renamed Output.priority to Output.urgent_message to emphasize its special role more clearly;
1 (* Title: Tools/quickcheck.ML
2 Author: Stefan Berghofer, Florian Haftmann, 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};
20 val test_params_of: Proof.context -> test_params
21 val report : Proof.context -> bool
22 val set_reporting : bool -> Context.generic -> Context.generic
24 string * (Proof.context -> term -> int -> term list option * (bool list * bool))
25 -> Context.generic -> Context.generic
26 (* testing terms and proof states *)
27 val gen_test_term: Proof.context -> bool -> string option -> int -> int -> term ->
28 (string * term) list option * ((string * int) list * ((int * report list) list) option)
29 val test_term: Proof.context -> bool -> string option -> int -> int -> term ->
30 (string * term) list option
31 val quickcheck: (string * string list) list -> int -> Proof.state -> (string * term) list option
34 structure Quickcheck : QUICKCHECK =
39 val auto = Unsynchronized.ref false;
41 val timing = Unsynchronized.ref false;
44 ProofGeneralPgip.add_preference Preferences.category_tracing
45 (Unsynchronized.setmp auto true (fn () =>
46 Preferences.bool_pref auto
48 "Run Quickcheck automatically.") ());
50 (* quickcheck report *)
52 datatype single_report = Run of bool list * bool | MatchExc
54 datatype report = Report of
55 { iterations : int, raised_match_errors : int,
56 satisfied_assms : int list, positive_concl_tests : int }
58 fun collect_single_report single_report
59 (Report {iterations = iterations, raised_match_errors = raised_match_errors,
60 satisfied_assms = satisfied_assms, positive_concl_tests = positive_concl_tests}) =
63 Report {iterations = iterations + 1, raised_match_errors = raised_match_errors + 1,
64 satisfied_assms = satisfied_assms, positive_concl_tests = positive_concl_tests}
65 | Run (assms, concl) =>
66 Report {iterations = iterations + 1, raised_match_errors = raised_match_errors,
68 map2 (fn b => fn s => if b then s + 1 else s) assms
69 (if null satisfied_assms then replicate (length assms) 0 else satisfied_assms),
70 positive_concl_tests = if concl then positive_concl_tests + 1 else positive_concl_tests}
74 datatype expectation = No_Expectation | No_Counterexample | Counterexample;
76 fun merge_expectation (expect1, expect2) =
77 if expect1 = expect2 then expect1 else No_Expectation
79 (* quickcheck configuration -- default parameters, test generators *)
81 datatype test_params = Test_Params of
82 { size: int, iterations: int, default_type: typ list, no_assms: bool,
83 expect : expectation, report: bool, quiet : bool};
85 fun dest_test_params (Test_Params { size, iterations, default_type, no_assms, expect, report, quiet }) =
86 ((size, iterations), ((default_type, no_assms), ((expect, report), quiet)));
88 fun make_test_params ((size, iterations), ((default_type, no_assms), ((expect, report), quiet))) =
89 Test_Params { size = size, iterations = iterations, default_type = default_type,
90 no_assms = no_assms, expect = expect, report = report, quiet = quiet };
92 fun map_test_params f (Test_Params { size, iterations, default_type, no_assms, expect, report, quiet }) =
93 make_test_params (f ((size, iterations), ((default_type, no_assms), ((expect, report), quiet))));
96 (Test_Params { size = size1, iterations = iterations1, default_type = default_type1,
97 no_assms = no_assms1, expect = expect1, report = report1, quiet = quiet1 },
98 Test_Params { size = size2, iterations = iterations2, default_type = default_type2,
99 no_assms = no_assms2, expect = expect2, report = report2, quiet = quiet2 }) =
100 make_test_params ((Int.max (size1, size2), Int.max (iterations1, iterations2)),
101 ((merge (op =) (default_type1, default_type2), no_assms1 orelse no_assms2),
102 ((merge_expectation (expect1, expect2), report1 orelse report2), quiet1 orelse quiet2)));
104 structure Data = Generic_Data
107 (string * (Proof.context -> term -> int -> term list option * (bool list * bool))) list
109 val empty = ([], Test_Params
110 { size = 10, iterations = 100, default_type = [], no_assms = false,
