src/HOL/Tools/ATP/atp_systems.ML
author blanchet
Sat, 18 Dec 2010 14:02:14 +0100
changeset 41269 abe867c29e55
parent 41238 78e4508d2e54
child 41313 a96ac4d180b7
permissions -rw-r--r--
tuning
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
38047
9033c03cc214 consequence of directory renaming
blanchet
parents: 38046
diff changeset
     1
(*  Title:      HOL/Tools/ATP/atp_systems.ML
28592
824f8390aaa2 renamed AtpThread to AtpWrapper;
wenzelm
parents:
diff changeset
     2
    Author:     Fabian Immler, TU Muenchen
36371
8c83ea1a7740 move the Sledgehammer menu options to "sledgehammer_isar.ML"
blanchet
parents: 36370
diff changeset
     3
    Author:     Jasmin Blanchette, TU Muenchen
28592
824f8390aaa2 renamed AtpThread to AtpWrapper;
wenzelm
parents:
diff changeset
     4
36376
e83d52a52449 renamed module "ATP_Wrapper" to "ATP_Systems"
blanchet
parents: 36371
diff changeset
     5
Setup for supported ATPs.
28592
824f8390aaa2 renamed AtpThread to AtpWrapper;
wenzelm
parents:
diff changeset
     6
*)
824f8390aaa2 renamed AtpThread to AtpWrapper;
wenzelm
parents:
diff changeset
     7
36376
e83d52a52449 renamed module "ATP_Wrapper" to "ATP_Systems"
blanchet
parents: 36371
diff changeset
     8
signature ATP_SYSTEMS =
28592
824f8390aaa2 renamed AtpThread to AtpWrapper;
wenzelm
parents:
diff changeset
     9
sig
39491
2416666e6f94 refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents: 39375
diff changeset
    10
  type failure = ATP_Proof.failure
38023
962b0a7f544b more refactoring
blanchet
parents: 38022
diff changeset
    11
40059
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
    12
  type atp_config =
38092
81a003f7de0d speed up the minimizer by using the time taken for the first iteration as a timeout for the following iterations, and fix a subtle bug in "string_for_failure"
blanchet
parents: 38090
diff changeset
    13
    {exec: string * string,
81a003f7de0d speed up the minimizer by using the time taken for the first iteration as a timeout for the following iterations, and fix a subtle bug in "string_for_failure"
blanchet
parents: 38090
diff changeset
    14
     required_execs: (string * string) list,
38023
962b0a7f544b more refactoring
blanchet
parents: 38022
diff changeset
    15
     arguments: bool -> Time.time -> string,
38645
4d5bbec1a598 be more generous towards SPASS's -SOS mode
blanchet
parents: 38631
diff changeset
    16
     has_incomplete_mode: bool,
38023
962b0a7f544b more refactoring
blanchet
parents: 38022
diff changeset
    17
     proof_delims: (string * string) list,
962b0a7f544b more refactoring
blanchet
parents: 38022
diff changeset
    18
     known_failures: (failure * string) list,
38744
2b6333f78a9e make relevance filter work in term of a "max_relevant" option + use Vampire SOS;
blanchet
parents: 38740
diff changeset
    19
     default_max_relevant: int,
38631
979a0b37f981 prefer TPTP "conjecture" tag to "hypothesis" on ATPs where this is possible;
blanchet
parents: 38603
diff changeset
    20
     explicit_forall: bool,
979a0b37f981 prefer TPTP "conjecture" tag to "hypothesis" on ATPs where this is possible;
blanchet
parents: 38603
diff changeset
    21
     use_conjecture_for_hypotheses: bool}
38023
962b0a7f544b more refactoring
blanchet
parents: 38022
diff changeset
    22
40059
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
    23
  val eN : string
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
    24
  val spassN : string
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
    25
  val vampireN : string
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
    26
  val sine_eN : string
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
    27
  val snarkN : string
40060
5ef6747aa619 first step in adding support for an SMT backend to Sledgehammer
blanchet
parents: 40059
diff changeset
    28
  val remote_prefix : string
40059
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
    29
  val add_atp : string * atp_config -> theory -> theory
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
    30
  val get_atp : theory -> string -> atp_config
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
    31
  val available_atps : theory -> string list
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
    32
  val is_atp_installed : theory -> string -> bool
35867
16279c4c7a33 move all ATP setup code into ATP_Wrapper
blanchet
parents: 35865
diff changeset
    33
  val refresh_systems_on_tptp : unit -> unit
16279c4c7a33 move all ATP setup code into ATP_Wrapper
blanchet
parents: 35865
diff changeset
    34
  val setup : theory -> theory
28592
824f8390aaa2 renamed AtpThread to AtpWrapper;
wenzelm
parents:
diff changeset
    35
end;
824f8390aaa2 renamed AtpThread to AtpWrapper;
wenzelm
parents:
diff changeset
    36
36376
e83d52a52449 renamed module "ATP_Wrapper" to "ATP_Systems"
blanchet
parents: 36371
diff changeset
    37
structure ATP_Systems : ATP_SYSTEMS =
28592
824f8390aaa2 renamed AtpThread to AtpWrapper;
wenzelm
