PoC of memory corruption issue in TS

[jgur-psyops]

Demonstrates a possible memory/process corruption issue that occurs on Linux when using LiteSVM with TS.

To preview:

cd scripts
yarn install
yarn repro

this will hang indefinitely on BigNumber("100000000").div("1000000"). Set MODE = "skip", or send a different ix (such as a simple SOL transfer) and it will complete. Creating an ATA always triggers the issue.

Possible Fix

Enabling stricter_abi_and_runtime_constraints fixes the issue. E.g. add

const featureSet = FeatureSet.allEnabled();
svm.withFeatureSet(featureSet);

Or comment back in the block starting with if (process.env.FEATURE_SOURCE === "repo") { and observe that

this hangs

FEATURE_SOURCE=repo MODE=send timeout 15s node --loader ts-node/esm repro.ts

this exits cleanly

FEATURE_SOURCE=repo EXTRA_FEATURES=Eoh7e1sDqtyPtuiWAhBNSJinvtJWTTDgeUMRi3RF8zWS MODE=send timeout 15s node --loader ts-node/esm repro.ts

Forcing the BPF loader to user the interpreter instead of the x86 SBPF JIT also seems to work.

It seems that with stricter_abi_and_runtime_constraints creating ATA runs through the old ABI path. A more clever permanent fix is probably possible but beyond the scope of this PR.

Versions Used:

Ubuntu 24.04.2 LTS
➜  scripts git:(ts-corruption-poc) node -v       
v24.16.0
➜  scripts git:(ts-corruption-poc) yarn -v       
4.14.1