better error reporting: detect missing E proofs and remove Vampire native format error
(* Title: HOL/Tools/ATP/atp_proof.ML
Author: Lawrence C. Paulson, Cambridge University Computer Laboratory
Author: Claire Quigley, Cambridge University Computer Laboratory
Author: Jasmin Blanchette, TU Muenchen
Abstract representation of ATP proofs and TSTP/Vampire/SPASS syntax.
*)
signature ATP_PROOF =
sig
type 'a fo_term = 'a ATP_Problem.fo_term
type ('a, 'b, 'c) formula = ('a, 'b, 'c) ATP_Problem.formula
datatype failure =
Unprovable |
IncompleteUnprovable |
ProofMissing |
UnsoundProof of bool |
CantConnect |
TimedOut |
OutOfResources |
SpassTooOld |
VampireTooOld |
NoPerl |
NoLibwwwPerl |
NoRealZ3 |
MalformedInput |
MalformedOutput |
Interrupted |
Crashed |
InternalError |
UnknownError of string
type step_name = string * string option
datatype 'a step =
Definition of step_name * 'a * 'a |
Inference of step_name * 'a * step_name list
type 'a proof = ('a, 'a, 'a fo_term) formula step list
val strip_spaces : (char -> bool) -> string -> string
val short_output : bool -> string -> string
val string_for_failure : failure -> string
val extract_important_message : string -> string
val extract_known_failure :
(failure * string) list -> string -> failure option
val extract_tstplike_proof_and_outcome :
bool -> bool -> bool -> int -> (string * string) list
-> (failure * string) list -> string -> string * failure option
val is_same_step : step_name * step_name -> bool
val atp_proof_from_tstplike_proof : string -> string proof
val map_term_names_in_atp_proof :
(string -> string) -> string proof -> string proof
val nasty_atp_proof : string Symtab.table -> string proof -> string proof
end;
structure ATP_Proof : ATP_PROOF =
struct
open ATP_Problem
datatype failure =
Unprovable |
IncompleteUnprovable |
ProofMissing |
UnsoundProof of bool |
CantConnect |
TimedOut |
OutOfResources |
SpassTooOld |
VampireTooOld |
NoPerl |
NoLibwwwPerl |
NoRealZ3 |
MalformedInput |
MalformedOutput |
Interrupted |
Crashed |
InternalError |
UnknownError of string
fun strip_spaces_in_list _ [] = []
| strip_spaces_in_list _ [c1] = if Char.isSpace c1 then [] else [str c1]
| strip_spaces_in_list is_evil [c1, c2] =
strip_spaces_in_list is_evil [c1] @ strip_spaces_in_list is_evil [c2]
| strip_spaces_in_list is_evil (c1 :: c2 :: c3 :: cs) =
if Char.isSpace c1 then
strip_spaces_in_list is_evil (c2 :: c3 :: cs)
else if Char.isSpace c2 then
if Char.isSpace c3 then
strip_spaces_in_list is_evil (c1 :: c3 :: cs)
else
str c1 :: (if forall is_evil [c1, c3] then [" "] else []) @
strip_spaces_in_list is_evil (c3 :: cs)
else
str c1 :: strip_spaces_in_list is_evil (c2 :: c3 :: cs)
fun strip_spaces is_evil =
implode o strip_spaces_in_list is_evil o String.explode
fun is_ident_char c = Char.isAlphaNum c orelse c = #"_"
val strip_spaces_except_between_ident_chars = strip_spaces is_ident_char
fun elide_string threshold s =
if size s > threshold then
String.extract (s, 0, SOME (threshold div 2 - 5)) ^ " ...... " ^
String.extract (s, size s - (threshold + 1) div 2 + 6, NONE)
else
s
fun short_output verbose output =
if verbose then
if output = "" then "No details available" else elide_string 1000 output
else
""
val missing_message_tail =
" appears to be missing. You will need to install it if you want to invoke \
\remote provers."
fun string_for_failure Unprovable =
"The problem is unprovable."
| string_for_failure IncompleteUnprovable =
"The prover gave up."
| string_for_failure ProofMissing =
"The prover claims the conjecture is a theorem but did not provide a proof."
| string_for_failure (UnsoundProof false) =
"The prover found a type-unsound proof (or, very unlikely, your axioms \
\are inconsistent). Try passing the \"full_types\" option to Sledgehammer \
\to avoid such spurious proofs."