parents:
diff changeset
    38
struct
28596
fcd463a6b6de tuned interfaces -- plain prover function, without thread;
wenzelm
parents: 28592
diff changeset
    39
39491
2416666e6f94 refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents: 39375
diff changeset
    40
open ATP_Proof
32864
a226f29d4bdc re-organized signature of AtpWrapper structure: records instead of unnamed parameters and return values,
boehmes
parents: 32740
diff changeset
    41
40059
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
    42
(* ATP configuration *)
32864
a226f29d4bdc re-organized signature of AtpWrapper structure: records instead of unnamed parameters and return values,
boehmes
parents: 32740
diff changeset
    43
40059
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
    44
type atp_config =
38092
81a003f7de0d speed up the minimizer by using the time taken for the first iteration as a timeout for the following iterations, and fix a subtle bug in "string_for_failure"
blanchet
parents: 38090
diff changeset
    45
  {exec: string * string,
81a003f7de0d speed up the minimizer by using the time taken for the first iteration as a timeout for the following iterations, and fix a subtle bug in "string_for_failure"
blanchet
parents: 38090
diff changeset
    46
   required_execs: (string * string) list,
37514
b147d01b8ebc if SPASS fails at finding a proof with the SOS option turned on, turn it off and try again
blanchet
parents: 37509
diff changeset
    47
   arguments: bool -> Time.time -> string,
38645
4d5bbec1a598 be more generous towards SPASS's -SOS mode
blanchet
parents: 38631
diff changeset
    48
   has_incomplete_mode: bool,
36370
a4f601daa175 centralized ATP-specific error handling in "atp_wrapper.ML"
blanchet
parents: 36369
diff changeset
    49
   proof_delims: (string * string) list,
a4f601daa175 centralized ATP-specific error handling in "atp_wrapper.ML"
blanchet
parents: 36369
diff changeset
    50
   known_failures: (failure * string) list,
38744
2b6333f78a9e make relevance filter work in term of a "max_relevant" option + use Vampire SOS;
blanchet
parents: 38740
diff changeset
    51
   default_max_relevant: int,
38631
979a0b37f981 prefer TPTP "conjecture" tag to "hypothesis" on ATPs where this is possible;
blanchet
parents: 38603
diff changeset
    52
   explicit_forall: bool,
979a0b37f981 prefer TPTP "conjecture" tag to "hypothesis" on ATPs where this is possible;
blanchet
parents: 38603
diff changeset
    53
   use_conjecture_for_hypotheses: bool}
28596
fcd463a6b6de tuned interfaces -- plain prover function, without thread;
wenzelm
parents: 28592
diff changeset
    54
38061
685d1f0f75b3 handle Perl and "libwww-perl" failures more gracefully, giving the user some clues about what goes on
blanchet
parents: 38049
diff changeset
    55
val known_perl_failures =
38094
d01b8119b2e0 better error and minimizer output
blanchet
parents: 38092
diff changeset
    56
  [(CantConnect, "HTTP error"),
d01b8119b2e0 better error and minimizer output
blanchet
parents: 38092
diff changeset
    57
   (NoPerl, "env: perl"),
38065
9069e1ad1527 improved ATP error handling some more
blanchet
parents: 38064
diff changeset
    58
   (NoLibwwwPerl, "Can't locate HTTP")]
28596
fcd463a6b6de tuned interfaces -- plain prover function, without thread;
wenzelm
parents: 28592
diff changeset
    59
40059
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
    60
(* named ATPs *)
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
    61
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
    62
val eN = "e"
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
    63
val spassN = "spass"
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
    64
val vampireN = "vampire"
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
    65
val sine_eN = "sine_e"
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
    66
val snarkN = "snark"
40060
5ef6747aa619 first step in adding support for an SMT backend to Sledgehammer
blanchet
parents: 40059
diff changeset
    67
val remote_prefix = "remote_"
38001
a9b47b85ca24 reintroduced more preprocessing steps to Sledgehammer, adapted to the new FOF setting
blanchet
parents: 38000
diff changeset
    68
38023
962b0a7f544b more refactoring
blanchet
parents: 38022
diff changeset
    69
structure Data = Theory_Data
962b0a7f544b more refactoring
blanchet
parents: 38022
diff changeset
    70
(
40059
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
    71
  type T = (atp_config * stamp) Symtab.table
38023
962b0a7f544b more refactoring
blanchet
parents: 38022
diff changeset
    72
  val empty = Symtab.empty
962b0a7f544b more refactoring
blanchet
parents: 38022
diff changeset
    73
  val extend = I
962b0a7f544b more refactoring
blanchet
parents: 38022
diff changeset
    74
  fun merge data : T = Symtab.merge (eq_snd op =) data
962b0a7f544b more refactoring
blanchet
parents: 38022
diff changeset
    75
    handle Symtab.DUP name => error ("Duplicate ATP: " ^ quote name ^ ".")
