author | blanchet |
Mon, 29 Jul 2013 16:13:35 +0200 | |
changeset 52756 | 1ac8a0d0ddb1 |
parent 52755 | 4183c3219745 |
child 52993 | dd28fbc5cecb |
permissions | -rw-r--r-- |
39452 | 1 |
(* Title: HOL/Tools/ATP/atp_proof.ML |
2 |
Author: Lawrence C. Paulson, Cambridge University Computer Laboratory |
|
3 |
Author: Claire Quigley, Cambridge University Computer Laboratory |
|
4 |
Author: Jasmin Blanchette, TU Muenchen |
|
5 |
||
42876
e336ef6313aa
more informative message when Sledgehammer finds an unsound proof
blanchet
parents:
42848
diff
changeset
|
6 |
Abstract representation of ATP proofs and TSTP/SPASS syntax. |
39452 | 7 |
*) |
8 |
||
9 |
signature ATP_PROOF = |
|
10 |
sig |
|
43678 | 11 |
type ('a, 'b) ho_term = ('a, 'b) ATP_Problem.ho_term |
50012
01cb92151a53
track formula roles in proofs and use that to determine whether the conjecture should be negated or not
blanchet
parents:
50011
diff
changeset
|
12 |
type formula_role = ATP_Problem.formula_role |
48135 | 13 |
type ('a, 'b, 'c, 'd) formula = ('a, 'b, 'c, 'd) ATP_Problem.formula |
42943 | 14 |
type 'a problem = 'a ATP_Problem.problem |
39452 | 15 |
|
42965
1403595ec38c
slightly gracefuller handling of LEO-II and Satallax output
blanchet
parents:
42962
diff
changeset
|
16 |
exception UNRECOGNIZED_ATP_PROOF of unit |
1403595ec38c
slightly gracefuller handling of LEO-II and Satallax output
blanchet
parents:
42962
diff
changeset
|
17 |
|
39491
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
18 |
datatype failure = |
42587
4fbb1de05169
fixed SPASS fact offset calculation and report unexpected unsound proofs with so-called sound encodings
blanchet
parents:
42550
diff
changeset
|
19 |
Unprovable | |
43050
59284a13abc4
support "metis" and "metisFT" as provers in the architecture, so they can be used for minimizing
blanchet
parents:
43029
diff
changeset
|
20 |
GaveUp | |
42587
4fbb1de05169
fixed SPASS fact offset calculation and report unexpected unsound proofs with so-called sound encodings
blanchet
parents:
42550
diff
changeset
|
21 |
ProofMissing | |
42882
391e41ac038b
make sure the Vampire incomplete proof detection code kicks in
blanchet
parents:
42876
diff
changeset
|
22 |
ProofIncomplete | |
44915
635ae0a73688
simplified unsound proof detection by removing impossible case
blanchet
parents:
44784
diff
changeset
|
23 |
UnsoundProof of bool * string list | |
42587
4fbb1de05169
fixed SPASS fact offset calculation and report unexpected unsound proofs with so-called sound encodings
blanchet
parents:
42550
diff
changeset
|
24 |
CantConnect | |
4fbb1de05169
fixed SPASS fact offset calculation and report unexpected unsound proofs with so-called sound encodings
blanchet
parents:
42550
diff
changeset
|
25 |
TimedOut | |
42953
26111aafab12
detect inappropriate problems and crashes better in Waldmeister
blanchet
parents:
42943
diff
changeset
|
26 |
Inappropriate | |
42587
4fbb1de05169
fixed SPASS fact offset calculation and report unexpected unsound proofs with so-called sound encodings
blanchet
parents:
42550
diff
changeset
|
27 |
OutOfResources | |
47950
9cb132898ac8
invite users to upgrade their SPASS (so we can get rid of old code)
blanchet
parents:
47947
diff
changeset
|
28 |
OldSPASS | |
42587
4fbb1de05169
fixed SPASS fact offset calculation and report unexpected unsound proofs with so-called sound encodings
blanchet
parents:
42550
diff
changeset
|
29 |
NoPerl | |
4fbb1de05169
fixed SPASS fact offset calculation and report unexpected unsound proofs with so-called sound encodings
blanchet
parents:
42550
diff
changeset
|
30 |
NoLibwwwPerl | |
4fbb1de05169
fixed SPASS fact offset calculation and report unexpected unsound proofs with so-called sound encodings
blanchet
parents:
42550
diff
changeset
|
31 |
MalformedInput | |
4fbb1de05169
fixed SPASS fact offset calculation and report unexpected unsound proofs with so-called sound encodings
blanchet
parents:
42550
diff
changeset
|
32 |
MalformedOutput | |
4fbb1de05169
fixed SPASS fact offset calculation and report unexpected unsound proofs with so-called sound encodings
blanchet
parents:
42550
diff
changeset
|
33 |
Interrupted | |
4fbb1de05169
fixed SPASS fact offset calculation and report unexpected unsound proofs with so-called sound encodings
blanchet
parents:
42550
diff
changeset
|
34 |
Crashed | |
4fbb1de05169
fixed SPASS fact offset calculation and report unexpected unsound proofs with so-called sound encodings
blanchet
parents:
42550
diff
changeset
|
35 |
InternalError | |
4fbb1de05169
fixed SPASS fact offset calculation and report unexpected unsound proofs with so-called sound encodings
blanchet
parents:
42550
diff
changeset
|
36 |
UnknownError of string |
39491
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
37 |
|
45551 | 38 |
type step_name = string * string list |
51201 | 39 |
type 'a step = step_name * formula_role * 'a * string * step_name list |
39452 | 40 |
|
48135 | 41 |
type 'a proof = ('a, 'a, ('a, 'a) ho_term, 'a) formula step list |
39452 | 42 |
|
41259
13972ced98d9
more precise error messages in "verbose" (or "debug") mode, following this morning's permission debacle
blanchet
parents:
41222
diff
changeset
|
43 |
val short_output : bool -> string -> string |
51998
f732a674db1b
renamed Sledgehammer functions with 'for' in their names to 'of'
blanchet
parents:
51881
diff
changeset
|
44 |
val string_of_failure : failure -> string |
39491
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
45 |
val extract_important_message : string -> string |
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
46 |
val extract_known_failure : |
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
47 |
(failure * string) list -> string -> failure option |
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
48 |
val extract_tstplike_proof_and_outcome : |
48716
1d2a12bb0640
stop distinguishing between complete and incomplete slices, since this is very fragile and has hardly any useful semantics to users
blanchet
parents:
48700
diff
changeset
|
49 |
bool -> (string * string) list -> (failure * string) list -> string |
43473
fb2713b803e6
deal with ATP time slices in a more flexible/robust fashion
blanchet
parents:
43465
diff
changeset
|
50 |
-> string * failure option |
42968
74415622d293
more work on parsing LEO-II proofs and extracting uses of extensionality
blanchet
parents:
42966
diff
changeset
|
51 |
val is_same_atp_step : step_name -> step_name -> bool |
42961 | 52 |
val scan_general_id : string list -> string * string list |
52073
ccb292952774
started adding agsyHOL as an experimental prover
blanchet
parents:
52031
diff
changeset
|
53 |
val agsyhol_coreN : string |
48539 | 54 |
val satallax_coreN : string |
55 |
val z3_tptp_coreN : string |
|
42961 | 56 |
val parse_formula : |
48132
9aa0fad4e864
added type arguments to "ATerm" constructor -- but don't use them yet
blanchet
parents:
48130
diff
changeset
|
57 |
string list |
48135 | 58 |
-> (string, 'a, (string, 'a) ho_term, string) formula * string list |
52031
9a9238342963
tuning -- renamed '_from_' to '_of_' in Sledgehammer
blanchet
parents:
51998
diff
changeset
|
59 |
val atp_proof_of_tstplike_proof : string problem -> string -> string proof |
42968
74415622d293
more work on parsing LEO-II proofs and extracting uses of extensionality
blanchet
parents:
42966
diff
changeset
|
60 |
val clean_up_atp_proof_dependencies : string proof -> string proof |
39454 | 61 |
val map_term_names_in_atp_proof : |
62 |
(string -> string) -> string proof -> string proof |
|
63 |
val nasty_atp_proof : string Symtab.table -> string proof -> string proof |
|
39452 | 64 |
end; |
65 |
||
66 |
structure ATP_Proof : ATP_PROOF = |
|
67 |
struct |
|
68 |
