mergeOSREntryValue is wrong when the incoming value does not match up with the flush format

[Saam Barati]

Our profiling is good, so we never really run into this issue. We'd probably hit this bug way more often if we random-fuzzed the value injection types.

However, once we've locked down a Variable's flushFormat, it's wrong to give it a type wider than that. E.g, we even assert that much in AI:

```
    case GetLocal: {
        VariableAccessData* variableAccessData = node->variableAccessData();
        AbstractValue value = m_state.operand(variableAccessData->local().offset());
        // The value in the local should already be checked.
        DFG_ASSERT(m_graph, node, value.isType(typeFilterFor(variableAccessData->flushFormat())));
        if (value.value())
            m_state.setFoundConstants(true);
        setForNode(node, value);
        break;
    }
```

[Saam Barati]

Created attachment 367438 [details]
patch

[Build Bot]

Attachment 367438 [details] did not pass style-queue:


ERROR: Source/JavaScriptCore/ChangeLog:17:  Please consider whether the use of security-sensitive phrasing could help someone exploit WebKit: fuzzer, fuzzing  [changelog/unwantedsecurityterms] [3]
Total errors found: 1 in 4 files


If any of these errors are false positives, please file a bug against check-webkit-style.

[Yusuke Suzuki]

Comment on attachment 367438 [details]
patch

r=me

[WebKit Commit Bot]

Comment on attachment 367438 [details]
patch

Clearing flags on attachment: 367438

Committed r244287: <https://trac.webkit.org/changeset/244287>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.

[Radar WebKit Bug Importer]

<rdar://problem/49915815>