962b0a7f544b more refactoring
blanchet
parents: 38022
diff changeset
    76
)
38017
3ad3e3ca2451 move Sledgehammer-specific code out of "Sledgehammer_TPTP_Format"
blanchet
parents: 38015
diff changeset
    77
38737
bdcb23701448 better workaround for E's off-by-one-second issue
blanchet
parents: 38691
diff changeset
    78
fun to_secs bonus time = (Time.toMilliseconds time + bonus + 999) div 1000
36142
f5e15e9aae10 make Sledgehammer "minimize" output less confusing + round up (not down) time limits to nearest second
blanchet
parents: 36064
diff changeset
    79
39491
2416666e6f94 refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents: 39375
diff changeset
    80
40059
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
    81
(* E *)
28596
fcd463a6b6de tuned interfaces -- plain prover function, without thread;
wenzelm
parents: 28592
diff changeset
    82
40344
df25b51af013 give E one more second, to prevent cases where it finds a proof but has no time to print it
blanchet
parents: 40060
diff changeset
    83
(* Give E an extra second to reconstruct the proof. Older versions even get two
df25b51af013 give E one more second, to prevent cases where it finds a proof but has no time to print it
blanchet
parents: 40060
diff changeset
    84
   seconds, because the "eproof" script wrongly subtracted an entire second to
df25b51af013 give E one more second, to prevent cases where it finds a proof but has no time to print it
blanchet
parents: 40060
diff changeset
    85
   account for the overhead of the script itself, which is in fact much
df25b51af013 give E one more second, to prevent cases where it finds a proof but has no time to print it
blanchet
parents: 40060
diff changeset
    86
   lower. *)
38737
bdcb23701448 better workaround for E's off-by-one-second issue
blanchet
parents: 38691
diff changeset
    87
fun e_bonus () =
bdcb23701448 better workaround for E's off-by-one-second issue
blanchet
parents: 38691
diff changeset
    88
  case getenv "E_VERSION" of
40344
df25b51af013 give E one more second, to prevent cases where it finds a proof but has no time to print it
blanchet
parents: 40060
diff changeset
    89
    "" => 2000
38737
bdcb23701448 better workaround for E's off-by-one-second issue
blanchet
parents: 38691
diff changeset
    90
  | version =>
40344
df25b51af013 give E one more second, to prevent cases where it finds a proof but has no time to print it
blanchet
parents: 40060
diff changeset
    91
    if exists (fn s => String.isPrefix s version) ["0.9", "1.0"] then 2000
df25b51af013 give E one more second, to prevent cases where it finds a proof but has no time to print it
blanchet
parents: 40060
diff changeset
    92
    else 1000
38737
bdcb23701448 better workaround for E's off-by-one-second issue
blanchet
parents: 38691
diff changeset
    93
36369
d2cd0d04b8e6 handle ATP proof delimiters in a cleaner, more extensible fashion
blanchet
parents: 36289
diff changeset
    94
val tstp_proof_delims =
d2cd0d04b8e6 handle ATP proof delimiters in a cleaner, more extensible fashion
blanchet
parents: 36289
diff changeset
    95
  ("# SZS output start CNFRefutation.", "# SZS output end CNFRefutation")
d2cd0d04b8e6 handle ATP proof delimiters in a cleaner, more extensible fashion
blanchet
parents: 36289
diff changeset
    96
40059
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
    97
val e_config : atp_config =
38092
81a003f7de0d speed up the minimizer by using the time taken for the first iteration as a timeout for the following iterations, and fix a subtle bug in "string_for_failure"
blanchet
parents: 38090
diff changeset
    98
  {exec = ("E_HOME", "eproof"),
81a003f7de0d speed up the minimizer by using the time taken for the first iteration as a timeout for the following iterations, and fix a subtle bug in "string_for_failure"
blanchet
parents: 38090
diff changeset
    99
   required_execs = [],
37514
b147d01b8ebc if SPASS fails at finding a proof with the SOS option turned on, turn it off and try again
blanchet
parents: 37509
diff changeset
   100
   arguments = fn _ => fn timeout =>
38691
fe5929dacd43 use a soft time limit for E
blanchet
parents: 38690
diff changeset
   101
     "--tstp-in --tstp-out -l5 -xAutoDev -tAutoDev --silent \
38737
bdcb23701448 better workaround for E's off-by-one-second issue
blanchet
parents: 38691
diff changeset
   102
     \--cpu-limit=" ^ string_of_int (to_secs (e_bonus ()) timeout),
38645
4d5bbec1a598 be more generous towards SPASS's -SOS mode
blanchet
parents: 38631
diff changeset
   103
   has_incomplete_mode = false,
36369
d2cd0d04b8e6 handle ATP proof delimiters in a cleaner, more extensible fashion
blanchet
parents: 36289
diff changeset
   104
   proof_delims = [tstp_proof_delims],
36265
41c9e755e552 distinguish between the different ATP errors in the user interface;
blanchet
parents: 36264
diff changeset
   105
   known_failures =