||
43085
0a2f5b86bdd7
first step in sharing more code between ATP and Metis translation
blanchet
parents:
43050
diff
changeset
|
69 |
open ATP_Util |
39491
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
70 |
open ATP_Problem |
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
71 |
|
42965
1403595ec38c
slightly gracefuller handling of LEO-II and Satallax output
blanchet
parents:
42962
diff
changeset
|
72 |
exception UNRECOGNIZED_ATP_PROOF of unit |
1403595ec38c
slightly gracefuller handling of LEO-II and Satallax output
blanchet
parents:
42962
diff
changeset
|
73 |
|
39491
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
74 |
datatype failure = |
42587
4fbb1de05169
fixed SPASS fact offset calculation and report unexpected unsound proofs with so-called sound encodings
blanchet
parents:
42550
diff
changeset
|
75 |
Unprovable | |
43050
59284a13abc4
support "metis" and "metisFT" as provers in the architecture, so they can be used for minimizing
blanchet
parents:
43029
diff
changeset
|
76 |
GaveUp | |
42587
4fbb1de05169
fixed SPASS fact offset calculation and report unexpected unsound proofs with so-called sound encodings
blanchet
parents:
42550
diff
changeset
|
77 |
ProofMissing | |
42882
391e41ac038b
make sure the Vampire incomplete proof detection code kicks in
blanchet
parents:
42876
diff
changeset
|
78 |
ProofIncomplete | |
44915
635ae0a73688
simplified unsound proof detection by removing impossible case
blanchet
parents:
44784
diff
changeset
|
79 |
UnsoundProof of bool * string list | |
42587
4fbb1de05169
fixed SPASS fact offset calculation and report unexpected unsound proofs with so-called sound encodings
blanchet
parents:
42550
diff
changeset
|
80 |
CantConnect | |
4fbb1de05169
fixed SPASS fact offset calculation and report unexpected unsound proofs with so-called sound encodings
blanchet
parents:
42550
diff
changeset
|
81 |
TimedOut | |
42953
26111aafab12
detect inappropriate problems and crashes better in Waldmeister
blanchet
parents:
42943
diff
changeset
|
82 |
Inappropriate | |
42587
4fbb1de05169
fixed SPASS fact offset calculation and report unexpected unsound proofs with so-called sound encodings
blanchet
parents:
42550
diff
changeset
|
83 |
OutOfResources | |
47950
9cb132898ac8
invite users to upgrade their SPASS (so we can get rid of old code)
blanchet
parents:
47947
diff
changeset
|
84 |
OldSPASS | |
42587
4fbb1de05169
fixed SPASS fact offset calculation and report unexpected unsound proofs with so-called sound encodings
blanchet
parents:
42550
diff
changeset
|
85 |
NoPerl | |
4fbb1de05169
fixed SPASS fact offset calculation and report unexpected unsound proofs with so-called sound encodings
blanchet
parents:
42550
diff
changeset
|
86 |
NoLibwwwPerl | |
4fbb1de05169
fixed SPASS fact offset calculation and report unexpected unsound proofs with so-called sound encodings
blanchet
parents:
42550
diff
changeset
|
87 |
MalformedInput | |
4fbb1de05169
fixed SPASS fact offset calculation and report unexpected unsound proofs with so-called sound encodings
blanchet
parents:
42550
diff
changeset
|
88 |
MalformedOutput | |
4fbb1de05169
fixed SPASS fact offset calculation and report unexpected unsound proofs with so-called sound encodings
blanchet
parents:
42550
diff
changeset
|
89 |
Interrupted | |
4fbb1de05169
fixed SPASS fact offset calculation and report unexpected unsound proofs with so-called sound encodings
blanchet
parents:
42550
diff
changeset
|
90 |
Crashed | |
4fbb1de05169
fixed SPASS fact offset calculation and report unexpected unsound proofs with so-called sound encodings
blanchet
parents:
42550
diff
changeset
|
91 |
InternalError | |
4fbb1de05169
fixed SPASS fact offset calculation and report unexpected unsound proofs with so-called sound encodings
blanchet
parents:
42550
diff
changeset
|
92 |
UnknownError of string |
39491
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
93 |
|
41259
13972ced98d9
more precise error messages in "verbose" (or "debug") mode, following this morning's permission debacle
blanchet
parents:
41222
diff
changeset
|
94 |
fun short_output verbose output = |
42060
889d767ce5f4
make Minimizer honor "verbose" and "debug" options better
blanchet
parents:
41944
diff
changeset
|
95 |
if verbose then |
889d767ce5f4
make Minimizer honor "verbose" and "debug" options better
blanchet
parents:
41944
diff
changeset
|
96 |
if output = "" then "No details available" else elide_string 1000 output |
889d767ce5f4
make Minimizer honor "verbose" and "debug" options better
blanchet
parents:
41944
diff
changeset
|
97 |
else |
889d767ce5f4
make Minimizer honor "verbose" and "debug" options better
blanchet
parents:
41944
diff
changeset
|
98 |
"" |
41259
13972ced98d9
more precise error messages in "verbose" (or "debug") mode, following this morning's permission debacle
blanchet
parents:
41222
diff
changeset
|
99 |
|
41744 | 100 |
val missing_message_tail = |
101 |
" appears to be missing. You will need to install it if you want to invoke \ |
|
102 |
\remote provers." |
|
39491
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
103 |
|
42876
e336ef6313aa
more informative message when Sledgehammer finds an unsound proof
blanchet
parents:
42848
diff
changeset
|
104 |
fun involving [] = "" |
43005
c96f06bffd90
merge timeout messages from several ATPs into one message to avoid clutter
blanchet
parents:
42998
diff
changeset
|
105 |
| involving ss = |
43029
3e060b1c844b
use helpers and tweak Quickcheck's priority to it comes second (to give Solve Direct slightly more time before another prover runs)
blanchet
parents:
43005
diff
changeset
|
106 |
"involving " ^ space_implode " " (Try.serial_commas "and" (map quote ss)) ^ |
3e060b1c844b
use helpers and tweak Quickcheck's priority to it comes second (to give Solve Direct slightly more time before another prover runs)
blanchet
parents:
43005
diff
changeset
|
107 |
" " |
42876
e336ef6313aa
more informative message when Sledgehammer finds an unsound proof
blanchet
parents:
42848
diff
changeset
|
108 |
|
51998
f732a674db1b
renamed Sledgehammer functions with 'for' in their names to 'of'
blanchet
parents:
51881
diff
changeset
|
109 |
fun string_of_failure Unprovable = "The generated problem is unprovable." |
f732a674db1b
renamed Sledgehammer functions with 'for' in their names to 'of'
blanchet
parents:
51881
diff
changeset
|
110 |
| string_of_failure GaveUp = "The prover gave up." |
f732a674db1b
renamed Sledgehammer functions with 'for' in their names to 'of'
blanchet
parents:
51881
diff
changeset
|
111 |
| string_of_failure ProofMissing = |
41744 | 112 |
"The prover claims the conjecture is a theorem but did not provide a proof." |
51998
f732a674db1b
renamed Sledgehammer functions with 'for' in their names to 'of'
blanchet
parents:
51881
diff
changeset
|
113 |
| string_of_failure ProofIncomplete = |
42882
391e41ac038b
make sure the Vampire incomplete proof detection code kicks in
blanchet
parents:
42876
diff
changeset
|
114 |
"The prover claims the conjecture is a theorem but provided an incomplete \ |
46427 | 115 |
\(or unparsable) proof." |
51998
f732a674db1b
renamed Sledgehammer functions with 'for' in their names to 'of'
blanchet
parents:
51881
diff
changeset
|
116 |
| string_of_failure (UnsoundProof (false, ss)) = |
51367
4b5a5e26161d
better message (type-unsoundnesses are becoming rare, usually the issue is elsewhere, e.g. in the TSTP proof parser)
blanchet
parents:
51211
diff
changeset
|
117 |
"The prover found an unsound proof " ^ involving ss ^ |
43823 | 118 |
"(or, less likely, your axioms are inconsistent). Specify a sound type \ |
119 |
\encoding or omit the \"type_enc\" option." |
|
51998
f732a674db1b
