Data flow: No magic in returnFlowCallableCand

[MathiasVP]

Adding implementations of PostUpdateNode for IR field flow in C++ (#3118) revealed a bad join order in DataFlowImpl::parameterThroughFlowCand when running on Wireshark/Wireshark:

[2020-03-27 10:24:07] (4076s) Tuple counts for DataFlowImpl::parameterThroughFlowCand#ff#antijoin_rhs:
                      2981323763 ~0%     {3} r1 = JOIN DataFlowImpl::parameterThroughFlowCand#ff#shared AS L WITH DataFlowUtil::InstructionNode::getEnclosingCallable_dispred#ff_10#join_rhs AS R ON FIRST 1 OUTPUT L.<2>, L.<1>, R.<1>
                      2968445848 ~2%     {4} r2 = JOIN r1 WITH DataFlowImplCommon::Cached::TParamUpdate#2#ff_10#join_rhs AS R ON FIRST 1 OUTPUT r1.<2>, R.<1>, r1.<1>, r1.<0>
                      832264     ~0%     {3} r3 = JOIN r2 WITH project#DataFlowUtil::ParameterNode::isParameterOf_dispred#2#fff AS R ON FIRST 2 OUTPUT r2.<2>, r2.<3>, r2.<0>
                                         return r3

The fix is to avoid inlining returnFlowCallableCand. I couldn't reproduce the bad join order in C++ without the changes in #3118, but I figured it'd be good to test the impact of this change on master in isolation.

[MathiasVP]

Differences jobs are as expected:

• https://jenkins.internal.semmle.com/job/Changes/job/CPP-Differences/994/
• https://jenkins.internal.semmle.com/job/Changes/job/Java-Differences/669/
• https://jenkins.internal.semmle.com/job/Changes/job/CSharp-Differences/75/

[hvitved]

LGTM. Now that you are implementing field-flow for C++, I'd be interested to see which effect #3110 has. It has a positive performance impact on large Java and C# projects, but performance for C++ was mostly unchanged.

[MathiasVP]

LGTM. Now that you are implementing field-flow for C++, I'd be interested to see which effect #3110 has. It has a positive performance impact on large Java and C# projects, but performance for C++ was mostly unchanged.

Good point! I'll try merging #3110 into #3118 locally and see what happens.

[jbj]

LGTM. @aschackmull, do you have any objection to merging this?