37995
06f02b15ef8a generate full first-order formulas (FOF) in Sledgehammer
blanchet
parents: 37994
diff changeset
   106
     [(Unprovable, "SZS status: CounterSatisfiable"),
06f02b15ef8a generate full first-order formulas (FOF) in Sledgehammer
blanchet
parents: 37994
diff changeset
   107
      (Unprovable, "SZS status CounterSatisfiable"),
36370
a4f601daa175 centralized ATP-specific error handling in "atp_wrapper.ML"
blanchet
parents: 36369
diff changeset
   108
      (TimedOut, "Failure: Resource limit exceeded (time)"),
a4f601daa175 centralized ATP-specific error handling in "atp_wrapper.ML"
blanchet
parents: 36369
diff changeset
   109
      (TimedOut, "time limit exceeded"),
a4f601daa175 centralized ATP-specific error handling in "atp_wrapper.ML"
blanchet
parents: 36369
diff changeset
   110
      (OutOfResources,
a4f601daa175 centralized ATP-specific error handling in "atp_wrapper.ML"
blanchet
parents: 36369
diff changeset
   111
       "# Cannot determine problem status within resource limit"),
a4f601daa175 centralized ATP-specific error handling in "atp_wrapper.ML"
blanchet
parents: 36369
diff changeset
   112
      (OutOfResources, "SZS status: ResourceOut"),
a4f601daa175 centralized ATP-specific error handling in "atp_wrapper.ML"
blanchet
parents: 36369
diff changeset
   113
      (OutOfResources, "SZS status ResourceOut")],
38744
2b6333f78a9e make relevance filter work in term of a "max_relevant" option + use Vampire SOS;
blanchet
parents: 38740
diff changeset
   114
   default_max_relevant = 500 (* FUDGE *),
38631
979a0b37f981 prefer TPTP "conjecture" tag to "hypothesis" on ATPs where this is possible;
blanchet
parents: 38603
diff changeset
   115
   explicit_forall = false,
979a0b37f981 prefer TPTP "conjecture" tag to "hypothesis" on ATPs where this is possible;
blanchet
parents: 38603
diff changeset
   116
   use_conjecture_for_hypotheses = true}
38454
9043eefe8d71 detect old Vampire and give a nicer error message
blanchet
parents: 38433
diff changeset
   117
40059
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   118
val e = (eN, e_config)
28596
fcd463a6b6de tuned interfaces -- plain prover function, without thread;
wenzelm
parents: 28592
diff changeset
   119
fcd463a6b6de tuned interfaces -- plain prover function, without thread;
wenzelm
parents: 28592
diff changeset
   120
39491
2416666e6f94 refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents: 39375
diff changeset
   121
(* SPASS *)
2416666e6f94 refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents: 39375
diff changeset
   122
36219
16670b4f0baa set SPASS option on the command-line, so that it doesn't vanish when moving to TPTP format
blanchet
parents: 36190
diff changeset
   123
(* The "-VarWeight=3" option helps the higher-order problems, probably by
16670b4f0baa set SPASS option on the command-line, so that it doesn't vanish when moving to TPTP format
blanchet
parents: 36190
diff changeset
   124
   counteracting the presence of "hAPP". *)
40059
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   125
val spass_config : atp_config =
38092
81a003f7de0d speed up the minimizer by using the time taken for the first iteration as a timeout for the following iterations, and fix a subtle bug in "string_for_failure"
blanchet
parents: 38090
diff changeset
   126
  {exec = ("ISABELLE_ATP", "scripts/spass"),
39002
a2d7be688ea1 add dependency of "spass" script
blanchet
parents: 38999
diff changeset
   127
   required_execs = [("SPASS_HOME", "SPASS"), ("SPASS_HOME", "tptp2dfg")],
37514
b147d01b8ebc if SPASS fails at finding a proof with the SOS option turned on, turn it off and try again
blanchet
parents: 37509
diff changeset
   128
   arguments = fn complete => fn timeout =>
37962
d7dbe01f48d7 keep track of clause numbers for SPASS now that we generate FOF rather than CNF problems;
blanchet
parents: 37926
diff changeset
   129
     ("-Auto -PGiven=0 -PProblem=0 -Splits=0 -FullRed=0 -DocProof \
38737
bdcb23701448 better workaround for E's off-by-one-second issue
blanchet
parents: 38691
diff changeset
   130
      \-VarWeight=3 -TimeLimit=" ^ string_of_int (to_secs 0 timeout))
37514
b147d01b8ebc if SPASS fails at finding a proof with the SOS option turned on, turn it off and try again
blanchet
parents: 37509
diff changeset
   131
     |> not complete ? prefix "-SOS=1 ",
38645
4d5bbec1a598 be more generous towards SPASS's -SOS mode
blanchet
parents: 38631
diff changeset
   132
   has_incomplete_mode = true,
36369
d2cd0d04b8e6 handle ATP proof delimiters in a cleaner, more extensible fashion
blanchet
parents: 36289
diff changeset
   133
   proof_delims = [("Here is a proof", "Formulae used in the proof")],
36289
f75b6a3e1450 set "atps" reference's default value to "(remote_)e (remote_)spass (remote_)vampire", based on what is installed
blanchet
parents: 36287
diff changeset
   134
   known_failures =