renamed Sledgehammer functions with 'for' in their names to 'of'
blanchet
parents:
51881
diff
changeset
|
120 |
| string_of_failure (UnsoundProof (true, ss)) = |
51367
4b5a5e26161d
better message (type-unsoundnesses are becoming rare, usually the issue is elsewhere, e.g. in the TSTP proof parser)
blanchet
parents:
51211
diff
changeset
|
121 |
"The prover found an unsound proof " ^ involving ss ^ |
4b5a5e26161d
better message (type-unsoundnesses are becoming rare, usually the issue is elsewhere, e.g. in the TSTP proof parser)
blanchet
parents:
51211
diff
changeset
|
122 |
"(or, less likely, your axioms are inconsistent). Please report this to \ |
4b5a5e26161d
better message (type-unsoundnesses are becoming rare, usually the issue is elsewhere, e.g. in the TSTP proof parser)
blanchet
parents:
51211
diff
changeset
|
123 |
\the Isabelle developers." |
51998
f732a674db1b
renamed Sledgehammer functions with 'for' in their names to 'of'
blanchet
parents:
51881
diff
changeset
|
124 |
| string_of_failure CantConnect = "Cannot connect to remote server." |
f732a674db1b
renamed Sledgehammer functions with 'for' in their names to 'of'
blanchet
parents:
51881
diff
changeset
|
125 |
| string_of_failure TimedOut = "Timed out." |
f732a674db1b
renamed Sledgehammer functions with 'for' in their names to 'of'
blanchet
parents:
51881
diff
changeset
|
126 |
| string_of_failure Inappropriate = |
47506 | 127 |
"The generated problem lies outside the prover's scope." |
51998
f732a674db1b
renamed Sledgehammer functions with 'for' in their names to 'of'
blanchet
parents:
51881
diff
changeset
|
128 |
| string_of_failure OutOfResources = "The prover ran out of resources." |
f732a674db1b
renamed Sledgehammer functions with 'for' in their names to 'of'
blanchet
parents:
51881
diff
changeset
|
129 |
| string_of_failure OldSPASS = |
47950
9cb132898ac8
invite users to upgrade their SPASS (so we can get rid of old code)
blanchet
parents:
47947
diff
changeset
|
130 |
"The version of SPASS you are using is obsolete. Please upgrade to \ |
9cb132898ac8
invite users to upgrade their SPASS (so we can get rid of old code)
blanchet
parents:
47947
diff
changeset
|
131 |
\SPASS 3.8ds. To install it, download and extract the package \ |
9cb132898ac8
invite users to upgrade their SPASS (so we can get rid of old code)
blanchet
parents:
47947
diff
changeset
|
132 |
\\"http://www21.in.tum.de/~blanchet/spass-3.8ds.tar.gz\" and add the \ |
9cb132898ac8
invite users to upgrade their SPASS (so we can get rid of old code)
blanchet
parents:
47947
diff
changeset
|
133 |
\\"spass-3.8ds\" directory's absolute path to " ^ |
9cb132898ac8
invite users to upgrade their SPASS (so we can get rid of old code)
blanchet
parents:
47947
diff
changeset
|
134 |
quote (Path.implode (Path.expand (Path.appends |
9cb132898ac8
invite users to upgrade their SPASS (so we can get rid of old code)
blanchet
parents:
47947
diff
changeset
|
135 |
(Path.variable "ISABELLE_HOME_USER" :: |
9cb132898ac8
invite users to upgrade their SPASS (so we can get rid of old code)
blanchet
parents:
47947
diff
changeset
|
136 |
map Path.basic ["etc", "components"])))) ^ |
9cb132898ac8
invite users to upgrade their SPASS (so we can get rid of old code)
blanchet
parents:
47947
diff
changeset
|
137 |
" on a line of its own." |
51998
f732a674db1b
renamed Sledgehammer functions with 'for' in their names to 'of'
blanchet
parents:
51881
diff
changeset
|
138 |
| string_of_failure NoPerl = "Perl" ^ missing_message_tail |
f732a674db1b
renamed Sledgehammer functions with 'for' in their names to 'of'
blanchet
parents:
51881
diff
changeset
|
139 |
| string_of_failure NoLibwwwPerl = |
41744 | 140 |
"The Perl module \"libwww-perl\"" ^ missing_message_tail |
51998
f732a674db1b
renamed Sledgehammer functions with 'for' in their names to 'of'
blanchet
parents:
51881
diff
changeset
|
141 |
| string_of_failure MalformedInput = |
41744 | 142 |
"The generated problem is malformed. Please report this to the Isabelle \ |
143 |
\developers." |
|
51998
f732a674db1b
renamed Sledgehammer functions with 'for' in their names to 'of'
blanchet
parents:
51881
diff
changeset
|
144 |
| string_of_failure MalformedOutput = "The prover output is malformed." |
f732a674db1b
renamed Sledgehammer functions with 'for' in their names to 'of'
blanchet
parents:
51881
diff
changeset
|
145 |
| string_of_failure Interrupted = "The prover was interrupted." |
f732a674db1b
renamed Sledgehammer functions with 'for' in their names to 'of'
blanchet
parents:
51881
diff
changeset
|
146 |
| string_of_failure Crashed = "The prover crashed." |
f732a674db1b
renamed Sledgehammer functions with 'for' in their names to 'of'
blanchet
parents:
51881
diff
changeset
|
147 |
| string_of_failure InternalError = "An internal prover error occurred." |
f732a674db1b
renamed Sledgehammer functions with 'for' in their names to 'of'
blanchet
parents:
51881
diff
changeset
|
148 |
| string_of_failure (UnknownError s) = |
41744 | 149 |
"A prover error occurred" ^ |
51998
f732a674db1b
renamed Sledgehammer functions with 'for' in their names to 'of'
blanchet
parents:
51881
diff
changeset
|
150 |
(if s = "" then ". (Pass the \"verbose\" option for details.)" |
f732a674db1b
renamed Sledgehammer functions with 'for' in their names to 'of'
blanchet
parents:
51881
diff
changeset
|
151 |
else ":\n" ^ s) |
39491
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
152 |
|
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
153 |
fun extract_delimited (begin_delim, end_delim) output = |
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
154 |
output |> first_field begin_delim |> the |> snd |
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
155 |
|> first_field end_delim |> the |> fst |
48539 | 156 |
|> perhaps (try (first_field "\n" #> the #> snd)) |
39491
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
157 |
handle Option.Option => "" |
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
158 |
|
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
159 |
val tstp_important_message_delims = |
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
160 |
("% SZS start RequiredInformation", "% SZS end RequiredInformation") |
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
161 |
|
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
162 |
fun extract_important_message output = |
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
163 |
case extract_delimited tstp_important_message_delims output of |
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
164 |
"" => "" |
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
165 |
| s => s |> space_explode "\n" |> filter_out (curry (op =) "") |
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
166 |
|> map (perhaps (try (unprefix "%"))) |
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
167 |
|> map (perhaps (try (unprefix " "))) |
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
168 |
|> space_implode "\n " |> quote |
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
169 |
|
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
170 |
(* Splits by the first possible of a list of delimiters. *) |
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
171 |
fun extract_tstplike_proof delims output = |
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
172 |
case pairself (find_first (fn s => String.isSubstring s output)) |
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
173 |
(ListPair.unzip delims) of |
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
174 |
(SOME begin_delim, SOME end_delim) => |
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
175 |
extract_delimited (begin_delim, end_delim) output |
