Made feedback set algorithm not ignore the nodes post cycle.

[orizi]

Fixes #5524

---

This change is 

[gilbens-starkware]

Reviewed 2 of 3 files at r1, 1 of 1 files at r2.
Reviewable status: all files reviewed (commit messages unreviewed), 1 unresolved discussion (waiting on @orizi and @yuvalsw)

---

crates/cairo-lang-utils/src/graph_algos/feedback_set.rs line 33 at r2 (raw file):

    /// reached, it indicates it's in some cycle that was not "resolved" yet.
    in_flight: UnorderedHashSet<Node::NodeId>,
    /// Node already visited in the current run.

Suggestion:

/// Already visited nodes in the current run.

[gilbens-starkware]

Reviewed 1 of 1 files at r3, all commit messages.
Reviewable status: complete! all files reviewed, all discussions resolved (waiting on @yuvalsw)

[gilbens-starkware]

Reviewed 1 of 1 files at r4, all commit messages.
Reviewable status: complete! all files reviewed, all discussions resolved (waiting on @yuvalsw)

[yuvalsw]

Reviewed 2 of 3 files at r1, 1 of 1 files at r4, all commit messages.
Reviewable status: all files reviewed, 2 unresolved discussions

---

crates/cairo-lang-utils/src/graph_algos/feedback_set.rs line 25 at r4 (raw file):

struct FeedbackSetAlgoContext<Node: ComputeScc> {
    /// Nodes that were discovered as reachable from the current run, but possibly were not yet
    /// visited.

please elaborate about the "possibly" here - what are the cases where they were visited and still pending? (IIUC, if visited after being added to pending)

Code quote:

    /// Nodes that were discovered as reachable from the current run, but possibly were not yet
    /// visited.

---

crates/cairo-lang-utils/src/graph_algos/feedback_set_test.rs line 144 at r4 (raw file):

        vec![1, 2],
        // 1:
        vec![1, 0],

Did this fail before the fix?
I think the bug was only relevant in the case of [0, 1]. Here you first find the size-1 cycle [1], and then don't add 0 to the fset and don't break. If it was [0, 1] (before the fix) you would not add 1, add 0 and break, and skip 2, missing the size-1 cycle [2].

Code quote:

vec![1, 0],

[orizi]

Reviewable status: all files reviewed, 2 unresolved discussions

---

crates/cairo-lang-utils/src/graph_algos/feedback_set_test.rs line 144 at r4 (raw file):

Previously, yuvalsw wrote…

Did this fail before the fix?
I think the bug was only relevant in the case of [0, 1]. Here you first find the size-1 cycle [1], and then don't add 0 to the fset and don't break. If it was [0, 1] (before the fix) you would not add 1, add 0 and break, and skip 2, missing the size-1 cycle [2].

it failed only on the case starting from 0.
a similar case failed for more. (where the edges were sorted)