38061
685d1f0f75b3 handle Perl and "libwww-perl" failures more gracefully, giving the user some clues about what goes on
blanchet
parents: 38049
diff changeset
   135
     known_perl_failures @
37413
e856582fe9c4 improve ATP-specific error messages
blanchet
parents: 37347
diff changeset
   136
     [(IncompleteUnprovable, "SPASS beiseite: Completion found"),
36370
a4f601daa175 centralized ATP-specific error handling in "atp_wrapper.ML"
blanchet
parents: 36369
diff changeset
   137
      (TimedOut, "SPASS beiseite: Ran out of time"),
36965
67ae217c6b5c identify common SPASS error more clearly
blanchet
parents: 36924
diff changeset
   138
      (OutOfResources, "SPASS beiseite: Maximal number of loops exceeded"),
37413
e856582fe9c4 improve ATP-specific error messages
blanchet
parents: 37347
diff changeset
   139
      (MalformedInput, "Undefined symbol"),
37414
d0cea0796295 expect SPASS 3.7, and give a friendly warning if an older version is used
blanchet
parents: 37413
diff changeset
   140
      (MalformedInput, "Free Variable"),
39263
e2a3c435334b more precise error messages when Vampire is interrupted or SPASS runs into an internal bug
blanchet
parents: 39262
diff changeset
   141
      (SpassTooOld, "tptp2dfg"),
e2a3c435334b more precise error messages when Vampire is interrupted or SPASS runs into an internal bug
blanchet
parents: 39262
diff changeset
   142
      (InternalError, "Please report this error")],
38744
2b6333f78a9e make relevance filter work in term of a "max_relevant" option + use Vampire SOS;
blanchet
parents: 38740
diff changeset
   143
   default_max_relevant = 350 (* FUDGE *),
38631
979a0b37f981 prefer TPTP "conjecture" tag to "hypothesis" on ATPs where this is possible;
blanchet
parents: 38603
diff changeset
   144
   explicit_forall = true,
979a0b37f981 prefer TPTP "conjecture" tag to "hypothesis" on ATPs where this is possible;
blanchet
parents: 38603
diff changeset
   145
   use_conjecture_for_hypotheses = true}
38454
9043eefe8d71 detect old Vampire and give a nicer error message
blanchet
parents: 38433
diff changeset
   146
40059
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   147
val spass = (spassN, spass_config)
28596
fcd463a6b6de tuned interfaces -- plain prover function, without thread;
wenzelm
parents: 28592
diff changeset
   148
38454
9043eefe8d71 detect old Vampire and give a nicer error message
blanchet
parents: 38433
diff changeset
   149
37509
f39464d971c4 factor out TPTP format output into file of its own, to facilitate further changes
blanchet
parents: 37506
diff changeset
   150
(* Vampire *)
f39464d971c4 factor out TPTP format output into file of its own, to facilitate further changes
blanchet
parents: 37506
diff changeset
   151
40059
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   152
val vampire_config : atp_config =
38092
81a003f7de0d speed up the minimizer by using the time taken for the first iteration as a timeout for the following iterations, and fix a subtle bug in "string_for_failure"
blanchet
parents: 38090
diff changeset
   153
  {exec = ("VAMPIRE_HOME", "vampire"),
81a003f7de0d speed up the minimizer by using the time taken for the first iteration as a timeout for the following iterations, and fix a subtle bug in "string_for_failure"
blanchet
parents: 38090
diff changeset
   154
   required_execs = [],
38744
2b6333f78a9e make relevance filter work in term of a "max_relevant" option + use Vampire SOS;
blanchet
parents: 38740
diff changeset
   155
   arguments = fn complete => fn timeout =>
41203
1393514094d7 fixed more issues with the Vampire output parser, and added support for Vampire's TSTP output (--proof tptp)
blanchet
parents: 41148
diff changeset
   156
     (* This would work too but it's less tested: "--proof tptp " ^ *)
1393514094d7 fixed more issues with the Vampire output parser, and added support for Vampire's TSTP output (--proof tptp)
blanchet
parents: 41148
diff changeset
   157
     "--mode casc -t " ^ string_of_int (to_secs 0 timeout) ^
1393514094d7 fixed more issues with the Vampire output parser, and added support for Vampire's TSTP output (--proof tptp)
blanchet
parents: 41148
diff changeset
   158
     " --thanks \"Andrei and Krystof\" --input_file"
38744
2b6333f78a9e make relevance filter work in term of a "max_relevant" option + use Vampire SOS;
blanchet
parents: 38740
diff changeset
   159
     |> not complete ? prefix "--sos on ",
2b6333f78a9e make relevance filter work in term of a "max_relevant" option + use Vampire SOS;
blanchet
parents: 38740
diff changeset
   160
   has_incomplete_mode = true,
37509
f39464d971c4 factor out TPTP format output into file of its own, to facilitate further changes
blanchet
parents: 37506
diff changeset
   161
   proof_delims =
f39464d971c4 factor out TPTP format output into file of its own, to facilitate further changes
blanchet
parents: 37506
diff changeset