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
176 |
| _ => "" |
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
177 |
|
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
178 |
fun extract_known_failure known_failures output = |
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
179 |
known_failures |
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
180 |
|> find_first (fn (_, pattern) => String.isSubstring pattern output) |
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
181 |
|> Option.map fst |
2416666e6f94
refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents:
39457
diff
changeset
|
182 |
|
48716
1d2a12bb0640
stop distinguishing between complete and incomplete slices, since this is very fragile and has hardly any useful semantics to users
blanchet
parents:
48700
diff
changeset
|
183 |
fun extract_tstplike_proof_and_outcome verbose proof_delims known_failures |
1d2a12bb0640
stop distinguishing between complete and incomplete slices, since this is very fragile and has hardly any useful semantics to users
blanchet
parents:
48700
diff
changeset
|
184 |
output = |
42882
391e41ac038b
make sure the Vampire incomplete proof detection code kicks in
blanchet
parents:
42876
diff
changeset
|
185 |
case (extract_tstplike_proof proof_delims output, |
391e41ac038b
make sure the Vampire incomplete proof detection code kicks in
blanchet
parents:
42876
diff
changeset
|
186 |
extract_known_failure known_failures output) of |
48700 | 187 |
(_, SOME ProofIncomplete) => ("", NONE) |
43246 | 188 |
| ("", SOME ProofMissing) => ("", NONE) |
189 |
| ("", NONE) => ("", SOME (UnknownError (short_output verbose output))) |
|
48716
1d2a12bb0640
stop distinguishing between complete and incomplete slices, since this is very fragile and has hardly any useful semantics to users
blanchet
parents:
48700
diff
changeset
|
190 |
| res as ("", _) => res |
42882
391e41ac038b
make sure the Vampire incomplete proof detection code kicks in
blanchet
parents:
42876
diff
changeset
|
191 |
| (tstplike_proof, _) => (tstplike_proof, NONE) |
39452 | 192 |
|
45551 | 193 |
type step_name = string * string list |
39452 | 194 |
|
42968
74415622d293
more work on parsing LEO-II proofs and extracting uses of extensionality
blanchet
parents:
42966
diff
changeset
|
195 |
fun is_same_atp_step (s1, _) (s2, _) = s1 = s2 |
74415622d293
more work on parsing LEO-II proofs and extracting uses of extensionality
blanchet
parents:
42966
diff
changeset
|
196 |
|
74415622d293
more work on parsing LEO-II proofs and extracting uses of extensionality
blanchet
parents:
42966
diff
changeset
|
197 |
val vampire_fact_prefix = "f" |
39452 | 198 |
|
52755
4183c3219745
simplified Vampire hack -- no need to run it for other ATPs
blanchet
parents:
52077
diff
changeset
|
199 |
fun vampire_step_name_ord p = |
39455 | 200 |
let val q = pairself fst p in |
52755
4183c3219745
simplified Vampire hack -- no need to run it for other ATPs
blanchet
parents:
52077
diff
changeset
|
201 |
(* The "unprefix" part is to cope with Vampire's output. *) |
42968
74415622d293
more work on parsing LEO-II proofs and extracting uses of extensionality
blanchet
parents:
42966
diff
changeset
|
202 |
case pairself (Int.fromString |
74415622d293
more work on parsing LEO-II proofs and extracting uses of extensionality
blanchet
parents:
42966
diff
changeset
|
203 |
o perhaps (try (unprefix vampire_fact_prefix))) q of |
52755
4183c3219745
simplified Vampire hack -- no need to run it for other ATPs
blanchet
parents:
52077
diff
changeset
|
204 |
(SOME i, SOME j) => int_ord (i, j) |
4183c3219745
simplified Vampire hack -- no need to run it for other ATPs
blanchet
parents:
52077
diff
changeset
|
205 |
| _ => raise Fail "not Vampire" |
39452 | 206 |
end |
207 |
||
51201 | 208 |
type 'a step = step_name * formula_role * 'a * string * step_name list |
39452 | 209 |
|
48135 | 210 |
type 'a proof = ('a, 'a, ('a, 'a) ho_term, 'a) formula step list |
39452 | 211 |
|
212 |
(**** PARSING OF TSTP FORMAT ****) |
|
213 |
||
42536 | 214 |
(* Strings enclosed in single quotes (e.g., file names) *) |
39452 | 215 |
val scan_general_id = |
47917
b287682bf917
improve parsing of Waldmeister dependencies (and kill obsolete hack)
blanchet
parents:
47787
diff
changeset
|
216 |
$$ "'" |-- Scan.repeat (~$$ "'") --| $$ "'" >> implode |
51881 | 217 |
|| Scan.repeat ($$ "$") -- Scan.many1 Symbol.is_letdig |
39452 | 218 |
>> (fn (ss1, ss2) => implode ss1 ^ implode ss2) |
219 |
||
51031
63d71b247323
more robustness in Isar proof reconstruction (cf. bug report by Ondrej)
blanchet
parents:
50704
diff
changeset
|
220 |
val scan_nat = Scan.repeat1 (Scan.one Symbol.is_ascii_digit) >> implode |
63d71b247323
more robustness in Isar proof reconstruction (cf. bug report by Ondrej)
blanchet
parents:
50704
diff
changeset
|
221 |
|
45235
7187bce94e88
more robust parsing of TSTP sources -- Vampire has nonstandard "introduced()" tags and Waldmeister(OnTPTP) has weird "theory(...)" dependencies
blanchet
parents:
45209
diff
changeset
|
222 |
val skip_term = |
45208
9a00f9cc8707
marginally cleaner proof parsing, that doesn't stumble upon LEO-II's E-step proofs
blanchet
parents:
45203
diff
changeset
|
223 |
let |
45235
7187bce94e88
more robust parsing of TSTP sources -- Vampire has nonstandard "introduced()" tags and Waldmeister(OnTPTP) has weird "theory(...)" dependencies
blanchet
parents:
45209
diff
changeset
|
224 |
fun skip _ accum [] = (accum, []) |
7187bce94e88
more robust parsing of TSTP sources -- Vampire has nonstandard "introduced()" tags and Waldmeister(OnTPTP) has weird "theory(...)" dependencies
blanchet
parents:
45209
diff
changeset
|
225 |
| skip 0 accum (ss as "," :: _) = (accum, ss) |
7187bce94e88
more robust parsing of TSTP sources -- Vampire has nonstandard "introduced()" tags and Waldmeister(OnTPTP) has weird "theory(...)" dependencies
blanchet
parents:
45209
diff
changeset
|
226 |
| skip 0 accum (ss as ")" :: _) = (accum, ss) |
7187bce94e88
more robust parsing of TSTP sources -- Vampire has nonstandard "introduced()" tags and Waldmeister(OnTPTP) has weird "theory(...)" dependencies
blanchet
parents:
45209
diff
changeset
|
227 |
| skip 0 accum (ss as "]" :: _) = (accum, ss) |
7187bce94e88
more robust parsing of TSTP sources -- Vampire has nonstandard "introduced()" tags and Waldmeister(OnTPTP) has weird "theory(...)" dependencies
blanchet
parents:
45209
diff
changeset
|
228 |
| skip n accum ((s as "(") :: ss) = skip (n + 1) (s :: accum) ss |
7187bce94e88
more robust parsing of TSTP sources -- Vampire has nonstandard "introduced()" tags and Waldmeister(OnTPTP) has weird "theory(...)" dependencies
blanchet
parents:
45209
diff
changeset
|
229 |
| skip n accum ((s as "[") :: ss) = skip (n + 1) (s :: accum) ss |
7187bce94e88
more robust parsing of TSTP sources -- Vampire has nonstandard "introduced()" tags and Waldmeister(OnTPTP) has weird "theory(...)" dependencies
blanchet
parents:
45209
diff
changeset
|
230 |
| skip n accum ((s as "]") :: ss) = skip (n - 1) (s :: accum) ss |
7187bce94e88
more robust parsing of TSTP sources -- Vampire has nonstandard "introduced()" tags and Waldmeister(OnTPTP) has weird "theory(...)" dependencies
blanchet
parents:
45209
diff
changeset
|
231 |
| skip n accum ((s as ")") :: ss) = skip (n - 1) (s :: accum) ss |
7187bce94e88
more robust parsing of TSTP sources -- Vampire has nonstandard "introduced()" tags and Waldmeister(OnTPTP) has weird "theory(...)" dependencies
blanchet
parents:
45209
diff
changeset
|
232 |
| skip n accum (s :: ss) = skip n (s :: accum) ss |
7187bce94e88
more robust parsing of TSTP sources -- Vampire has nonstandard "introduced()" tags and Waldmeister(OnTPTP) has weird "theory(...)" dependencies