| string_for_failure (UnsoundProof true) =
"The prover found a type-unsound proof even though a supposedly type-sound \
\encoding was used (or, very unlikely, your axioms are inconsistent). You \
\might want to report this to the Isabelle developers."
| string_for_failure CantConnect = "Cannot connect to remote server."
| string_for_failure TimedOut = "Timed out."
| string_for_failure OutOfResources = "The prover ran out of resources."
| string_for_failure SpassTooOld =
"Isabelle requires a more recent version of SPASS with support for the \
\TPTP syntax. To install it, download and extract the package \
\\"http://isabelle.in.tum.de/dist/contrib/spass-3.7.tar.gz\" and add the \
\\"spass-3.7\" directory's absolute path to " ^
Path.print (Path.expand (Path.appends
(Path.variable "ISABELLE_HOME_USER" ::
map Path.basic ["etc", "components"]))) ^
" on a line of its own."
| string_for_failure VampireTooOld =
"Isabelle requires a more recent version of Vampire. To install it, follow \
\the instructions from the Sledgehammer manual (\"isabelle doc\
\ sledgehammer\")."
| string_for_failure NoPerl = "Perl" ^ missing_message_tail
| string_for_failure NoLibwwwPerl =
"The Perl module \"libwww-perl\"" ^ missing_message_tail
| string_for_failure NoRealZ3 =
"The environment variable \"Z3_REAL_SOLVER\" must be set to Z3's full path."
| string_for_failure MalformedInput =
"The generated problem is malformed. Please report this to the Isabelle \
\developers."
| string_for_failure MalformedOutput = "The prover output is malformed."
| string_for_failure Crashed = "The prover crashed."
| string_for_failure InternalError = "An internal prover error occurred."
| string_for_failure (UnknownError string) =
"A prover error occurred" ^
(if string = "" then ". (Pass the \"verbose\" option for details.)"
else ":\n" ^ string)
fun extract_delimited (begin_delim, end_delim) output =
output |> first_field begin_delim |> the |> snd
|> first_field end_delim |> the |> fst
|> first_field "\n" |> the |> snd
handle Option.Option => ""
val tstp_important_message_delims =
("% SZS start RequiredInformation", "% SZS end RequiredInformation")
fun extract_important_message output =
case extract_delimited tstp_important_message_delims output of
"" => ""
| s => s |> space_explode "\n" |> filter_out (curry (op =) "")
|> map (perhaps (try (unprefix "%")))
|> map (perhaps (try (unprefix " ")))
|> space_implode "\n " |> quote
(* Splits by the first possible of a list of delimiters. *)
fun extract_tstplike_proof delims output =
case pairself (find_first (fn s => String.isSubstring s output))
(ListPair.unzip delims) of
(SOME begin_delim, SOME end_delim) =>
extract_delimited (begin_delim, end_delim) output
| _ => ""
fun extract_known_failure known_failures output =
known_failures
|> find_first (fn (_, pattern) => String.isSubstring pattern output)
|> Option.map fst
fun extract_tstplike_proof_and_outcome debug verbose complete res_code
proof_delims known_failures output =
case extract_tstplike_proof proof_delims output of
"" =>
("", SOME (if res_code = 0 andalso (not debug orelse output = "") then
ProofMissing
else case extract_known_failure known_failures output of
SOME failure =>
if failure = IncompleteUnprovable andalso complete then
Unprovable
else
failure
| NONE => UnknownError (short_output verbose output)))
| tstplike_proof => (tstplike_proof, NONE)
fun mk_anot (AConn (ANot, [phi])) = phi
| mk_anot phi = AConn (ANot, [phi])
fun mk_aconn c (phi1, phi2) = AConn (c, [phi1, phi2])
