The forc debug
CLI enables debugging a live transaction on a running Fuel Client node.
First, we need a project to debug, so create a new project using
forc new --script dbg_example && cd dbg_example
And then add some content to src/main.sw
, for example:
script;
use std::logging::log;
fn factorial(n: u64) -> u64 {
let mut result = 1;
let mut counter = 0;
while counter < n {
counter = counter + 1;
result = result * counter;
}
return result;
}
fn main() {
log::<u64>(factorial(5)); // 120
}
Now we are ready to build the project.
forc build
After this the resulting binary should be located at out/debug/dbg_example.bin
. Because we are interested in the resulting bytecode, we can read that with:
forc parse-bytecode out/debug/dbg_example.bin
Which should give us something like
half-word byte op raw notes
0 0 JI { imm: 4 } 90 00 00 04 jump to byte 16
1 4 NOOP 47 00 00 00
2 8 InvalidOpcode 00 00 00 00 data section offset lo (0)
3 12 InvalidOpcode 00 00 00 44 data section offset hi (68)
4 16 LW { ra: 63, rb: 12, imm: 1 } 5d fc c0 01
5 20 ADD { ra: 63, rb: 63, rc: 12 } 10 ff f3 00
6 24 MOVE { ra: 18, rb: 1 } 1a 48 10 00
7 28 MOVE { ra: 17, rb: 0 } 1a 44 00 00
8 32 LW { ra: 16, rb: 63, imm: 0 } 5d 43 f0 00
9 36 LT { ra: 16, rb: 17, rc: 16 } 16 41 14 00
10 40 JNZI { ra: 16, imm: 13 } 73 40 00 0d conditionally jump to byte 52
11 44 LOG { ra: 18, rb: 0, rc: 0, rd: 0 } 33 48 00 00
12 48 RET { ra: 0 } 24 00 00 00
13 52 ADD { ra: 17, rb: 17, rc: 1 } 10 45 10 40
14 56 MUL { ra: 18, rb: 18, rc: 17 } 1b 49 24 40
15 60 JI { imm: 8 } 90 00 00 08 jump to byte 32
16 64 NOOP 47 00 00 00
17 68 InvalidOpcode 00 00 00 00
18 72 InvalidOpcode 00 00 00 05
We can recognize the while
loop by the conditional jumps JNZI
. The condition just before the first jump can be identified by LT
instruction (for <
). Some notable instructions that are generated only once in our code include MUL
for multiplication and LOG {.., 0, 0, 0}
from the log
function.
We can start up the debug infrastructure. On a new terminal session run fuel-core run --db-type in-memory --debug
; we need to have that running because it actually executes the program. Now we can fire up the debugger itself: forc-debug
. Now
if everything is set up correctly, you should see the debugger prompt (>>
). You can use help
command to list available commands.
Now we would like to inspect the program while it's running. To do this, we first need to send the script to the executor, i.e. fuel-core
. To do so, we need a transaction specification, tx.json
. It looks something like this:
{
"Script": {
"body": {
"script_gas_limit": 1000000,
"script": [
144,
0,
0,
4,
71,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
68,
93,
252,
192,
1,
16,
255,
243,
0,
26,
72,
16,
0,
26,
68,
0,
0,
93,
67,
240,
0,
22,
65,
20,
0,
115,
64,
0,
13,
51,
72,
0,
0,
36,
0,
0,
0,
16,
69,
16,
64,
27,
73,
36,
64,
144,
0,
0,
8,
71,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
5
],
"script_data": [],
"receipts_root": "0000000000000000000000000000000000000000000000000000000000000000"
},
"policies": {
"bits": "MaxFee",
"values": [
0,
0,
0,
0
]
},
"inputs": [
{
"CoinSigned": {
"utxo_id": {
"tx_id": "c49d65de61cf04588a764b557d25cc6c6b4bc0d7429227e2a21e61c213b3a3e2",
"output_index": 18
},
"owner": "f1e92c42b90934aa6372e30bc568a326f6e66a1a0288595e6e3fbd392a4f3e6e",
"amount": 10599410012256088000,
"asset_id": "2cafad611543e0265d89f1c2b60d9ebf5d56ad7e23d9827d6b522fd4d6e44bc3",
"tx_pointer": {
"block_height": 0,
"tx_index": 0
},
"witness_index": 0,
"maturity": 0,
"predicate_gas_used": null,
"predicate": null,
"predicate_data": null
}
}
],
"outputs": [],
"witnesses": [
{
"data": [
156,
254,
34,
102,
65,
96,
133,
170,
254,
105,
147,
35,
196,
199,
179,
133,
132,
240,
208,
149,
11,
46,
30,
96,
44,
91,
121,
195,
145,
184,
159,
235,
117,
82,
135,
41,
84,
154,
102,
61,
61,
16,
99,
123,
58,
173,
75,
226,
219,
139,
62,
33,
41,
176,
16,
18,
132,
178,
8,
125,
130,
169,
32,
108
]
}
]
}
}
However, the key script
should contain the actual bytecode to execute, i.e. the contents of out/debug/dbg_example.bin
as a JSON array. The following command can be used to generate it:
python3 -c 'print(list(open("out/debug/dbg_example.bin", "rb").read()))'
So now we replace the script array with the result, and save it as tx.json
.