blanchet
parents:
45209
diff
changeset
|
233 |
in skip 0 [] #>> (rev #> implode) end |
45208
9a00f9cc8707
marginally cleaner proof parsing, that doesn't stumble upon LEO-II's E-step proofs
blanchet
parents:
45203
diff
changeset
|
234 |
|
9a00f9cc8707
marginally cleaner proof parsing, that doesn't stumble upon LEO-II's E-step proofs
blanchet
parents:
45203
diff
changeset
|
235 |
datatype source = |
9a00f9cc8707
marginally cleaner proof parsing, that doesn't stumble upon LEO-II's E-step proofs
blanchet
parents:
45203
diff
changeset
|
236 |
File_Source of string * string option | |
45209 | 237 |
Inference_Source of string * string list |
45208
9a00f9cc8707
marginally cleaner proof parsing, that doesn't stumble upon LEO-II's E-step proofs
blanchet
parents:
45203
diff
changeset
|
238 |
|
48132
9aa0fad4e864
added type arguments to "ATerm" constructor -- but don't use them yet
blanchet
parents:
48130
diff
changeset
|
239 |
val dummy_phi = AAtom (ATerm (("", []), [])) |
45235
7187bce94e88
more robust parsing of TSTP sources -- Vampire has nonstandard "introduced()" tags and Waldmeister(OnTPTP) has weird "theory(...)" dependencies
blanchet
parents:
45209
diff
changeset
|
240 |
val dummy_inference = Inference_Source ("", []) |
7187bce94e88
more robust parsing of TSTP sources -- Vampire has nonstandard "introduced()" tags and Waldmeister(OnTPTP) has weird "theory(...)" dependencies
blanchet
parents:
45209
diff
changeset
|
241 |
|
47917
b287682bf917
improve parsing of Waldmeister dependencies (and kill obsolete hack)
blanchet
parents:
47787
diff
changeset
|
242 |
(* "skip_term" is there to cope with Waldmeister nonsense such as |
b287682bf917
improve parsing of Waldmeister dependencies (and kill obsolete hack)
blanchet
parents:
47787
diff
changeset
|
243 |
"theory(equality)". *) |
50011 | 244 |
fun parse_dependency x = |
245 |
(parse_inference_source >> snd |
|
246 |
|| scan_general_id --| skip_term >> single) x |
|
247 |
and parse_dependencies x = |
|
248 |
(parse_dependency ::: Scan.repeat ($$ "," |-- parse_dependency) |
|
249 |
>> flat) x |
|
250 |
and parse_file_source x = |
|
251 |
(Scan.this_string "file" |-- $$ "(" |-- scan_general_id |
|
252 |
-- Scan.option ($$ "," |-- scan_general_id) --| $$ ")") x |
|
253 |
and parse_inference_source x = |
|
254 |
(Scan.this_string "inference" |-- $$ "(" |-- scan_general_id |
|
255 |
--| skip_term --| $$ "," --| skip_term --| $$ "," --| $$ "[" |
|
256 |
-- parse_dependencies --| $$ "]" --| $$ ")") x |
|
51031
63d71b247323
more robustness in Isar proof reconstruction (cf. bug report by Ondrej)
blanchet
parents:
50704
diff
changeset
|
257 |
and skip_introduced x = |
63d71b247323
more robustness in Isar proof reconstruction (cf. bug report by Ondrej)
blanchet
parents:
50704
diff
changeset
|
258 |
(Scan.this_string "introduced" |-- $$ "(" |-- skip_term --| $$ ")") x |
50011 | 259 |
and parse_source x = |
260 |
(parse_file_source >> File_Source |
|
261 |
|| parse_inference_source >> Inference_Source |
|
51031
63d71b247323
more robustness in Isar proof reconstruction (cf. bug report by Ondrej)
blanchet
parents:
50704
diff
changeset
|
262 |
|| skip_introduced >> K dummy_inference (* for Vampire *) |
52756
1ac8a0d0ddb1
parse nonnumeric identifiers in E proofs correctly
blanchet
parents:
52755
diff
changeset
|
263 |
|| scan_general_id >> (fn s => Inference_Source ("", [s])) (* for E *) |
45235
7187bce94e88
more robust parsing of TSTP sources -- Vampire has nonstandard "introduced()" tags and Waldmeister(OnTPTP) has weird "theory(...)" dependencies
blanchet
parents:
45209
diff
changeset
|
264 |
|| skip_term >> K dummy_inference) x |
39452 | 265 |
|
42966
4e2d6c1e5392
more work on parsing LEO-II proofs without lambdas
blanchet
parents:
42965
diff
changeset
|
266 |
fun list_app (f, args) = |
48132
9aa0fad4e864
added type arguments to "ATerm" constructor -- but don't use them yet
blanchet
parents:
48130
diff
changeset
|
267 |
fold (fn arg => fn f => ATerm ((tptp_app, []), [f, arg])) args f |
42966
4e2d6c1e5392
more work on parsing LEO-II proofs without lambdas
blanchet
parents:
42965
diff
changeset
|
268 |
|
45881 | 269 |
(* We currently ignore TFF and THF types. *) |
42968
74415622d293
more work on parsing LEO-II proofs and extracting uses of extensionality
blanchet
parents:
42966
diff
changeset
|
270 |
fun parse_type_stuff x = |
74415622d293
more work on parsing LEO-II proofs and extracting uses of extensionality
blanchet
parents:
42966
diff
changeset
|
271 |
Scan.repeat (($$ tptp_has_type || $$ tptp_fun_type) |-- parse_arg) x |
74415622d293
more work on parsing LEO-II proofs and extracting uses of extensionality
blanchet
parents:
42966
diff
changeset
|
272 |
and parse_arg x = |
74415622d293
more work on parsing LEO-II proofs and extracting uses of extensionality
blanchet
parents:
42966
diff
changeset
|
273 |
($$ "(" |-- parse_term --| $$ ")" --| parse_type_stuff |
74415622d293
more work on parsing LEO-II proofs and extracting uses of extensionality
blanchet
parents:
42966
diff
changeset
|
274 |
|| scan_general_id --| parse_type_stuff |
74415622d293
more work on parsing LEO-II proofs and extracting uses of extensionality
blanchet
parents:
42966
diff
changeset
|
275 |
-- Scan.optional ($$ "(" |-- parse_terms --| $$ ")") [] |
48132
9aa0fad4e864
added type arguments to "ATerm" constructor -- but don't use them yet
blanchet
parents:
48130
diff
changeset
|
276 |
>> (ATerm o apfst (rpair []))) x |
45881 | 277 |
and parse_term x = |
42968
74415622d293
more work on parsing LEO-II proofs and extracting uses of extensionality
blanchet
parents:
42966
diff
changeset
|
278 |
(parse_arg -- Scan.repeat ($$ tptp_app |-- parse_arg) >> list_app) x |
42966
4e2d6c1e5392
more work on parsing LEO-II proofs without lambdas
blanchet
parents:
42965
diff
changeset
|
279 |
and parse_terms x = |
4e2d6c1e5392
more work on parsing LEO-II proofs without lambdas
blanchet
parents:
42965
diff
changeset
|
280 |
(parse_term ::: Scan.repeat ($$ "," |-- parse_term)) x |
39452 | 281 |
|
39598 | 282 |
fun parse_atom x = |
42968
74415622d293
more work on parsing LEO-II proofs and extracting uses of extensionality
blanchet
parents:
42966
diff
changeset
|
283 |
(parse_term -- Scan.option (Scan.option ($$ tptp_not_infix) --| $$ tptp_equal |
74415622d293
more work on parsing LEO-II proofs and extracting uses of extensionality
blanchet
parents:
42966
diff
changeset
|
284 |
-- parse_term) |
39598 | 285 |
>> (fn (u1, NONE) => AAtom u1 |
45881 | 286 |
| (u1, SOME (neg, u2)) => |
48132
9aa0fad4e864
added type arguments to "ATerm" constructor -- but don't use them yet
blanchet
parents:
48130
diff
changeset
|
287 |
AAtom (ATerm (("equal", []), [u1, u2])) |> is_some neg ? mk_anot)) x |
39452 | 288 |
|
289 |
(* TPTP formulas are fully parenthesized, so we don't need to worry about |
|
290 |
operator precedence. *) |
|
42605
8734eb0033b3
Vampire sometimes generates formulas with ~ (not) followed by a quantified subformula, without parentheses -- parse these correctly
blanchet
parents:
42603
diff
changeset
|
291 |
fun parse_literal x = |
42968
74415622d293
more work on parsing LEO-II proofs and extracting uses of extensionality
blanchet
parents:
42966
diff
changeset
|
292 |
((Scan.repeat ($$ tptp_not) >> length) |
42605
8734eb0033b3
Vampire sometimes generates formulas with ~ (not) followed by a quantified subformula, without parentheses -- parse these correctly
blanchet
parents:
42603
diff
changeset
|
293 |
-- ($$ "(" |-- parse_formula --| $$ ")" |
8734eb0033b3
Vampire sometimes generates formulas with ~ (not) followed by a quantified subformula, without parentheses -- parse these correctly