type step_name = string * string option
fun is_same_step p = p |> pairself fst |> op =
fun step_name_ord p =
let val q = pairself fst p in
(* The "unprefix" part is to cope with remote Vampire's output. The proper
solution would be to perform a topological sort, e.g. using the nice
"Graph" functor. *)
case pairself (Int.fromString o perhaps (try (unprefix "f"))) q of
(NONE, NONE) => string_ord q
| (NONE, SOME _) => LESS
| (SOME _, NONE) => GREATER
| (SOME i, SOME j) => int_ord (i, j)
end
datatype 'a step =
Definition of step_name * 'a * 'a |
Inference of step_name * 'a * step_name list
type 'a proof = ('a, 'a, 'a fo_term) formula step list
fun step_name (Definition (name, _, _)) = name
| step_name (Inference (name, _, _)) = name
(**** PARSING OF TSTP FORMAT ****)
(* Strings enclosed in single quotes (e.g., file names) *)
val scan_general_id =
$$ "'" |-- Scan.repeat (~$$ "'") --| $$ "'" >> implode
|| Scan.repeat ($$ "$") -- Scan.many1 Symbol.is_letdig
>> (fn (ss1, ss2) => implode ss1 ^ implode ss2)
(* Generalized first-order terms, which include file names, numbers, etc. *)
fun parse_annotation x =
((scan_general_id ::: Scan.repeat ($$ " " |-- scan_general_id))
-- Scan.optional parse_annotation [] >> op @
|| $$ "(" |-- parse_annotations --| $$ ")"
|| $$ "[" |-- parse_annotations --| $$ "]") x
and parse_annotations x =
(Scan.optional (parse_annotation
::: Scan.repeat ($$ "," |-- parse_annotation)) []
>> flat) x
fun parse_term x =
(scan_general_id
--| Scan.option ($$ ":" -- scan_general_id) (* ignore TFF types for now *)
-- Scan.optional ($$ "(" |-- parse_terms --| $$ ")") []
>> ATerm) x
and parse_terms x = (parse_term ::: Scan.repeat ($$ "," |-- parse_term)) x
fun parse_atom x =
(parse_term -- Scan.option (Scan.option ($$ "!") --| $$ "=" -- parse_term)
>> (fn (u1, NONE) => AAtom u1
| (u1, SOME (NONE, u2)) => AAtom (ATerm ("c_equal", [u1, u2]))
| (u1, SOME (SOME _, u2)) =>
mk_anot (AAtom (ATerm ("c_equal", [u1, u2]))))) x
fun fo_term_head (ATerm (s, _)) = s
(* TPTP formulas are fully parenthesized, so we don't need to worry about
operator precedence. *)
fun parse_literal x =
((Scan.repeat ($$ "~") >> length)
-- ($$ "(" |-- parse_formula --| $$ ")"
|| parse_quantified_formula
|| parse_atom)
>> (fn (n, phi) => phi |> n mod 2 = 1 ? mk_anot)) x
and parse_formula x =
(parse_literal
-- Scan.option ((Scan.this_string "=>" >> K AImplies
|| Scan.this_string "<=>" >> K AIff
|| Scan.this_string "<~>" >> K ANotIff
|| Scan.this_string "<=" >> K AIf
|| $$ "|" >> K AOr
|| $$ "&" >> K AAnd)
-- parse_formula)
>> (fn (phi1, NONE) => phi1
| (phi1, SOME (c, phi2)) => mk_aconn c (phi1, phi2))) x
and parse_quantified_formula x =
(($$ "!" >> K AForall || $$ "?" >> K AExists)
--| $$ "[" -- parse_terms --| $$ "]" --| $$ ":" -- parse_literal
>> (fn ((q, ts), phi) =>
(* FIXME: TFF *)
AQuant (q, map (rpair NONE o fo_term_head) ts, phi))) x
val parse_tstp_extra_arguments =
Scan.optional ($$ "," |-- parse_annotation
--| Scan.option ($$ "," |-- parse_annotations)) []
val vampire_unknown_fact = "unknown"
val tofof_fact_prefix = "fof_"
(* Syntax: (cnf|fof|tff)\(<num>, <formula_role>, <formula> <extra_arguments>\).
The <num> could be an identifier, but we assume integers. *)
fun parse_tstp_line x =
(((Scan.this_string "cnf" || Scan.this_string "fof" || Scan.this_string "tff")
-- $$ "(")
|-- scan_general_id --| $$ "," -- Symbol.scan_id --| $$ ","
-- parse_formula -- parse_tstp_extra_arguments --| $$ ")" --| $$ "."