   162
     [("=========== Refutation ==========",
f39464d971c4 factor out TPTP format output into file of its own, to facilitate further changes
blanchet
parents: 37506
diff changeset
   163
       "======= End of refutation ======="),
38033
df99f022751d support latest version of Vampire (1.0) locally
blanchet
parents: 38032
diff changeset
   164
      ("% SZS output start Refutation", "% SZS output end Refutation"),
df99f022751d support latest version of Vampire (1.0) locally
blanchet
parents: 38032
diff changeset
   165
      ("% SZS output start Proof", "% SZS output end Proof")],
37509
f39464d971c4 factor out TPTP format output into file of its own, to facilitate further changes
blanchet
parents: 37506
diff changeset
   166
   known_failures =
f39464d971c4 factor out TPTP format output into file of its own, to facilitate further changes
blanchet
parents: 37506
diff changeset
   167
     [(Unprovable, "UNPROVABLE"),
f39464d971c4 factor out TPTP format output into file of its own, to facilitate further changes
blanchet
parents: 37506
diff changeset
   168
      (IncompleteUnprovable, "CANNOT PROVE"),
38092
81a003f7de0d speed up the minimizer by using the time taken for the first iteration as a timeout for the following iterations, and fix a subtle bug in "string_for_failure"
blanchet
parents: 38090
diff changeset
   169
      (TimedOut, "SZS status Timeout"),
37509
f39464d971c4 factor out TPTP format output into file of its own, to facilitate further changes
blanchet
parents: 37506
diff changeset
   170
      (Unprovable, "Satisfiability detected"),
38647
5500241da479 play with fudge factor + parse one more Vampire error
blanchet
parents: 38646
diff changeset
   171
      (Unprovable, "Termination reason: Satisfiable"),
39263
e2a3c435334b more precise error messages when Vampire is interrupted or SPASS runs into an internal bug
blanchet
parents: 39262
diff changeset
   172
      (VampireTooOld, "not a valid option"),
e2a3c435334b more precise error messages when Vampire is interrupted or SPASS runs into an internal bug
blanchet
parents: 39262
diff changeset
   173
      (Interrupted, "Aborted by signal SIGINT")],
38744
2b6333f78a9e make relevance filter work in term of a "max_relevant" option + use Vampire SOS;
blanchet
parents: 38740
diff changeset
   174
   default_max_relevant = 400 (* FUDGE *),
38631
979a0b37f981 prefer TPTP "conjecture" tag to "hypothesis" on ATPs where this is possible;
blanchet
parents: 38603
diff changeset
   175
   explicit_forall = false,
38680
634a6d400c2e revert unintended change
blanchet
parents: 38678
diff changeset
   176
   use_conjecture_for_hypotheses = true}
38454
9043eefe8d71 detect old Vampire and give a nicer error message
blanchet
parents: 38433
diff changeset
   177
40059
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   178
val vampire = (vampireN, vampire_config)
37509
f39464d971c4 factor out TPTP format output into file of its own, to facilitate further changes
blanchet
parents: 37506
diff changeset
   179
38454
9043eefe8d71 detect old Vampire and give a nicer error message
blanchet
parents: 38433
diff changeset
   180
40059
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   181
(* Remote ATP invocation via SystemOnTPTP *)
28596
fcd463a6b6de tuned interfaces -- plain prover function, without thread;
wenzelm
parents: 28592
diff changeset
   182
38061
685d1f0f75b3 handle Perl and "libwww-perl" failures more gracefully, giving the user some clues about what goes on
blanchet
parents: 38049
diff changeset
   183
val systems = Synchronized.var "atp_systems" ([] : string list)
31835
b686d4df54c2 check for current versions on server
immler@in.tum.de
parents: 31832
diff changeset
   184
b686d4df54c2 check for current versions on server
immler@in.tum.de
parents: 31832
diff changeset
   185
fun get_systems () =
38061
685d1f0f75b3 handle Perl and "libwww-perl" failures more gracefully, giving the user some clues about what goes on
blanchet
parents: 38049
diff changeset
   186
  case bash_output "\"$ISABELLE_ATP/scripts/remote_atp\" -w 2>&1" of
39491
2416666e6f94 refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents: 39375
diff changeset
   187
    (output, 0) => split_lines output
2416666e6f94 refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents: 39375
diff changeset
   188
  | (output, _) =>
2416666e6f94 refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents: 39375
diff changeset
   189
    error (case extract_known_failure known_perl_failures output of
40669
5c316d1327d4 more precise error handling for Z3;
blanchet
parents: 40596
diff changeset
   190
             SOME failure => string_for_failure "ATP" failure
39491
2416666e6f94 refactoring: move ATP proof and error extraction code to "ATP_Proof" module
blanchet
parents: 39375
diff changeset
   191
           | NONE => perhaps (try (unsuffix "\n")) output ^ ".")