blanchet
parents:
42603
diff
changeset
|
294 |
|| parse_quantified_formula |
8734eb0033b3
Vampire sometimes generates formulas with ~ (not) followed by a quantified subformula, without parentheses -- parse these correctly
blanchet
parents:
42603
diff
changeset
|
295 |
|| parse_atom) |
8734eb0033b3
Vampire sometimes generates formulas with ~ (not) followed by a quantified subformula, without parentheses -- parse these correctly
blanchet
parents:
42603
diff
changeset
|
296 |
>> (fn (n, phi) => phi |> n mod 2 = 1 ? mk_anot)) x |
8734eb0033b3
Vampire sometimes generates formulas with ~ (not) followed by a quantified subformula, without parentheses -- parse these correctly
blanchet
parents:
42603
diff
changeset
|
297 |
and parse_formula x = |
8734eb0033b3
Vampire sometimes generates formulas with ~ (not) followed by a quantified subformula, without parentheses -- parse these correctly
blanchet
parents:
42603
diff
changeset
|
298 |
(parse_literal |
43163 | 299 |
-- Scan.option ((Scan.this_string tptp_implies |
300 |
|| Scan.this_string tptp_iff |
|
301 |
|| Scan.this_string tptp_not_iff |
|
302 |
|| Scan.this_string tptp_if |
|
303 |
|| $$ tptp_or |
|
304 |
|| $$ tptp_and) -- parse_formula) |
|
39452 | 305 |
>> (fn (phi1, NONE) => phi1 |
43163 | 306 |
| (phi1, SOME (c, phi2)) => |
307 |
if c = tptp_implies then mk_aconn AImplies phi1 phi2 |
|
308 |
else if c = tptp_iff then mk_aconn AIff phi1 phi2 |
|
309 |
else if c = tptp_not_iff then mk_anot (mk_aconn AIff phi1 phi2) |
|
310 |
else if c = tptp_if then mk_aconn AImplies phi2 phi1 |
|
311 |
else if c = tptp_or then mk_aconn AOr phi1 phi2 |
|
312 |
else if c = tptp_and then mk_aconn AAnd phi1 phi2 |
|
313 |
else raise Fail ("impossible connective " ^ quote c))) x |
|
42605
8734eb0033b3
Vampire sometimes generates formulas with ~ (not) followed by a quantified subformula, without parentheses -- parse these correctly
blanchet
parents:
42603
diff
changeset
|
314 |
and parse_quantified_formula x = |
42968
74415622d293
more work on parsing LEO-II proofs and extracting uses of extensionality
blanchet
parents:
42966
diff
changeset
|
315 |
(($$ tptp_forall >> K AForall || $$ tptp_exists >> K AExists) |
42605
8734eb0033b3
Vampire sometimes generates formulas with ~ (not) followed by a quantified subformula, without parentheses -- parse these correctly
blanchet
parents:
42603
diff
changeset
|
316 |
--| $$ "[" -- parse_terms --| $$ "]" --| $$ ":" -- parse_literal |
8734eb0033b3
Vampire sometimes generates formulas with ~ (not) followed by a quantified subformula, without parentheses -- parse these correctly
blanchet
parents:
42603
diff
changeset
|
317 |
>> (fn ((q, ts), phi) => |
42966
4e2d6c1e5392
more work on parsing LEO-II proofs without lambdas
blanchet
parents:
42965
diff
changeset
|
318 |
(* We ignore TFF and THF types for now. *) |
48132
9aa0fad4e864
added type arguments to "ATerm" constructor -- but don't use them yet
blanchet
parents:
48130
diff
changeset
|
319 |
AQuant (q, map (fn ATerm ((s, []), _) => (s, NONE)) ts, phi))) x |
39452 | 320 |
|
321 |
val parse_tstp_extra_arguments = |
|
45235
7187bce94e88
more robust parsing of TSTP sources -- Vampire has nonstandard "introduced()" tags and Waldmeister(OnTPTP) has weird "theory(...)" dependencies
blanchet
parents:
45209
diff
changeset
|
322 |
Scan.optional ($$ "," |-- parse_source --| Scan.option ($$ "," |-- skip_term)) |
7187bce94e88
more robust parsing of TSTP sources -- Vampire has nonstandard "introduced()" tags and Waldmeister(OnTPTP) has weird "theory(...)" dependencies
blanchet
parents:
45209
diff
changeset
|
323 |
dummy_inference |
39452 | 324 |
|
47927
c35238d19bb9
repair the Waldmeister endgame only for Waldmeister proofs
blanchet
parents:
47926
diff
changeset
|
325 |
val waldmeister_conjecture_name = "conjecture_1" |
42943 | 326 |
|
42536 | 327 |
val tofof_fact_prefix = "fof_" |
41203
1393514094d7
fixed more issues with the Vampire output parser, and added support for Vampire's TSTP output (--proof tptp)
blanchet
parents:
41201
diff
changeset
|
328 |
|
42943 | 329 |
fun is_same_term subst tm1 tm2 = |
330 |
let |
|
331 |
fun do_term_pair _ NONE = NONE |
|
48132
9aa0fad4e864
added type arguments to "ATerm" constructor -- but don't use them yet
blanchet
parents:
48130
diff
changeset
|
332 |
| do_term_pair (ATerm ((s1, _), tm1), ATerm ((s2, _), tm2)) (SOME subst) = |
42998
1c80902d0456
fully support all type system encodings in typed formats (TFF, THF)
blanchet
parents:
42975
diff
changeset
|
333 |
case pairself is_tptp_variable (s1, s2) of |
42943 | 334 |
(true, true) => |
335 |
(case AList.lookup (op =) subst s1 of |
|
336 |
SOME s2' => if s2' = s2 then SOME subst else NONE |
|
337 |
| NONE => |
|
338 |
if null (AList.find (op =) subst s2) then SOME ((s1, s2) :: subst) |
|
339 |
else NONE) |
|
340 |
| (false, false) => |
|
341 |
if s1 = s2 andalso length tm1 = length tm2 then |
|
342 |
SOME subst |> fold do_term_pair (tm1 ~~ tm2) |
|
343 |
else |
|
344 |
NONE |
|
345 |
| _ => NONE |
|
346 |
in SOME subst |> do_term_pair (tm1, tm2) |> is_some end |
|
347 |
||
47921
fc26d5538868
ensure the "show" equation is not reoriented by Waldmeister
blanchet
parents:
47917
diff
changeset
|
348 |
fun is_same_formula comm subst (AQuant (q1, xs1, phi1)) (AQuant (q2, xs2, phi2)) = |
42943 | 349 |
q1 = q2 andalso length xs1 = length xs2 andalso |
47921
fc26d5538868
ensure the "show" equation is not reoriented by Waldmeister
blanchet
parents:
47917
diff
changeset
|
350 |
is_same_formula comm ((map fst xs1 ~~ map fst xs2) @ subst) phi1 phi2 |
fc26d5538868
ensure the "show" equation is not reoriented by Waldmeister
blanchet
parents:
47917
diff
changeset
|
351 |
| is_same_formula comm subst (AConn (c1, phis1)) (AConn (c2, phis2)) = |
42943 | 352 |
c1 = c2 andalso length phis1 = length phis2 andalso |
47921
fc26d5538868
ensure the "show" equation is not reoriented by Waldmeister
blanchet
parents:
47917
diff
changeset
|
353 |
forall (uncurry (is_same_formula comm subst)) (phis1 ~~ phis2) |
48132
9aa0fad4e864
added type arguments to "ATerm" constructor -- but don't use them yet
blanchet
parents:
48130
diff
changeset
|
354 |
| is_same_formula comm subst |
9aa0fad4e864
added type arguments to "ATerm" constructor -- but don't use them yet
blanchet
parents:
48130
diff
changeset
|
355 |
(AAtom (tm1 as ATerm (("equal", []), [tm11, tm12]))) (AAtom tm2) = |
47926 | 356 |
is_same_term subst tm1 tm2 orelse |
48132
9aa0fad4e864
added type arguments to "ATerm" constructor -- but don't use them yet
blanchet
parents:
48130
diff
changeset
|
357 |
(comm andalso is_same_term subst (ATerm (("equal", []), [tm12, tm11])) tm2) |
47921
fc26d5538868
ensure the "show" equation is not reoriented by Waldmeister
blanchet
parents:
47917
diff
changeset
|
358 |
| is_same_formula _ subst (AAtom tm1) (AAtom tm2) = is_same_term subst tm1 tm2 |
fc26d5538868
ensure the "show" equation is not reoriented by Waldmeister
blanchet
parents:
47917
diff
changeset
|
359 |
| is_same_formula _ _ _ _ = false |
42943 | 360 |
|
50521
bec828f3364e
generate comments with original names for debugging
blanchet
parents:
50236
diff
changeset
|
361 |
fun matching_formula_line_identifier phi (Formula ((ident, _), _, phi', _, _)) = |
47921
fc26d5538868
ensure the "show" equation is not reoriented by Waldmeister
blanchet
parents:
47917
diff
changeset
|
362 |
if is_same_formula true [] phi phi' then SOME (ident, phi') else NONE |
42943 | 363 |
| matching_formula_line_identifier _ _ = NONE |
364 |
||
365 |
fun find_formula_in_problem problem phi = |
|
366 |
problem |> maps snd |> map_filter (matching_formula_line_identifier phi) |
|
45551 | 367 |