>> (fn (((num, role), phi), deps) =>
let
val (name, deps) =
case deps of
["file", _, s] =>
((num,
if s = vampire_unknown_fact then NONE
else SOME (s |> perhaps (try (unprefix tofof_fact_prefix)))),
[])
| _ => ((num, NONE), deps)
in
case role of
"definition" =>
(case phi of
AConn (AIff, [phi1 as AAtom _, phi2]) =>
Definition (name, phi1, phi2)
| AAtom (ATerm ("c_equal", _)) =>
(* Vampire's equality proxy axiom *)
Inference (name, phi, map (rpair NONE) deps)
| _ => raise Fail "malformed definition")
| _ => Inference (name, phi, map (rpair NONE) deps)
end)) x
(**** PARSING OF SPASS OUTPUT ****)
(* SPASS returns clause references of the form "x.y". We ignore "y", whose role
is not clear anyway. *)
val parse_dot_name = scan_general_id --| $$ "." --| scan_general_id
val parse_spass_annotations =
Scan.optional ($$ ":" |-- Scan.repeat (parse_dot_name
--| Scan.option ($$ ","))) []
(* It is not clear why some literals are followed by sequences of stars and/or
pluses. We ignore them. *)
fun parse_decorated_atom x =
(parse_atom --| Scan.repeat ($$ "*" || $$ "+" || $$ " ")) x
fun mk_horn ([], []) = AAtom (ATerm ("c_False", []))
| mk_horn ([], pos_lits) = foldr1 (mk_aconn AOr) pos_lits
| mk_horn (neg_lits, []) = mk_anot (foldr1 (mk_aconn AAnd) neg_lits)
| mk_horn (neg_lits, pos_lits) =
mk_aconn AImplies (foldr1 (mk_aconn AAnd) neg_lits,
foldr1 (mk_aconn AOr) pos_lits)
fun parse_horn_clause x =
(Scan.repeat parse_decorated_atom --| $$ "|" --| $$ "|"
-- Scan.repeat parse_decorated_atom --| $$ "-" --| $$ ">"
-- Scan.repeat parse_decorated_atom
>> (mk_horn o apfst (op @))) x
(* Syntax: <num>[0:<inference><annotations>]
<atoms> || <atoms> -> <atoms>. *)
fun parse_spass_line x =
(scan_general_id --| $$ "[" --| $$ "0" --| $$ ":" --| Symbol.scan_id
-- parse_spass_annotations --| $$ "]" -- parse_horn_clause --| $$ "."
>> (fn ((num, deps), u) =>
Inference ((num, NONE), u, map (rpair NONE) deps))) x
fun parse_line x = (parse_tstp_line || parse_spass_line) x
fun parse_proof s =
s |> strip_spaces_except_between_ident_chars
|> raw_explode
|> Scan.finite Symbol.stopper
(Scan.error (!! (fn _ => raise Fail "unrecognized ATP output")
(Scan.repeat1 parse_line)))
|> fst
fun clean_up_dependency seen dep = find_first (curry is_same_step dep) seen
fun clean_up_dependencies _ [] = []
| clean_up_dependencies seen ((step as Definition (name, _, _)) :: steps) =
step :: clean_up_dependencies (name :: seen) steps
| clean_up_dependencies seen (Inference (name, u, deps) :: steps) =
Inference (name, u, map_filter (clean_up_dependency seen) deps) ::
clean_up_dependencies (name :: seen) steps
fun atp_proof_from_tstplike_proof "" = []
| atp_proof_from_tstplike_proof s =
s ^ "$" (* the $ sign acts as a sentinel (FIXME: needed?) *)
|> parse_proof
|> sort (step_name_ord o pairself step_name)
|> clean_up_dependencies []
fun map_term_names_in_term f (ATerm (s, ts)) =
ATerm (f s, map (map_term_names_in_term f) ts)
fun map_term_names_in_formula f (AQuant (q, xs, phi)) =
AQuant (q, xs, map_term_names_in_formula f phi)
| map_term_names_in_formula f (AConn (c, phis)) =
AConn (c, map (map_term_names_in_formula f) phis)
| map_term_names_in_formula f (AAtom t) = AAtom (map_term_names_in_term f t)
fun map_term_names_in_step f (Definition (name, phi1, phi2)) =
Definition (name, map_term_names_in_formula f phi1,
map_term_names_in_formula f phi2)
| map_term_names_in_step f (Inference (name, phi, deps)) =
Inference (name, map_term_names_in_formula f phi, deps)
fun map_term_names_in_atp_proof f = map (map_term_names_in_step f)
fun nasty_name pool s = s |> Symtab.lookup pool |> the_default s
fun nasty_atp_proof pool =
if Symtab.is_empty pool then I
else map_term_names_in_atp_proof (nasty_name pool)
end;