-
Notifications
You must be signed in to change notification settings - Fork 50
/
eval_env.py
245 lines (217 loc) · 8.46 KB
/
eval_env.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
# Environment for interacting with Coq theorems during testing
from serapi import SerAPI, CoqExn, CoqTimeout
import re
from copy import deepcopy
from time import time
from collections import OrderedDict
import json
import sexpdata
from utils import update_env
import os
from glob import glob
import pdb
class ProofEnv:
"""
the interactive environment for proving a theorem
should be created by FileEnv
"""
def __init__(self, proof, serapi, max_num_tactics, timeout):
"""
proof - the data structure for a proof in CoqGym
serapi - the SerAPI at the beginning of the proof
max_num_tactics - the maxium number of interactions for proving a theorem
timeout - the maxium amount of time (in seconds) for each theorem
"""
self.proof = proof
self.serapi = serapi
self.num_tactics_left = max_num_tactics
self.timeout = timeout
self.success = False
self.failure = False
def init(self):
"return the initial goals and the global environment of the theorem"
fg_goals, bg_goals, shelved_goals, given_up_goals = self.serapi.query_goals()
assert shelved_goals + given_up_goals == []
self.start_time = time()
return self.feedback(
"PROVING",
env=self.proof["env"],
fg_goals=fg_goals,
bg_goals=bg_goals,
shelved_goals=shelved_goals,
given_up_goals=given_up_goals,
)
def step(self, command):
"""
perform a single interaction
the agent provides a command and get feedback from Coq
valid commands include:
tactics
Admitted - to give up the proof
Undo - to backtrack one step
other valid Coq commands
"""
if self.success:
return self.feedback("ALREADY_SUCCEEDED")
if self.failure:
return self.feedback("ALREADY_FAILED")
time_left = self.timeout - (time() - self.start_time)
print("%d: %s: %.02f" % (self.num_tactics_left, command, time_left))
if time_left <= 0:
return self.feedback("MAX_TIME_REACHED")
self.serapi.push() # save the state before executing the command
try:
ast = sexpdata.dumps(self.serapi.query_ast(command))
if "VernacExtend" in ast: # is a tactic
if self.num_tactics_left <= 0:
self.serapi.pop()
return self.feedback("MAX_NUM_TACTICS_REACHED")
self.num_tactics_left -= 1
command = "timeout %d (%s)." % (time_left, command[:-1])
responses, _ = self.serapi.execute(command)
states_cnt = self.serapi.pull() # delete the saved state if no error
except CoqExn as ex:
self.serapi.pop() # restore the state
return self.feedback("ERROR", error=ex)
except CoqTimeout as ex:
self.serapi.shutdown()
return self.feedback("ERROR", error=ex)
if "(VernacEndProof Admitted)" in ast:
self.failure = True
return self.feedback("GIVEN_UP")
try:
(
fg_goals,
bg_goals,
shelved_goals,
given_up_goals,
) = self.serapi.query_goals()
time_left = self.timeout - (time() - self.start_time)
if time_left < 0:
return self.feedback("MAX_TIME_REACHED")
except CoqTimeout as ex:
self.serapi.shutdown()
return self.feedback("ERROR", error=ex)
if fg_goals + bg_goals + shelved_goals + given_up_goals == []: # no goals left
self.success = True
return self.feedback("SUCCESS")
elif shelved_goals + given_up_goals != []:
self.serapi.pop_n(states_cnt)
return self.feedback("ERROR", error=(shelved_goals, given_up_goals))
else:
return self.feedback(
"PROVING",
fg_goals=fg_goals,
bg_goals=bg_goals,
shelved_goals=shelved_goals,