|> try (single o hd) |> the_default [] |
42943 | 368 |
|
48132
9aa0fad4e864
added type arguments to "ATerm" constructor -- but don't use them yet
blanchet
parents:
48130
diff
changeset
|
369 |
fun commute_eq (AAtom (ATerm ((s, []), tms))) = AAtom (ATerm ((s, []), rev tms)) |
47946
33afcfad3f8d
add an experimental "aggressive" mode to Sledgehammer, to experiment with more complete translations of higher-order features without breaking "metis"
blanchet
parents:
47927
diff
changeset
|
370 |
| commute_eq _ = raise Fail "expected equation" |
47921
fc26d5538868
ensure the "show" equation is not reoriented by Waldmeister
blanchet
parents:
47917
diff
changeset
|
371 |
|
50012
01cb92151a53
track formula roles in proofs and use that to determine whether the conjecture should be negated or not
blanchet
parents:
50011
diff
changeset
|
372 |
fun role_of_tptp_string "axiom" = Axiom |
01cb92151a53
track formula roles in proofs and use that to determine whether the conjecture should be negated or not
blanchet
parents:
50011
diff
changeset
|
373 |
| role_of_tptp_string "definition" = Definition |
01cb92151a53
track formula roles in proofs and use that to determine whether the conjecture should be negated or not
blanchet
parents:
50011
diff
changeset
|
374 |
| role_of_tptp_string "lemma" = Lemma |
01cb92151a53
track formula roles in proofs and use that to determine whether the conjecture should be negated or not
blanchet
parents:
50011
diff
changeset
|
375 |
| role_of_tptp_string "hypothesis" = Hypothesis |
01cb92151a53
track formula roles in proofs and use that to determine whether the conjecture should be negated or not
blanchet
parents:
50011
diff
changeset
|
376 |
| role_of_tptp_string "conjecture" = Conjecture |
01cb92151a53
track formula roles in proofs and use that to determine whether the conjecture should be negated or not
blanchet
parents:
50011
diff
changeset
|
377 |
| role_of_tptp_string "negated_conjecture" = Negated_Conjecture |
01cb92151a53
track formula roles in proofs and use that to determine whether the conjecture should be negated or not
blanchet
parents:
50011
diff
changeset
|
378 |
| role_of_tptp_string "plain" = Plain |
01cb92151a53
track formula roles in proofs and use that to determine whether the conjecture should be negated or not
blanchet
parents:
50011
diff
changeset
|
379 |
| role_of_tptp_string _ = Unknown |
01cb92151a53
track formula roles in proofs and use that to determine whether the conjecture should be negated or not
blanchet
parents:
50011
diff
changeset
|
380 |
|
42962
3b50fdeb6cfc
started adding support for THF output (but no lambdas)
blanchet
parents:
42961
diff
changeset
|
381 |
(* Syntax: (cnf|fof|tff|thf)\(<num>, <formula_role>, |
3b50fdeb6cfc
started adding support for THF output (but no lambdas)
blanchet
parents:
42961
diff
changeset
|
382 |
<formula> <extra_arguments>\). |
39452 | 383 |
The <num> could be an identifier, but we assume integers. *) |
42943 | 384 |
fun parse_tstp_line problem = |
42968
74415622d293
more work on parsing LEO-II proofs and extracting uses of extensionality
blanchet
parents:
42966
diff
changeset
|
385 |
((Scan.this_string tptp_cnf || Scan.this_string tptp_fof |
74415622d293
more work on parsing LEO-II proofs and extracting uses of extensionality
blanchet
parents:
42966
diff
changeset
|
386 |
|| Scan.this_string tptp_tff || Scan.this_string tptp_thf) -- $$ "(") |
50236 | 387 |
|-- scan_general_id --| $$ "," -- Symbol.scan_ascii_id --| $$ "," |
45235
7187bce94e88
more robust parsing of TSTP sources -- Vampire has nonstandard "introduced()" tags and Waldmeister(OnTPTP) has weird "theory(...)" dependencies
blanchet
parents:
45209
diff
changeset
|
388 |
-- (parse_formula || skip_term >> K dummy_phi) -- parse_tstp_extra_arguments |
7187bce94e88
more robust parsing of TSTP sources -- Vampire has nonstandard "introduced()" tags and Waldmeister(OnTPTP) has weird "theory(...)" dependencies
blanchet
parents:
45209
diff
changeset
|
389 |
--| $$ ")" --| $$ "." |
42943 | 390 |
>> (fn (((num, role), phi), deps) => |
391 |
let |
|
47921
fc26d5538868
ensure the "show" equation is not reoriented by Waldmeister
blanchet
parents:
47917
diff
changeset
|
392 |
val ((name, phi), rule, deps) = |
42943 | 393 |
(* Waldmeister isn't exactly helping. *) |
394 |
case deps of |
|
45208
9a00f9cc8707
marginally cleaner proof parsing, that doesn't stumble upon LEO-II's E-step proofs
blanchet
parents:
45203
diff
changeset
|
395 |
File_Source (_, SOME s) => |
47927
c35238d19bb9
repair the Waldmeister endgame only for Waldmeister proofs
blanchet
parents:
47926
diff
changeset
|
396 |
(if s = waldmeister_conjecture_name then |
47921
fc26d5538868
ensure the "show" equation is not reoriented by Waldmeister
blanchet
parents:
47917
diff
changeset
|
397 |
case find_formula_in_problem problem (mk_anot phi) of |
fc26d5538868
ensure the "show" equation is not reoriented by Waldmeister
blanchet
parents:
47917
diff
changeset
|
398 |
(* Waldmeister hack: Get the original orientation of the |
fc26d5538868
ensure the "show" equation is not reoriented by Waldmeister
blanchet
parents:
47917
diff
changeset
|
399 |
equation to avoid confusing Isar. *) |
fc26d5538868
ensure the "show" equation is not reoriented by Waldmeister
blanchet
parents:
47917
diff
changeset
|
400 |
[(s, phi')] => |
fc26d5538868
ensure the "show" equation is not reoriented by Waldmeister
blanchet
parents:
47917
diff
changeset
|
401 |
((num, [s]), |
47926 | 402 |
phi |> not (is_same_formula false [] (mk_anot phi) phi') |
47921
fc26d5538868
ensure the "show" equation is not reoriented by Waldmeister
blanchet
parents:
47917
diff
changeset
|
403 |
? commute_eq) |
fc26d5538868
ensure the "show" equation is not reoriented by Waldmeister
blanchet
parents:
47917
diff
changeset
|
404 |
| _ => ((num, []), phi) |
fc26d5538868
ensure the "show" equation is not reoriented by Waldmeister
blanchet
parents:
47917
diff
changeset
|
405 |
else |
fc26d5538868
ensure the "show" equation is not reoriented by Waldmeister
blanchet
parents:
47917
diff
changeset
|
406 |
((num, [s |> perhaps (try (unprefix tofof_fact_prefix))]), |
fc26d5538868
ensure the "show" equation is not reoriented by Waldmeister
blanchet
parents:
47917
diff
changeset
|
407 |
phi), |
fc26d5538868
ensure the "show" equation is not reoriented by Waldmeister
blanchet
parents:
47917
diff
changeset
|
408 |
"", []) |
45208
9a00f9cc8707
marginally cleaner proof parsing, that doesn't stumble upon LEO-II's E-step proofs
blanchet
parents:
45203
diff
changeset
|
409 |
| File_Source _ => |
47921
fc26d5538868
ensure the "show" equation is not reoriented by Waldmeister
blanchet
parents:
47917
diff
changeset
|
410 |
(((num, phi |> find_formula_in_problem problem |> map fst), |
fc26d5538868
ensure the "show" equation is not reoriented by Waldmeister
blanchet
parents:
47917
diff
changeset
|
411 |
phi), "", []) |
fc26d5538868
ensure the "show" equation is not reoriented by Waldmeister
blanchet
parents:
47917
diff
changeset
|
412 |
| Inference_Source (rule, deps) => (((num, []), phi), rule, deps) |
47787 | 413 |
fun mk_step () = |
51201 | 414 |
(name, role_of_tptp_string role, phi, rule, map (rpair []) deps) |
42943 | 415 |
in |
50012
01cb92151a53
track formula roles in proofs and use that to determine whether the conjecture should be negated or not
blanchet
parents:
50011
diff
changeset
|
416 |
case role_of_tptp_string role of |
01cb92151a53
track formula roles in proofs and use that to determine whether the conjecture should be negated or not
blanchet
parents:
50011
diff
changeset
|
417 |
Definition => |
42943 | 418 |
(case phi of |
51198 | 419 |
AAtom (ATerm (("equal", []), _)) => |
42943 | 420 |
