Cleanup vm module memory leakage

[laverdet]

There are some paths here that led to dangling contexts. By being
smarter with handle management we can get rid of all the cleanup code
and fix those issues.

[isaacs]

Well, looks fine to me and doesn't seem to break anything too badly.

@bnoordhuis care to comment? You've also been in and around the vm code somewhat, if memory serves.

@laverdet If it seems like I'm being overly nitpicky, it's only because the vm module is one of the most grimy areas of node, and it's easy for small missteps to result in really subtle issues in other places (as we saw!)

[laverdet]

Yeah I understand being picky. I'm not 100% confident about the robustness of this patch, but I feel pretty good about it. All the unit tests pass, but this function does SO MANY THINGS that it's easy to mess up.

[bnoordhuis]

Can you add a comment here explaining why the code looks like that?

[isaacs]

So, this fixes the errors we were seeing in debug mode. I'm going to land it.

However, @laverdet, test/pummel/test-vm-memleak.js is still failing with this patch. That's not so bad, since it was failing on master without the patch, so it's not making it any worse. But was the intent to make that test pass? Or was this another thing, and I'm mis-remembering?

[isaacs]

Landed on 7063575

[laverdet]

@isaacs: This commit was to fix another bug. In some cases when you used the vm module there would be memory leaks, and sometimes those leaks would lead to crashes; those are gone now.

The crash in the pummel test is a different issue. I looked into it and it appears to be a v8 bug which has already been fixed. You can reproduce this easily in v8 3.9.24 (current v8 version in node master) with this quick program:

#include <v8.h>
using namespace v8;

static Handle<Value> thrower(const Arguments& args) {
    TryCatch try_catch;
    Local<Script> script = Script::Compile(String::New("throw 1;"));
    Local<Value> result = script->Run();
    return try_catch.HasCaught() ? try_catch.ReThrow() : result;
}

int main() {
    HandleScope scope;
    Persistent<Context> context = Context::New();
    Context::Scope context_scope(context);
    context->Global()->Set(String::New("thrower"), FunctionTemplate::New(thrower)->GetFunction());
    Local<Script> script = Script::Compile(String::New("for (var ii = 0; ii < 200; ++ii) try { thrower(); } catch(e) {}"));
    script->Run();
    context.Dispose();
    return 0;
}

The same program will not raise errors on bleeding_edge. I stopped researching it when I realized it's fixed in bleeding_edge so I don't know if there's a v8 ticket for it or not, or when it will land in a release. I believe it's a bug in the way v8 handles some AST walking. You can easily "fix" the unit test by changing require('vm').runInNewContext('throw 1;'); to require('vm').runInNewContext('new function() { throw 1; }');. Just putting the throw in a self-calling function is enough to workaround the bug.

[mfncooper]

Any chance we can get the vm memory leak fix back-ported to v0.6? We just ran into this issue in a production system, where the context leaks pile up over time and periodically cause the process to crash. Let me know if there's anything we can do to help get it back-ported.

[bnoordhuis]

@mfncooper I backported the patch in 7865c5c. It'll be in the next v0.6 release.

[mfncooper]

Thanks!