_int o Symbol.explode) s
398 of (k, []) => if k >= 0 then k
399 else error ("Not a natural number: " ^ s)
400 | (_, _ :: _) => error ("Not a natural number: " ^ s);
402 fun read_bool "false" = false
403 | read_bool "true" = true
404 | read_bool s = error ("Not a Boolean value: " ^ s)
407 case (Real.fromString s) of
409 | NONE => error ("Not a real number: " ^ s)
411 fun read_expectation "no_expectation" = No_Expectation
412 | read_expectation "no_counterexample" = No_Counterexample
413 | read_expectation "counterexample" = Counterexample
414 | read_expectation s = error ("Not an expectation value: " ^ s)
416 fun parse_test_param ctxt ("size", [arg]) =
417 (apfst o apfst o K) (read_nat arg)
418 | parse_test_param ctxt ("iterations", [arg]) =
419 (apfst o apsnd o K) (read_nat arg)
420 | parse_test_param ctxt ("default_type", arg) =
421 (apsnd o apfst o apfst o K) (map (ProofContext.read_typ ctxt) arg)
422 | parse_test_param ctxt ("no_assms", [arg]) =
423 (apsnd o apfst o apsnd o K) (read_bool arg)
424 | parse_test_param ctxt ("expect", [arg]) =
425 (apsnd o apsnd o apfst o apfst o K) (read_expectation arg)
426 | parse_test_param ctxt ("report", [arg]) =
427 (apsnd o apsnd o apfst o apsnd o K) (read_bool arg)
428 | parse_test_param ctxt ("quiet", [arg]) =
429 (apsnd o apsnd o apsnd o apfst o K) (read_bool arg)
430 | parse_test_param ctxt ("timeout", [arg]) =
431 (apsnd o apsnd o apsnd o apsnd o K) (read_real arg)
432 | parse_test_param ctxt (name, _) =
433 error ("Unknown test parameter: " ^ name);
435 fun parse_test_param_inst ctxt ("generator", [arg]) =
436 (apsnd o apfst o K o SOME) arg
437 | parse_test_param_inst ctxt (name, arg) =
438 case try (ProofContext.read_typ ctxt) name
439 of SOME (TFree (v, _)) => (apsnd o apsnd o AList.update (op =))
440 (v, ProofContext.read_typ ctxt (the_single arg))
441 | _ => (apfst o parse_test_param ctxt) (name, arg);
443 fun quickcheck_params_cmd args thy =
445 val ctxt = ProofContext.init_global thy
446 val f = fold (parse_test_param ctxt) args;
449 |> (Context.theory_map o map_test_params) f
452 fun gen_quickcheck args i state =
454 val ctxt = Proof.context_of state;
455 val default_params = (dest_test_params o snd o Data.get o Context.Proof) ctxt;
456 val f = fold (parse_test_param_inst ctxt) args;
457 val (test_params, (generator_name, insts)) = f (default_params, (NONE, []));
460 |> Proof.map_context (Context.proof_map (map_test_params (K test_params)))
461 |> (fn state' => test_goal generator_name insts i state'
462 |> tap (fn (SOME x, _) => if expect (Proof.context_of state') = No_Counterexample then
463 error ("quickcheck expected to find no counterexample but found one") else ()
464 | (NONE, _) => if expect (Proof.context_of state') = Counterexample then
465 error ("quickcheck expected to find a counterexample but did not find one") else ()))
468 fun quickcheck args i state = fst (gen_quickcheck args i state);
470 fun quickcheck_cmd args i state =
471 gen_quickcheck args i (Toplevel.proof_of state)
472 |> Pretty.writeln o pretty_counterex_and_reports (Toplevel.context_of state) false;
474 val parse_arg = Parse.name -- (Scan.optional (Parse.$$$ "=" |--
475 (((Parse.name || Parse.float_number) >> single) || (Parse.$$$ "[" |-- Parse.list1 Parse.name --| Parse.$$$ "]"))) ["true"]);
477 val parse_args = Parse.$$$ "[" |-- Parse.list1 parse_arg --| Parse.$$$ "]"
481 Outer_Syntax.command "quickcheck_params" "set parameters for random testing" Keyword.thy_decl
482 (parse_args >> (fn args => Toplevel.theory (quickcheck_params_cmd args)));
485 Outer_Syntax.improper_command "quickcheck" "try to find counterexample for subgoal" Keyword.diag
486 (parse_args -- Scan.optional Parse.nat 1
487 >> (fn (args, i) => Toplevel.no_timing o Toplevel.keep (quickcheck_cmd args i)));
492 val auto_quickcheck = Quickcheck.auto;