(* Vampire's equality proxy axiom *) |
51201 | 421 |
(name, Definition, phi, rule, map (rpair []) deps) |
47787 | 422 |
| _ => mk_step ()) |
423 |
| _ => mk_step () |
|
42943 | 424 |
end) |
39452 | 425 |
|
426 |
(**** PARSING OF SPASS OUTPUT ****) |
|
427 |
||
428 |
(* SPASS returns clause references of the form "x.y". We ignore "y", whose role |
|
429 |
is not clear anyway. *) |
|
430 |
val parse_dot_name = scan_general_id --| $$ "." --| scan_general_id |
|
431 |
||
432 |
val parse_spass_annotations = |
|
433 |
Scan.optional ($$ ":" |-- Scan.repeat (parse_dot_name |
|
434 |
--| Scan.option ($$ ","))) [] |
|
435 |
||
436 |
(* It is not clear why some literals are followed by sequences of stars and/or |
|
437 |
pluses. We ignore them. *) |
|
39602 | 438 |
fun parse_decorated_atom x = |
439 |
(parse_atom --| Scan.repeat ($$ "*" || $$ "+" || $$ " ")) x |
|
39452 | 440 |
|
48132
9aa0fad4e864
added type arguments to "ATerm" constructor -- but don't use them yet
blanchet
parents:
48130
diff
changeset
|
441 |
fun mk_horn ([], []) = AAtom (ATerm (("c_False", []), [])) |
42943 | 442 |
| mk_horn ([], pos_lits) = foldr1 (uncurry (mk_aconn AOr)) pos_lits |
443 |
| mk_horn (neg_lits, []) = mk_anot (foldr1 (uncurry (mk_aconn AAnd)) neg_lits) |
|
39452 | 444 |
| mk_horn (neg_lits, pos_lits) = |
42943 | 445 |
mk_aconn AImplies (foldr1 (uncurry (mk_aconn AAnd)) neg_lits) |
446 |
(foldr1 (uncurry (mk_aconn AOr)) pos_lits) |
|
39452 | 447 |
|
39645 | 448 |
fun parse_horn_clause x = |
449 |
(Scan.repeat parse_decorated_atom --| $$ "|" --| $$ "|" |
|
450 |
-- Scan.repeat parse_decorated_atom --| $$ "-" --| $$ ">" |
|
451 |
-- Scan.repeat parse_decorated_atom |
|
452 |
>> (mk_horn o apfst (op @))) x |
|
39452 | 453 |
|
46390 | 454 |
val parse_spass_debug = |
455 |
Scan.option ($$ "(" |-- Scan.repeat (scan_general_id --| Scan.option ($$ ",")) |
|
456 |
--| $$ ")") |
|
457 |
||
46427 | 458 |
(* Syntax: <num>[0:<inference><annotations>] <atoms> || <atoms> -> <atoms>. |
459 |
derived from formulae <ident>* *) |
|
48005
eeede26f2721
killed SPASS 3.5/3.7 FLOTTER hack -- requires users to upgrade to SPASS 3.8
blanchet
parents:
47972
diff
changeset
|
460 |
fun parse_spass_line x = |
51211 | 461 |
(parse_spass_debug |-- scan_general_id --| $$ "[" --| |
462 |
Scan.many1 Symbol.is_digit --| $$ ":" -- Symbol.scan_ascii_id |
|
463 |
-- parse_spass_annotations --| $$ "]" |
|
48005
eeede26f2721
killed SPASS 3.5/3.7 FLOTTER hack -- requires users to upgrade to SPASS 3.8
blanchet
parents:
47972
diff
changeset
|
464 |
-- parse_horn_clause --| $$ "." |
eeede26f2721
killed SPASS 3.5/3.7 FLOTTER hack -- requires users to upgrade to SPASS 3.8
blanchet
parents:
47972
diff
changeset
|
465 |
-- Scan.option (Scan.this_string "derived from formulae " |
eeede26f2721
killed SPASS 3.5/3.7 FLOTTER hack -- requires users to upgrade to SPASS 3.8
blanchet
parents:
47972
diff
changeset
|
466 |
|-- Scan.repeat (scan_general_id --| Scan.option ($$ " "))) |
eeede26f2721
killed SPASS 3.5/3.7 FLOTTER hack -- requires users to upgrade to SPASS 3.8
blanchet
parents:
47972
diff
changeset
|
467 |
>> (fn ((((num, rule), deps), u), names) => |
51201 | 468 |
((num, these names), Unknown, u, rule, map (rpair []) deps))) x |
45162 | 469 |
|
52073
ccb292952774
started adding agsyHOL as an experimental prover
blanchet
parents:
52031
diff
changeset
|
470 |
val agsyhol_coreN = "__agsyhol_core" (* arbitrary *) |
48539 | 471 |
val satallax_coreN = "__satallax_core" (* arbitrary *) |
472 |
val z3_tptp_coreN = "__z3_tptp_core" (* arbitrary *) |
|
473 |
||
52077 | 474 |
fun core_inference inf fact = ((fact, [fact]), Unknown, dummy_phi, inf, []) |
475 |
||
48539 | 476 |
(* Syntax: core(<name>,[<name>,...,<name>]). *) |
477 |
fun parse_z3_tptp_line x = |
|
478 |
(scan_general_id --| $$ "," --| $$ "[" -- parse_dependencies --| $$ "]" |
|
479 |
>> (fn (name, names) => |
|
51201 | 480 |
(("", name :: names), Unknown, dummy_phi, z3_tptp_coreN, []))) x |
47947 | 481 |
|
45162 | 482 |
(* Syntax: <name> *) |
45203 | 483 |
fun parse_satallax_line x = |
52077 | 484 |
(scan_general_id --| Scan.option ($$ " ") >> core_inference satallax_coreN) x |
43481 | 485 |
|
48005
eeede26f2721
killed SPASS 3.5/3.7 FLOTTER hack -- requires users to upgrade to SPASS 3.8
blanchet
parents:
47972
diff
changeset
|
486 |
fun parse_line problem = |
48539 | 487 |
parse_tstp_line problem || parse_spass_line || parse_z3_tptp_line |
488 |
|| parse_satallax_line |
|
50590 | 489 |
fun parse_proof problem = |
490 |
strip_spaces_except_between_idents |
|
491 |
#> raw_explode |
|
492 |
#> Scan.error (!! (fn _ => raise UNRECOGNIZED_ATP_PROOF ()) |
|
493 |
(Scan.finite Symbol.stopper |
|
494 |
(Scan.repeat1 (parse_line problem)))) |
|
495 |
#> fst |
|
43481 | 496 |
|
52077 | 497 |
fun core_of_agsyhol_proof s = |
498 |
case split_lines s of |
|
499 |
"The transformed problem consists of the following conjectures:" :: conj :: |
|
500 |
_ :: proof_term :: _ => |
|
501 |
SOME (unprefix " " conj :: find_enclosed "<<" ">>" proof_term) |
|
502 |
| _ => NONE |
|
503 |
||
52031
9a9238342963
tuning -- renamed '_from_' to '_of_' in Sledgehammer
blanchet
parents:
51998
diff
changeset
|
504 |
fun atp_proof_of_tstplike_proof _ "" = [] |
9a9238342963
tuning -- renamed '_from_' to '_of_' in Sledgehammer
blanchet
parents:
51998
diff
changeset
|
505 |
| atp_proof_of_tstplike_proof problem tstp = |
52077 | 506 |
case core_of_agsyhol_proof tstp of |
507 |
SOME facts => facts |> map (core_inference agsyhol_coreN) |
|
508 |
| NONE => |
|
52756
1ac8a0d0ddb1
parse nonnumeric identifiers in E proofs correctly
blanchet
parents:
52755
diff
changeset
|
509 |
tstp ^ "$" (* the $ sign acts as a sentinel (FIXME: needed?) *) |
52077 | 510 |
|> parse_proof problem |
52755
4183c3219745
simplified Vampire hack -- no need to run it for other ATPs
blanchet
parents:
52077
diff
changeset
|
511 |
|> perhaps (try (sort (vampire_step_name_ord o pairself #1))) |
42968
74415622d293
more work on parsing LEO-II proofs and extracting uses of extensionality
blanchet
parents:
42966
diff
changeset
|
512 |
|
74415622d293
more work on parsing LEO-II proofs and extracting uses of extensionality
blanchet
parents:
42966
diff
changeset
|
513 |
fun clean_up_dependencies _ [] = [] |
51201 | 514 |
| clean_up_dependencies seen ((name, role, u, rule, deps) :: steps) = |
515 |
(name, role, u, rule, |
|
516 |
map_filter (fn dep => find_first (is_same_atp_step dep) seen) deps) :: |
|
42968
74415622d293
more work on parsing LEO-II proofs and extracting uses of extensionality
blanchet
parents:
42966
diff
changeset
|
517 |
clean_up_dependencies (name :: seen) steps |
74415622d293
more work on parsing LEO-II proofs and extracting uses of extensionality
blanchet
parents:
42966
diff
changeset
|
518 |
|
42975 | 519 |
fun clean_up_atp_proof_dependencies proof = clean_up_dependencies [] proof |
39452 | 520 |
|
50704 | 521 |
fun map_term_names_in_atp_proof f = |
522 |
let |
|
523 |
fun do_term (ATerm ((s, tys), ts)) = ATerm ((f s, tys), map do_term ts) |
|
524 |
fun do_formula (AQuant (q, xs, phi)) = |
|
525 |
AQuant (q, map (apfst f) xs, do_formula phi) |
|
526 |
| do_formula (AConn (c, phis)) = AConn (c, map do_formula phis) |
|
527 |
| do_formula (AAtom t) = AAtom (do_term t) |
|
51201 | 528 |
fun do_step (name, role, phi, rule, deps) = |
529 |
(name, role, do_formula phi, rule, deps) |
|
50704 | 530 |
in map do_step end |
39454 | 531 |
|
532 |
fun nasty_name pool s = s |> Symtab.lookup pool |> the_default s |
|
50704 | 533 |
|
39454 | 534 |
fun nasty_atp_proof pool = |
50704 | 535 |
not (Symtab.is_empty pool) ? map_term_names_in_atp_proof (nasty_name pool) |
39454 | 536 |
|
39452 | 537 |
end; |