given_up_goals=given_up_goals,
)
def feedback(self, result, **kwargs):
obs = {
"result": result,
"num_tactics_left": self.num_tactics_left,
"time_left": self.timeout - (time() - self.start_time),
}
for k, v in kwargs.items():
obs[k] = v
return obs
class FileEnv:
"""
a generator of the theorems in a Coq file
the agent iterate through them and interact with each theorem sequentially
"""
def __init__(
self, filename, max_num_tactics, timeout, with_hammer=None, hammer_timeout=None
):
"""
filename - the json file in CoqGym
max_num_tactics - the maxium number of interactions for proving a theorem
timeout - the maxium amount of time (in seconds) for each theorem
"""
file_data = json.load(open(filename))
self.vernac_cmds = [cmd[:2] for cmd in file_data["vernac_cmds"]]
self.proofs = file_data["proofs"]
self.max_num_tactics = max_num_tactics
self.timeout = timeout
self.with_hammer = with_hammer
self.hammer_timeout = hammer_timeout
self.serapi = self.initialize_serapi()
def initialize_serapi(self):
serapi = SerAPI(timeout=1200)
if self.with_hammer is not None:
atp_limit = 29 * self.hammer_timeout // 60
reconstr_limit = 28 * self.hammer_timeout // 60
crush_limit = 3 * self.hammer_timeout // 60
serapi.execute(
"From Hammer Require Import Hammer. Set Hammer ATPLimit %d. Set Hammer ReconstrLimit %d. Set Hammer CrushLimit %d."
% (atp_limit, reconstr_limit, crush_limit)
)
if self.with_hammer == "Z3":
serapi.execute(
"Unset Hammer Vampire. Unset Hammer Eprover. Unset Hammer CVC4."
)
elif self.with_hammer == "Vampire":
serapi.execute(
"Unset Hammer Z3. Unset Hammer Eprover. Unset Hammer CVC4."
)
elif self.with_hammer == "Eprover":
serapi.execute(
"Unset Hammer Z3. Unset Hammer Vampire. Unset Hammer CVC4."
)
elif self.with_hammer == "CVC4":
serapi.execute(
"Unset Hammer Z3. Unset Hammer Vampire. Unset Hammer Eprover."
)
else:
assert self.with_hammer == "All"
return serapi
def __iter__(self):
self.cmd_idx = 0
self.next_proof_idx = 0
self.env = {"constants": [], "inductives": []}
return self
def __next__(self):
if self.next_proof_idx >= len(self.proofs): # no more theorem
raise StopIteration
next_proof = self.proofs[self.next_proof_idx]
self.env = update_env(self.env, next_proof["env_delta"])
del next_proof["env_delta"]
next_proof["env"] = self.env
if self.serapi.dead:
self.serapi = self.initialize_serapi()
self.cmd_idx = 0
elif self.next_proof_idx > 0: # rollback to the start of the current proof
self.serapi.pop()
if (
self.next_proof_idx > 0 and self.next_proof_idx % 30 == 0
): # renew the SerAPI to prevent stack overflow
self.serapi.clean()
self.serapi = self.initialize_serapi()
self.cmd_idx = 0
while self.cmd_idx <= next_proof["line_nb"]:
cmd, _ = self.vernac_cmds[self.cmd_idx]
self.serapi.execute(cmd)
self.cmd_idx += 1
self.serapi.push()
self.next_proof_idx += 1
return ProofEnv(next_proof, self.serapi, self.max_num_tactics, self.timeout)
def __len__(self):
return len(self.proofs)
def test(self):
assert self.cmd_idx is None
self.serapi.push()
for cmd, _ in self.vernac_cmds:
self.serapi.execute(cmd)
self.serapi.pop()
def __enter__(self):
return self
def __exit__(self, exc_type, exc_value, traceback):
self.serapi.clean()
if __name__ == "__main__":
f = "data/StructTact/Assoc.json"
print(f)
with FileEnv(f, max_num_tactics=100, timeout=600) as file_env:
for proof_env in file_env:
print(proof_env.proof["name"])
proof_env.init()
print(proof_env.step("Admitted."))