111 expect = No_Expectation, report = false, quiet = false});
113 fun merge ((generators1, params1), (generators2, params2)) : T =
114 (AList.merge (op =) (K true) (generators1, generators2),
115 merge_test_params (params1, params2));
118 val test_params_of = snd o Data.get o Context.Proof;
120 val report = snd o fst o snd o snd o dest_test_params o test_params_of
122 fun map_report f (Test_Params { size, iterations, default_type, no_assms, expect, report, quiet }) =
123 make_test_params ((size, iterations), ((default_type, no_assms), ((expect, f report), quiet)));
125 fun set_reporting report = Data.map (apsnd (map_report (K report)))
127 val add_generator = Data.map o apfst o AList.update (op =);
129 (* generating tests *)
131 fun mk_tester_select name ctxt =
132 case AList.lookup (op =) ((fst o Data.get o Context.Proof) ctxt) name
133 of NONE => error ("No such quickcheck generator: " ^ name)
134 | SOME generator => generator ctxt;
136 fun mk_testers ctxt t =
137 (map snd o fst o Data.get o Context.Proof) ctxt
138 |> map_filter (fn generator => try (generator ctxt) t);
140 fun mk_testers_strict ctxt t =
142 val generators = ((map snd o fst o Data.get o Context.Proof) ctxt)
143 val testers = map (fn generator => Exn.capture (generator ctxt) t) generators;
144 in if forall (is_none o Exn.get_result) testers
145 then [(Exn.release o snd o split_last) testers]
146 else map_filter Exn.get_result testers
150 (* testing propositions *)
152 fun prep_test_term t =
154 val _ = (null (Term.add_tvars t []) andalso null (Term.add_tfrees t [])) orelse
155 error "Term to be tested contains type variables";
156 val _ = null (Term.add_vars t []) orelse
157 error "Term to be tested contains schematic variables";
158 val frees = Term.add_frees t [];
159 in (map fst frees, list_abs_free (frees, t)) end
161 fun cpu_time description f =
163 val start = start_timing ()
164 val result = Exn.capture f ()
165 val time = Time.toMilliseconds (#cpu (end_timing start))
166 in (Exn.release result, (description, time)) end
168 fun gen_test_term ctxt quiet generator_name size i t =
170 val (names, t') = prep_test_term t;
171 val (testers, comp_time) = cpu_time "quickcheck compilation"
172 (fn () => (case generator_name
173 of NONE => if quiet then mk_testers ctxt t' else mk_testers_strict ctxt t'
174 | SOME name => [mk_tester_select name ctxt t']));
175 fun iterate f 0 report = (NONE, report)
176 | iterate f j report =
178 val (test_result, single_report) = apsnd Run (f ()) handle Match => (if quiet then ()
179 else warning "Exception Match raised during quickcheck"; (NONE, MatchExc))
180 val report = collect_single_report single_report report
182 case test_result of NONE => iterate f (j - 1) report | SOME q => (SOME q, report)
184 val empty_report = Report { iterations = 0, raised_match_errors = 0,
185 satisfied_assms = [], positive_concl_tests = 0 }
186 fun with_testers k [] = (NONE, [])
187 | with_testers k (tester :: testers) =
188 case iterate (fn () => tester (k - 1)) i empty_report
189 of (NONE, report) => apsnd (cons report) (with_testers k testers)
190 | (SOME q, report) => (SOME q, [report]);
191 fun with_size k reports =
192 if k > size then (NONE, reports)
194 (if quiet then () else Output.urgent_message ("Test data size: " ^ string_of_int k);
196 val (result, new_report) = with_testers k testers
197 val reports = ((k, new_report) :: reports)
198 in case result of NONE => with_size (k + 1) reports | SOME q => (SOME q, reports) end);
199 val ((result, reports), exec_time) = cpu_time "quickcheck execution"
201 (fn result => case result of NONE => NONE
202 | SOME ts => SOME (names ~~ ts)) (with_size 1 []))
204 (result, ([exec_time, comp_time], if report ctxt then SOME reports else NONE))
207 fun test_term ctxt quiet generator_name size i t =
209 |> Context.proof_map (set_reporting false)
210 |> (fn ctxt' => fst (gen_test_term ctxt' quiet generator_name size i t))
212 exception WELLSORTED of string
214 fun monomorphic_term thy insts default_T =
216 fun subst (T as TFree (v, S)) =
218 val T' = AList.lookup (op =) insts v