Now we can actually execute the script:
>> start_tx tx.json
Receipt: Log { id: 0000000000000000000000000000000000000000000000000000000000000000, ra: 120, rb: 0, rc: 0, rd: 0, pc: 10380, is: 10336 }
Receipt: Return { id: 0000000000000000000000000000000000000000000000000000000000000000, val: 0, pc: 10384, is: 10336 }
Receipt: ScriptResult { result: Success, gas_used: 60 }
Terminated
Looking at the first output line, we can see that it logged ra: 120
which is the correct return value for factorial(5)
. It also tells us that the execution terminated without hitting any breakpoints. That's unsurprising, because we haven't set up any. We can do so with breakpoint
command:
>> breakpoint 0
>> start_tx tx.json
Receipt: ScriptResult { result: Success, gas_used: 0 }
Stopped on breakpoint at address 0 of contract 0x0000000000000000000000000000000000000000000000000000000000000000
Now we have stopped execution at the breakpoint on entry (address 0
). We can now inspect the initial state of the VM.
>> register ggas
reg[0x9] = 1000000 # ggas
>> memory 0x10 0x8
000010: e9 5c 58 86 c8 87 26 dd
However, that's not too interesting either, so let's just execute until the end, and then reset the VM to remove the breakpoints.
>> continue
Receipt: Log { id: 0000000000000000000000000000000000000000000000000000000000000000, ra: 120, rb: 0, rc: 0, rd: 0, pc: 10380, is: 10336 }
Receipt: Return { id: 0000000000000000000000000000000000000000000000000000000000000000, val: 0, pc: 10384, is: 10336 }
Terminated
>> reset
Next, we will setup a breakpoint to check the state on each iteration of the while
loop. For instance, if we'd like to see what numbers get multiplied together, we could set up a breakpoint before the operation. The bytecode has only a single MUL
instruction:
half-word byte op raw notes
14 56 MUL { ra: 18, rb: 18, rc: 17 } 1b 49 24 40
We can set a breakpoint on its address, at halfword-offset 14
.
>>> breakpoint 14
>> start_tx tx.json
Receipt: ScriptResult { result: Success, gas_used: 9 }
Stopped on breakpoint at address 56 of contract 0x0000000000000000000000000000000000000000000000000000000000000000
Now we can inspect the inputs to multiply. Looking at the specification tells us that the instruction MUL { ra: 18, rb: 18, rc: 17 }
means reg[18] = reg[18] * reg[17]
. So inspecting the inputs tells us that
>> r 18 17
reg[0x12] = 1 # reg18
reg[0x11] = 1 # reg17
So on the first round the numbers are 1
and 1
, so we can continue to the next iteration:
>> c
Stopped on breakpoint at address 56 of contract 0x0000000000000000000000000000000000000000000000000000000000000000
>> r 18 17
reg[0x12] = 1 # reg18
reg[0x11] = 2 # reg17
And the next one:
>> c
Stopped on breakpoint at address 56 of contract 0x0000000000000000000000000000000000000000000000000000000000000000
>> r 18 17
reg[0x12] = 2 # reg18
reg[0x11] = 3 # reg17
And fourth one:
>> c
Stopped on breakpoint at address 56 of contract 0x0000000000000000000000000000000000000000000000000000000000000000
>> r 18 17
reg[0x12] = 6 # reg18
reg[0x11] = 4 # reg17
And round 5:
>> c
Stopped on breakpoint at address 56 of contract 0x0000000000000000000000000000000000000000000000000000000000000000
>> r 18 17
reg[0x12] = 24 # reg18
reg[0x11] = 5 # reg17
At this point we can look at the values
17 | 18 |
---|---|
1 | 1 |
2 | 1 |
3 | 2 |
4 | 6 |
5 | 24 |
From this we can clearly see that the left side, register 17
is the counter
variable, and register 18
is result
. Now the counter equals the given factorial function argument 5
, and the loop terminates. So when we continue, the program finishes without encountering any more breakpoints:
>> c
Receipt: Log { id: 0000000000000000000000000000000000000000000000000000000000000000, ra: 120, rb: 0, rc: 0, rd: 0, pc: 10380, is: 10336 }
Receipt: Return { id: 0000000000000000000000000000000000000000000000000000000000000000, val: 0, pc: 10384, is: 10336 }
Terminated