[Bug/Enhance]: Loop fails to terminate when loop argument is negative.

[Enkryp]

Bug

Infinite loops for repeat (negative) [...]

Fix

In visitLoop in builder.py:

change
loopCond = ChironAST.ConditionCommand(ChironAST.NEQ(counterVar, constZero))
to
loopCond = ChironAST.ConditionCommand(ChironAST.GT(counterVar, constZero))

otherwise fuzzers will go into infinite loops if repeat has a fuzzing var.

version: 1.0.4

[lahiri-phdworks]

@Enkryp This would change the semantics of the way we handle loops; if you check the IR for loops, we need the loop variable to be positive since we perform a decrement. Example below, the __rep_counter_3 must be 0 or greater, since we just want to loop through, we don't support conditional loops yet.

[L17] :__rep_counter_3 = 5 [1]
[L18] (:__rep_counter_3 != 0) [5]
[L19] forward :p [1]
[L20] left 72 [1]
[L21] :__rep_counter_3 = (:__rep_counter_3 - 1) [1]

[lahiri-phdworks]

However, GTE may work fine I guess.

[Enkryp]

@Enkryp This would change the semantics of the way we handle loops; if you check the IR for loops, we need the loop variable to be positive since we perform a decrement. Example below, the __rep_counter_3 must be 0 or greater, since we just want to loop through, we don't support conditional loops yet.

[L17] :__rep_counter_3 = 5 [1]
[L18] (:__rep_counter_3 != 0) [5]
[L19] forward :p [1]
[L20] left 72 [1]
[L21] :__rep_counter_3 = (:__rep_counter_3 - 1) [1]

What I meant was if there is a statement :

repeat x [
...
]

and this is passed to fuzzer, fuzzer might try to check this for x = -1 say, that will never terminate.

[lahiri-phdworks]

Yeah, so we will change to greater than equal, for x = -1, we won't even enter the loop

x = -1
if (x >= 0) {
....
....
x = x - 1;
}

[lahiri-phdworks]

We shall fix it in the next release. This also affects regular execution. Thanks @Enkryp

[Enkryp]

Yeah, so we will change to greater than equal, for x = -1, we won't even enter the loop

x = -1
if (x >= 0) {
....
....
x = x - 1;
}

shouldn't it be greater than ? if x = 0 we still dont want to enter.

[lahiri-phdworks]

I see. You are right here.

[lahiri-phdworks]

@Enkryp This has now been fixed. Thanks!