Fix SQLStatementBackend::m_resultSet to use CrossThreadPersistent

Fix SQLStatementBackend::m_resultSet to use CrossThreadPersistent

The SQLResultSet in SQLStatementBackend::m_resultSet is created on the main thread (in SQLStatementBackend::SQLStatementBackend) but was accessed from the db thread in SQLStatementBackend::execute.

BUG=681488

[keishi, 2017-01-17 12:41:39]

keishi@chromium.org changed reviewers:
+ haraken@chromium.org, oilpan-reviews@chromium.org

[keishi, 2017-01-17 12:41:40]

PTAL

m_resultSet is created on the main thread and SQLStatementBackend::execute runs
on the db thread so m_resultSet needs to be CrossThreadPersistent. 
In SQLStatementBackend::execute, the reference to SQLResultSetRowList from the
stack should be CrossThreadPersistent.
Changed SQLResultSet::m_rows to const so that m_rows is guaranteed to be alive
while we have a CrossThreadPersistent to SQLResultSet in m_resultSet.

[sof, 2017-01-17 14:02:22]

On 2017/01/17 12:41:40, keishi wrote:
> PTAL
> 
> m_resultSet is created on the main thread and SQLStatementBackend::execute
runs
> on the db thread so m_resultSet needs to be CrossThreadPersistent. 
> In SQLStatementBackend::execute, the reference to SQLResultSetRowList from the
> stack should be CrossThreadPersistent.
> Changed SQLResultSet::m_rows to const so that m_rows is guaranteed to be alive
> while we have a CrossThreadPersistent to SQLResultSet in m_resultSet.

I don't understand why |m_resultSet| needs to be a CrossThreadPersistent as it
is created on the main thread alongside SQLStatementBackend, for which a CTP is
kept (correct?) while the db thread executes the statement.

[sof, 2017-01-17 14:13:03]

sigbjornf@opera.com changed reviewers:
+ sigbjornf@opera.com

[sof, 2017-01-17 14:13:03]

https://codereview.chromium.org/2631403002/diff/20001/third_party/WebKit/Sour...
File third_party/WebKit/Source/modules/webdatabase/SQLStatementBackend.cpp
(right):

https://codereview.chromium.org/2631403002/diff/20001/third_party/WebKit/Sour...
third_party/WebKit/Source/modules/webdatabase/SQLStatementBackend.cpp:195:
CrossThreadPersistent<SQLResultSetRowList> rows = m_resultSet->rows();
This too I'm also not understanding -- if a thread is working over an object O
from another heap, then keeping O alive with a CTP while the thread uses it
should be sufficient to keep everything it refers to alive.

Having to create CTPs like this for stack references, shouldn't be needed.. and
looks tricky to maintain correctness for.

[sof, 2017-01-18 11:38:16]

Something noticed while going over https://crrev.com/1909813002 details -- is
the clearing of m_wrapper in SQLTransactionBackend::doCleanup() a "cross-heap"
write that's safe, strictly speaking?

( Would it be worth considering having a barriered cross-heap Member<>::assign()
operation? )