Preventing infinite inlining of cycle functions.

[orizi]

Summary

Fixes a hang that occurred when lowering self-referential Destruct implementations in no-gas mode (issue #9894). The cycle detection for inlining now uses LoweringStage::PreOptimizations instead of LoweringStage::Monomorphized, which correctly identifies recursive destructors as cyclic and prevents infinite inlining. The TODO comment about needing better logic to avoid inlining destructors that call themselves has been removed, as this fix addresses that concern.

A no_gas test flag and corresponding LoweringDatabaseForTesting::with_no_gas() snapshot are introduced to enable lowering tests that disable withdraw_gas insertion, along with a regression test covering the self-referential Destruct case.

---

Type of change

Please check one:

• Bug fix (fixes incorrect behavior)
• New feature
• Performance improvement
• Documentation change with concrete technical impact
• Style, wording, formatting, or typo-only change

---

Why is this change needed?

In no-gas mode, a struct with a Destruct impl that calls itself (directly or indirectly) caused the compiler to hang indefinitely during inlining. The cycle check was running against LoweringStage::Monomorphized, which did not detect the cycle at the point where inlining decisions are made, allowing unbounded recursive inlining attempts.

---

What was the behavior or documentation before?

Compiling a self-referential Destruct impl with gas disabled would cause the compiler to hang without producing diagnostics or output.

---

What is the behavior or documentation after?

The compiler correctly detects the cycle, skips inlining for the recursive destructor, and produces valid lowered output without hanging.

---

Related issue or discussion (if any)

Fixes #9894

---

Additional context

The SHARED_DB static in test_utils.rs was also changed from pub to private, as it was not intended to be part of the public API.

[reviewable-StarkWare]

This change is 

[orizi]

• Preventing infinite inlining of cycle functions. #9917 👈 (View in Graphite)
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[cursor]

PR Summary

Medium Risk
Changes the inliner’s cycle detection stage, which can alter inlining behavior and generated lowered code in recursive/cyclic call scenarios. Adds new no-gas testing mode and regression coverage, reducing risk of future hangs but still touching compiler optimization logic.

Overview
Fixes an inlining hang in no-gas lowering by switching the inliner’s cycle check in priv_should_inline to use LoweringStage::PreOptimizations, preventing recursive/cyclic callees (notably self-referential Destruct impls) from being inlined indefinitely.

Extends the lowering test runner with a no_gas flag backed by LoweringDatabaseForTesting::with_no_gas() (disables ADD_WITHDRAW_GAS) and adds a regression case in test_data/cycles; also removes an outdated TODO about destructor self-inlining.

^{Reviewed by Cursor Bugbot for commit 7789074. Bugbot is set up for automated code reviews on this repo. Configure here.}

[TomerStarkware]

@TomerStarkware reviewed 4 files and all commit messages, and made 1 comment.
Reviewable status: complete! all files reviewed, all discussions resolved (waiting on orizi).