31835
b686d4df54c2 check for current versions on server
immler@in.tum.de
parents: 31832
diff changeset
   192
38690
38a926e033ad make remote ATP versions more robust, by starting with "preferred" version numbers and falling back on any version
blanchet
parents: 38685
diff changeset
   193
fun find_system name [] systems = find_first (String.isPrefix name) systems
38a926e033ad make remote ATP versions more robust, by starting with "preferred" version numbers and falling back on any version
blanchet
parents: 38685
diff changeset
   194
  | find_system name (version :: versions) systems =
38a926e033ad make remote ATP versions more robust, by starting with "preferred" version numbers and falling back on any version
blanchet
parents: 38685
diff changeset
   195
    case find_first (String.isPrefix (name ^ "---" ^ version)) systems of
38a926e033ad make remote ATP versions more robust, by starting with "preferred" version numbers and falling back on any version
blanchet
parents: 38685
diff changeset
   196
      NONE => find_system name versions systems
38a926e033ad make remote ATP versions more robust, by starting with "preferred" version numbers and falling back on any version
blanchet
parents: 38685
diff changeset
   197
    | res => res
38a926e033ad make remote ATP versions more robust, by starting with "preferred" version numbers and falling back on any version
blanchet
parents: 38685
diff changeset
   198
38a926e033ad make remote ATP versions more robust, by starting with "preferred" version numbers and falling back on any version
blanchet
parents: 38685
diff changeset
   199
fun get_system name versions =
38589
b03f8fe043ec added "max_relevant_per_iter" option to Sledgehammer
blanchet
parents: 38588
diff changeset
   200
  Synchronized.change_result systems
b03f8fe043ec added "max_relevant_per_iter" option to Sledgehammer
blanchet
parents: 38588
diff changeset
   201
      (fn systems => (if null systems then get_systems () else systems)
38690
38a926e033ad make remote ATP versions more robust, by starting with "preferred" version numbers and falling back on any version
blanchet
parents: 38685
diff changeset
   202
                     |> `(find_system name versions))
32864
a226f29d4bdc re-organized signature of AtpWrapper structure: records instead of unnamed parameters and return values,
boehmes
parents: 32740
diff changeset
   203
38690
38a926e033ad make remote ATP versions more robust, by starting with "preferred" version numbers and falling back on any version
blanchet
parents: 38685
diff changeset
   204
fun the_system name versions =
38a926e033ad make remote ATP versions more robust, by starting with "preferred" version numbers and falling back on any version
blanchet
parents: 38685
diff changeset
   205
  case get_system name versions of
39010
344028ecc00e show real CPU time
blanchet
parents: 39002
diff changeset
   206
    SOME sys => sys
41269
blanchet
parents: 41238
diff changeset
   207
  | NONE => error ("System " ^ quote name ^
blanchet
parents: 41238
diff changeset
   208
                   " is not available at SystemOnTPTP.")
31835
b686d4df54c2 check for current versions on server
immler@in.tum.de
parents: 31832
diff changeset
   209
41148
f5229ab54284 added timeout max for remote server invocation
blanchet
parents: 40669
diff changeset
   210
val max_remote_secs = 240 (* give Geoff Sutcliffe's servers a break *)
f5229ab54284 added timeout max for remote server invocation
blanchet
parents: 40669
diff changeset
   211
38690
38a926e033ad make remote ATP versions more robust, by starting with "preferred" version numbers and falling back on any version
blanchet
parents: 38685
diff changeset
   212
fun remote_config system_name system_versions proof_delims known_failures
38997
78ac4468cf9d got rid of the "theory_relevant" option;
blanchet
parents: 38817
diff changeset
   213
                  default_max_relevant use_conjecture_for_hypotheses
40059
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   214
                  : atp_config =
38092
81a003f7de0d speed up the minimizer by using the time taken for the first iteration as a timeout for the following iterations, and fix a subtle bug in "string_for_failure"
blanchet
parents: 38090
diff changeset
   215
  {exec = ("ISABELLE_ATP", "scripts/remote_atp"),
81a003f7de0d speed up the minimizer by using the time taken for the first iteration as a timeout for the following iterations, and fix a subtle bug in "string_for_failure"
blanchet
parents: 38090
diff changeset
   216
   required_execs = [],
37514
b147d01b8ebc if SPASS fails at finding a proof with the SOS option turned on, turn it off and try again
blanchet
parents: 37509
diff changeset
   217
   arguments = fn _ => fn timeout =>
41148
f5229ab54284 added timeout max for remote server invocation
blanchet
parents: 40669
diff changeset
   218
     " -t " ^ string_of_int (Int.min (max_remote_secs, (to_secs 0 timeout)))
f5229ab54284 added timeout max for remote server invocation
blanchet
parents: 40669
diff changeset
   219
     ^ " -s " ^ the_system system_name system_versions,
38645
4d5bbec1a598 be more generous towards SPASS's -SOS mode
blanchet
parents: 38631
diff changeset
   220
   has_incomplete_mode = false,
36369
d2cd0d04b8e6 handle ATP proof delimiters in a cleaner, more extensible fashion
blanchet
parents: 36289
diff changeset
   221
   proof_delims = insert (op =) tstp_proof_delims proof_delims,
38061
685d1f0f75b3 handle Perl and "libwww-perl" failures more gracefully, giving the user some clues about what goes on
blanchet
parents: 38049
diff changeset
   222
   known_failures =
685d1f0f75b3 handle Perl and "libwww-perl" failures more gracefully, giving the user some clues about what goes on
blanchet
parents: 38049
diff changeset
   223
     known_failures @ known_perl_failures @
38094
d01b8119b2e0 better error and minimizer output
blanchet
parents: 38092
diff changeset
   224
     [(TimedOut, "says Timeout")],
38744
2b6333f78a9e make relevance filter work in term of a "max_relevant" option + use Vampire SOS;
blanchet
parents: 38740
diff changeset
   225
   default_max_relevant = default_max_relevant,
38631
979a0b37f981 prefer TPTP "conjecture" tag to "hypothesis" on ATPs where this is possible;
blanchet
parents: 38603
diff changeset
   226
   explicit_forall = true,
979a0b37f981 prefer TPTP "conjecture" tag to "hypothesis" on ATPs where this is possible;
