Move every example onto Anchor v2's LiteSVM test harness

[mikemaccana]

Anchor v2 ships its own LiteSVM wrapper, anchor-v2-testing. Tests that call LiteSVM::new() directly still run and still pass, but they record nothing, so anchor test --profile, anchor debugger and anchor coverage all come back empty. Going through anchor_v2_testing::svm() is what connects a test to those three tools, and it is LiteSVM::new() until the profile feature turns tracing on, so the default path costs nothing.

What changes

Per program crate, the dev-dependencies drop litesvm, solana-signer and solana-keypair, which the harness re-exports, and gain the harness plus a profile feature forwarding to it:

[dev-dependencies]
anchor-v2-testing = { git = "https://github.com/otter-sec/anchor.git", rev = "3e38a121b969d1a77230baeb0056baf5cc9b3c1a" }

[features]
profile = ["anchor-v2-testing/profile"]

Tests swap LiteSVM::new() for anchor_v2_testing::svm() and their imports follow. The execution model does not change: same in-process LiteSVM, same solana-kite helpers, which take the same LiteSVM value.

Why a git dependency, and why a pinned rev

crates.io has no anchor-v2-testing; a git dependency is what the v2 testing documentation prescribes. That documentation says branch = "anchor-next". This pins a revision instead, because Cargo.lock is not tracked here, so CI resolves fresh on every run and a branch would let an upstream push change a build with no commit in this repository to explain it. Bumping the rev then becomes a deliberate, reviewable change. Happy to switch to the branch if you would rather track it.

What this unlocks

None of these work today, because nothing records traces:

• anchor test --profile writes per-test SBF traces and renders flamegraphs
• anchor debugger opens an instruction stepper with DWARF source mapping and CPI frames
• anchor coverage emits LCOV at target/coverage/sbf.lcov

Verified on basics/counter: anchor coverage resolved 294 of 312 program counters to source across 19 files, 91 lines covered, with per-line hit counts in the LCOV.

Verification

A full local build-and-test sweep over all 55 Anchor projects: 54 pass, 286 tests.

The one failure, finance/vault-strategy's test_add_asset_enforces_max_assets, is not caused by this change. It fails identically on unmigrated origin/main with ProgramLoad("Entrypoint out of bounds") while loading mock_swap_router.so, and it passes in CI, so it is an artefact of the locally-built cargo-build-sbf 4.2.0 rather than the Solana 3.1.14 toolchain CI installs. Flagging it rather than folding a fix into this PR.

The sweep earned its keep. The first pass of this migration failed on 44 of 51 projects with unresolved import anchor_lang::anchor_v2_testing: the import was being inserted at the first alphabetically-later entry without regard to nesting depth, so in a block like

use {
    anchor_lang::{
        solana_program::instruction::Instruction, ...
    },

it landed inside anchor_lang. Five further files named the replaced crates by full path rather than importing them, mostly in helper signatures like fn setup() -> (LiteSVM, solana_keypair::Keypair), and dropping the dev-dependencies left those dangling. Both are fixed in the second commit.

Not included

The one test importing solana_transaction::Transaction keeps that dependency, because the harness re-exports VersionedTransaction and not Transaction. mock-swap-router has no tests, so it gains nothing.

---

Generated by Claude Code