219 |> the_default default_T
220 in if Sign.of_sort thy (T', S) then T'
221 else raise (WELLSORTED ("For instantiation with default_type " ^ Syntax.string_of_typ_global thy default_T ^
222 ":\n" ^ Syntax.string_of_typ_global thy T' ^
223 " to be substituted for variable " ^
224 Syntax.string_of_typ_global thy T ^ " does not have sort " ^
225 Syntax.string_of_sort_global thy S))
228 in (map_types o map_atyps) subst end;
230 datatype wellsorted_error = Wellsorted_Error of string | Term of term
232 fun test_goal quiet generator_name size iterations default_Ts no_assms insts i state =
234 val lthy = Proof.context_of state;
235 val thy = Proof.theory_of state;
236 fun strip (Const ("all", _) $ Abs (_, _, t)) = strip t
238 val {goal = st, ...} = Proof.raw_goal state;
239 val (gi, frees) = Logic.goal_params (prop_of st) i;
240 val some_locale = case (Option.map #target o Named_Target.peek) lthy
243 | SOME locale => SOME locale;
244 val assms = if no_assms then [] else case some_locale
245 of NONE => Assumption.all_assms_of lthy
246 | SOME locale => Assumption.local_assms_of lthy (Locale.init locale thy);
247 val proto_goal = Logic.list_implies (map Thm.term_of assms, subst_bounds (frees, strip gi));
248 val check_goals = case some_locale
249 of NONE => [proto_goal]
250 | SOME locale => map (fn (_, phi) => Morphism.term phi proto_goal) (Locale.registrations_of (Context.Theory thy) (*FIXME*) locale);
251 val inst_goals = maps (fn check_goal => map (fn T =>
252 Term ((Object_Logic.atomize_term thy o monomorphic_term thy insts T) check_goal)
253 handle WELLSORTED s => Wellsorted_Error s) default_Ts) check_goals
254 val error_msg = cat_lines (map_filter (fn Term t => NONE | Wellsorted_Error s => SOME s) inst_goals)
255 val correct_inst_goals =
256 case map_filter (fn Term t => SOME t | Wellsorted_Error s => NONE) inst_goals of
257 [] => error error_msg
259 val _ = if quiet then () else warning error_msg
260 fun collect_results f reports [] = (NONE, rev reports)
261 | collect_results f reports (t :: ts) =
263 (SOME res, report) => (SOME res, rev (report :: reports))
264 | (NONE, report) => collect_results f (report :: reports) ts
265 in collect_results (gen_test_term lthy quiet generator_name size iterations) [] correct_inst_goals end;
267 (* pretty printing *)
269 fun pretty_counterex ctxt NONE = Pretty.str "Quickcheck found no counterexample."
270 | pretty_counterex ctxt (SOME cex) =
271 Pretty.chunks (Pretty.str "Quickcheck found a counterexample:\n" ::
273 Pretty.block [Pretty.str (s ^ " ="), Pretty.brk 1, Syntax.pretty_term ctxt t]) cex);
275 fun pretty_report (Report {iterations = iterations, raised_match_errors = raised_match_errors,
276 satisfied_assms = satisfied_assms, positive_concl_tests = positive_concl_tests}) =
278 fun pretty_stat s i = Pretty.block ([Pretty.str (s ^ ": " ^ string_of_int i)])
280 ([pretty_stat "iterations" iterations,
281 pretty_stat "match exceptions" raised_match_errors]
282 @ map_index (fn (i, n) => pretty_stat ("satisfied " ^ string_of_int (i + 1) ^ ". assumption") n)
284 @ [pretty_stat "positive conclusion tests" positive_concl_tests])
287 fun pretty_reports' [report] = [Pretty.chunks (pretty_report report)]
288 | pretty_reports' reports =
289 map_index (fn (i, report) =>
290 Pretty.chunks (Pretty.str (string_of_int (i + 1) ^ ". generator:\n") :: pretty_report report))
293 fun pretty_reports ctxt (SOME reports) =
294 Pretty.chunks (Pretty.str "Quickcheck report:" ::
295 maps (fn (size, reports) =>
296 Pretty.str ("size " ^ string_of_int size ^ ":") :: pretty_reports' reports @ [Pretty.brk 1])
298 | pretty_reports ctxt NONE = Pretty.str ""
300 fun pretty_counterex_and_reports ctxt (cex, timing_and_reports) =
301 Pretty.chunks (pretty_counterex ctxt cex :: map (pretty_reports ctxt) (map snd timing_and_reports))
303 (* automatic testing *)
305 fun auto_quickcheck state =
310 val ctxt = Proof.context_of state;
311 val Test_Params {size, iterations, default_type, no_assms, ...} =
312 (snd o Data.get o Context.Proof) ctxt;
315 |> Proof.map_context (Context.proof_map (set_reporting false))
