fix: raise RUST_MIN_STACK repo-wide to fix e2e test stack overflows

[wbbradley]

Problem

E2e tests (test_blob_attribute_add_and_remove, test_byte_range_read_client, test_aggregator_blob_endpoints_corner_cases, and others run via --run-ignored) abort with fatal runtime error: stack overflow when built with the default debug profile (opt-level = 0), for example a plain local cargo nextest run --run-ignored ....

The root cause is deep async call chains through the Sui SDK futures (Wallet::execute_transaction_may_fail ≈ 10 KB, WalletContext::execute_transaction_may_fail ≈ 9.7 KB, and friends). In debug builds these frames are large and un-inlined, so they compose past the default thread stack.

This is a local-only issue, not a CI failure

CI is not affected and is green on main. The test job runs with --cargo-profile test-opt (Cargo.toml:202, opt-level = 1), which shrinks those stack frames enough to stay under the default stack even though it runs the same --run-ignored all e2e tests. The real axis is debug-vs-optimized builds, not macOS-vs-Linux.

Fix

Add a workspace-root .cargo/config.toml with a Cargo [env] table setting RUST_MIN_STACK to 32 MiB. This lets developers run the e2e suite with a default debug build without manually exporting RUST_MIN_STACK per invocation. Cargo applies [env] to all processes it launches (test binaries under both cargo test and cargo nextest run, build scripts, cargo run), so it is honored uniformly. The value is non-force, so a developer who already exports RUST_MIN_STACK keeps their own value.

The libtest harness runs each test on its own spawned thread (the overflow message names the thread after the test, for example thread 'deep_stack' has overflowed its stack), and that thread honors RUST_MIN_STACK at creation time. cargo-nextest 0.9.108 propagates the Cargo [env] table into the test process.

Why this over alternatives

• build.rs linker flags (-Wl,-stack_size / -Wl,-z,stacksize): only reliably resize the main thread stack on macOS, and tests don't run on the main thread anyway — so they wouldn't help the per-test spawned threads where the overflow actually occurs.
• multi_thread, worker_threads = 1: changes async test semantics workspace-wide — risky and broad for no benefit.
• Box more futures: diminishing returns; the largest frames live in the Sui SDK.
• Per-invocation RUST_MIN_STACK export: easy to forget and undocumented; .cargo/config.toml makes it automatic and discoverable.

Notes

• 32 MiB is a minimum reserve per spawned thread and is virtual address space only; RSS grows solely for touched pages. No measurable runtime cost.
• Harmless in CI: CI already builds optimized and never overflowed, so this is a no-op there. It applies regardless, which is fine.
• Does not affect deployed node binaries (executed directly, not via cargo run).

Verification

• chk — clean (config-only change; no compiler impact).
• cargo nextest run — 1106 tests passed, no regressions.

🤖 Generated with Claude Code