perf: sanitize build artifacts using ASAN/LSAN

[bartlomieju]

This PR is intended to sanitize more build targets.

Currently only libdeno_test is built and run with ASAN/LSAN

[ry]

Great - so this failed with a very readable error:

Direct leak of 3393160 byte(s) in 1 object(s) allocated from:
    #0 0x5586899ae30d in operator new[](unsigned long) /b/swarming/w/ir/cache/builder/src/third_party/llvm/compiler-rt/lib/asan/asan_new_delete.cc:102:3
    #1 0x55868b01e8ee in v8::internal::Snapshot::CreateSnapshotBlob(v8::internal::SnapshotData const*, v8::internal::SnapshotData const*, std::__1::vector<v8::internal::SnapshotData*, std::__1::allocator<v8::internal::SnapshotData*> > const&, bool) (/home/travis/build/denoland/deno/target/debug/deno_core_test+0x28198ee)
    #2 0x558689a262fd in v8::SnapshotCreator::CreateBlob(v8::SnapshotCreator::FunctionCodeHandling) (/home/travis/build/denoland/deno/target/debug/deno_core_test+0x12212fd)
    #3 0x55868cd9a820 in deno_snapshot_new core/libdeno/api.cc:101:37
    #4 0x55868cb88ebd in deno_core_test_bin::isolate::Isolate::snapshot::h8f29014824d4a2a2 core/isolate.rs:471:28
    #5 0x5586899e3af9 in __rust_maybe_catch_panic /rustc/eae3437dfe991621e8afdc82734f4a172d7ddf9b/src/libpanic_unwind/lib.rs:82:7
SUMMARY: AddressSanitizer: 3393160 byte(s) leaked in 1 allocation(s).

It says we're leaking the snapshot blob. That seems not unlikely.

[bartlomieju]

@ry adding this bit makes LSAN happy. I followed this test and it indeed seems to fix the issue.

[bartlomieju]

But makes tests fail 😓 still digging...

[bartlomieju]

Okay so I figured what causes this leak, however I'm not yet sure how to fix it.

The problem is in will_snapshot test case in //cli/isolate.rs; this test does a snapshot of a short script and then tries to instantiate a new isolate with that snapshot. Isolate.snapshot() returns Snapshot1 which explicitly says in comment:

/// The type returned from deno_snapshot_new. Needs to be dropped.

During creation of new isolate (Isolate::new) this snapshot is passed in deno_config as load_snapshot (which has a type of Snapshot2 not Snapshot1). However there is no way to tell that this is the type of snapshot on C++ side and Rust's borrow checker prevents an explicitly drop after calling libdeno::deno_new.

[bartlomieju]

@ry in light of #2825 I believe Isolate::snapshot will become absolete and this PR won't make much sense. I'm closing it for now and we can revisit this topic after #2825 lands.

[bartlomieju]

@ry in light of #2825 I believe Isolate::snapshot will become absolete and this PR won't make much sense. I'm closing it for now and we can revisit this topic after #2825 lands.

After further reviewing latest changes this PR still makes sense

[ry]

@bartlomieju Isolate::snapshot will stick around, it's how deno_typescript makes snapshots.

That said, I think more work needs to be put into fixing the memory ownership situation and merging the Snapshot1 Snapshot2 types.

[ry]

Closing because it's old. We've since removed the ASAN test completely when we moved to "cargo build"... Hoping to bring that back at some point when Rust supports it in stable.