316 |> try (test_goal true NONE size iterations default_type no_assms [] 1);
319 NONE => (false, state)
320 | SOME (NONE, report) => (false, state)
321 | SOME (cex, report) => (true, Proof.goal_message (K (Pretty.chunks [Pretty.str "",
322 Pretty.mark Markup.hilite (pretty_counterex ctxt cex)])) state)
325 val setup = Auto_Tools.register_tool ("quickcheck", auto_quickcheck)
330 fun read_nat s = case (Library.read_int o Symbol.explode) s
331 of (k, []) => if k >= 0 then k
332 else error ("Not a natural number: " ^ s)
333 | (_, _ :: _) => error ("Not a natural number: " ^ s);
335 fun read_bool "false" = false
336 | read_bool "true" = true
337 | read_bool s = error ("Not a Boolean value: " ^ s)
339 fun read_expectation "no_expectation" = No_Expectation
340 | read_expectation "no_counterexample" = No_Counterexample
341 | read_expectation "counterexample" = Counterexample
342 | read_expectation s = error ("Not an expectation value: " ^ s)
344 fun parse_test_param ctxt ("size", [arg]) =
345 (apfst o apfst o K) (read_nat arg)
346 | parse_test_param ctxt ("iterations", [arg]) =
347 (apfst o apsnd o K) (read_nat arg)
348 | parse_test_param ctxt ("default_type", arg) =
349 (apsnd o apfst o apfst o K) (map (ProofContext.read_typ ctxt) arg)
350 | parse_test_param ctxt ("no_assms", [arg]) =
351 (apsnd o apfst o apsnd o K) (read_bool arg)
352 | parse_test_param ctxt ("expect", [arg]) =
353 (apsnd o apsnd o apfst o apfst o K) (read_expectation arg)
354 | parse_test_param ctxt ("report", [arg]) =
355 (apsnd o apsnd o apfst o apsnd o K) (read_bool arg)
356 | parse_test_param ctxt ("quiet", [arg]) =
357 (apsnd o apsnd o apsnd o K) (read_bool arg)
358 | parse_test_param ctxt (name, _) =
359 error ("Unknown test parameter: " ^ name);
361 fun parse_test_param_inst ctxt ("generator", [arg]) =
362 (apsnd o apfst o K o SOME) arg
363 | parse_test_param_inst ctxt (name, arg) =
364 case try (ProofContext.read_typ ctxt) name
365 of SOME (TFree (v, _)) => (apsnd o apsnd o AList.update (op =))
366 (v, ProofContext.read_typ ctxt (the_single arg))
367 | _ => (apfst o parse_test_param ctxt) (name, arg);
369 fun quickcheck_params_cmd args thy =
371 val ctxt = ProofContext.init_global thy
372 val f = fold (parse_test_param ctxt) args;
375 |> (Context.theory_map o Data.map o apsnd o map_test_params) f
378 fun gen_quickcheck args i state =
380 val ctxt = Proof.context_of state;
381 val default_params = (dest_test_params o snd o Data.get o Context.Proof) ctxt;
382 val f = fold (parse_test_param_inst ctxt) args;
383 val (((size, iterations), ((default_type, no_assms), ((expect, report), quiet))), (generator_name, insts)) =
384 f (default_params, (NONE, []));
387 |> Proof.map_context (Context.proof_map (set_reporting report))
388 |> test_goal quiet generator_name size iterations default_type no_assms insts i
389 |> tap (fn (SOME x, _) => if expect = No_Counterexample then
390 error ("quickcheck expected to find no counterexample but found one") else ()
391 | (NONE, _) => if expect = Counterexample then
392 error ("quickcheck expected to find a counterexample but did not find one") else ())
395 fun quickcheck args i state = fst (gen_quickcheck args i state);
397 fun quickcheck_cmd args i state =
398 gen_quickcheck args i (Toplevel.proof_of state)
399 |> Pretty.writeln o pretty_counterex_and_reports (Toplevel.context_of state);
401 val parse_arg = Parse.name -- (Scan.optional (Parse.$$$ "=" |--
402 ((Parse.name >> single) || (Parse.$$$ "[" |-- Parse.list1 Parse.name --| Parse.$$$ "]"))) ["true"]);
404 val parse_args = Parse.$$$ "[" |-- Parse.list1 parse_arg --| Parse.$$$ "]"
408 Outer_Syntax.command "quickcheck_params" "set parameters for random testing" Keyword.thy_decl
409 (parse_args >> (fn args => Toplevel.theory (quickcheck_params_cmd args)));
412 Outer_Syntax.improper_command "quickcheck" "try to find counterexample for subgoal" Keyword.diag
413 (parse_args -- Scan.optional Parse.nat 1
414 >> (fn (args, i) => Toplevel.no_timing o Toplevel.keep (quickcheck_cmd args i)));
419 val auto_quickcheck = Quickcheck.auto;