blanchet
parents: 38603
diff changeset
   227
   use_conjecture_for_hypotheses = use_conjecture_for_hypotheses}
28596
fcd463a6b6de tuned interfaces -- plain prover function, without thread;
wenzelm
parents: 28592
diff changeset
   228
38690
38a926e033ad make remote ATP versions more robust, by starting with "preferred" version numbers and falling back on any version
blanchet
parents: 38685
diff changeset
   229
fun remotify_config system_name system_versions
38744
2b6333f78a9e make relevance filter work in term of a "max_relevant" option + use Vampire SOS;
blanchet
parents: 38740
diff changeset
   230
        ({proof_delims, known_failures, default_max_relevant,
40059
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   231
          use_conjecture_for_hypotheses, ...} : atp_config) : atp_config =
38690
38a926e033ad make remote ATP versions more robust, by starting with "preferred" version numbers and falling back on any version
blanchet
parents: 38685
diff changeset
   232
  remote_config system_name system_versions proof_delims known_failures
38997
78ac4468cf9d got rid of the "theory_relevant" option;
blanchet
parents: 38817
diff changeset
   233
                default_max_relevant use_conjecture_for_hypotheses
38023
962b0a7f544b more refactoring
blanchet
parents: 38022
diff changeset
   234
40059
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   235
fun remote_atp name system_name system_versions proof_delims known_failures
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   236
               default_max_relevant use_conjecture_for_hypotheses =
40060
5ef6747aa619 first step in adding support for an SMT backend to Sledgehammer
blanchet
parents: 40059
diff changeset
   237
  (remote_prefix ^ name,
38690
38a926e033ad make remote ATP versions more robust, by starting with "preferred" version numbers and falling back on any version
blanchet
parents: 38685
diff changeset
   238
   remote_config system_name system_versions proof_delims known_failures
38997
78ac4468cf9d got rid of the "theory_relevant" option;
blanchet
parents: 38817
diff changeset
   239
                 default_max_relevant use_conjecture_for_hypotheses)
40059
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   240
fun remotify_atp (name, config) system_name system_versions =
40060
5ef6747aa619 first step in adding support for an SMT backend to Sledgehammer
blanchet
parents: 40059
diff changeset
   241
  (remote_prefix ^ name, remotify_config system_name system_versions config)
28592
824f8390aaa2 renamed AtpThread to AtpWrapper;
wenzelm
parents:
diff changeset
   242
40059
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   243
val remote_e = remotify_atp e "EP" ["1.0", "1.1", "1.2"]
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   244
val remote_vampire = remotify_atp vampire "Vampire" ["0.6", "9.0", "1.0"]
38603
a57d04dd1b25 fix SInE's error handling + run "vampire" locally if either SPASS or E is missing
blanchet
parents: 38598
diff changeset
   245
val remote_sine_e =
40059
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   246
  remote_atp sine_eN "SInE" [] [] [(IncompleteUnprovable, "says Unknown")]
41238
78e4508d2e54 fewer facts to SInE-E
blanchet
parents: 41203
diff changeset
   247
                600 (* FUDGE *) true
38598
ce117ef51999 added remote SInE and remote SNARK
blanchet
parents: 38596
diff changeset
   248
val remote_snark =
40059
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   249
  remote_atp snarkN "SNARK---" [] [("refutation.", "end_refutation.")] []
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   250
             250 (* FUDGE *) true
38454
9043eefe8d71 detect old Vampire and give a nicer error message
blanchet
parents: 38433
diff changeset
   251
9043eefe8d71 detect old Vampire and give a nicer error message
blanchet
parents: 38433
diff changeset
   252
(* Setup *)
9043eefe8d71 detect old Vampire and give a nicer error message
blanchet
parents: 38433
diff changeset
   253
40059
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   254
fun add_atp (name, config) thy =
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   255
  Data.map (Symtab.update_new (name, (config, stamp ()))) thy
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   256
  handle Symtab.DUP name => error ("Duplicate ATP: " ^ quote name ^ ".")
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   257
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   258
fun get_atp thy name =
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   259
  the (Symtab.lookup (Data.get thy) name) |> fst
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   260
  handle Option.Option => error ("Unknown ATP: " ^ name ^ ".")
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   261
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   262
val available_atps = Symtab.keys o Data.get
36371
8c83ea1a7740 move the Sledgehammer menu options to "sledgehammer_isar.ML"
blanchet
parents: 36370
diff changeset
   263
40059
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   264
fun is_atp_installed thy name =
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   265
  let val {exec, required_execs, ...} = get_atp thy name in
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   266
    forall (curry (op <>) "" o getenv o fst) (exec :: required_execs)
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   267
  end
36371
8c83ea1a7740 move the Sledgehammer menu options to "sledgehammer_isar.ML"
blanchet
parents: 36370
diff changeset
   268
40059
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   269
fun refresh_systems_on_tptp () =
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   270
  Synchronized.change systems (fn _ => get_systems ())
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   271
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   272
val atps = [e, spass, vampire, remote_e, remote_vampire, remote_sine_e,
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   273
            remote_snark]
6ad9081665db use consistent terminology in Sledgehammer: "prover = ATP or SMT solver or ..."
blanchet
parents: 39491
diff changeset
   274
val setup = fold add_atp atps
35867
16279c4c7a33 move all ATP setup code into ATP_Wrapper
blanchet
parents: 35865
diff changeset
   275
28592
824f8390aaa2 renamed AtpThread to AtpWrapper;
wenzelm
parents:
diff changeset
   276
end;