Skip to content

Dataflow: Flow-state changing steps should always be in path explanations#8474

Merged
MathiasVP merged 3 commits into
github:mainfrom
hvitved:flow-state-changing-steps-should-be-in-path-explanation-alternative
Mar 18, 2022
Merged

Dataflow: Flow-state changing steps should always be in path explanations#8474
MathiasVP merged 3 commits into
github:mainfrom
hvitved:flow-state-changing-steps-should-be-in-path-explanation-alternative

Conversation

@hvitved
Copy link
Copy Markdown
Contributor

@hvitved hvitved commented Mar 17, 2022

Alternative implementation to #8381.

@MathiasVP
Copy link
Copy Markdown
Contributor

MathiasVP commented Mar 17, 2022

Thanks for fixing this, Tom! This does indeed look much better 👍.

@MathiasVP MathiasVP mentioned this pull request Mar 17, 2022
Closed
@jketema
Copy link
Copy Markdown
Contributor

jketema commented Mar 17, 2022

Testing this out on the query I'm rewriting: localFlowStepPlus seems significantly slower now (compared to @MathiasVP's proposal).

@jketema
Copy link
Copy Markdown
Contributor

jketema commented Mar 17, 2022 

[2022-03-17 13:04:23] (135s) Starting to evaluate predicate DataFlowImpl::LocalFlowBigStep::localFlowStepPlus#fffffff/7@i7#e19c7wjr (iteration 7)
[2022-03-17 13:04:33] (144s) Tuple counts for DataFlowImpl::LocalFlowBigStep::localFlowStepPlus#fffffff/7@i7#e19c7wjr after 9.7s:
                      16646     ~5%      {7} r1 = SCAN DataFlowImpl::LocalFlowBigStep::localFlowStepPlus#fffffff#prev_delta OUTPUT In.1, In.5, In.0 'node1', In.2, In.3 'preservesValue', In.4 't', In.6 'cc'
                      313900036 ~0%      {8} r2 = JOIN r1 WITH project#DataFlowImpl::Stage2::revFlow#2#ffffff_120#join_rhs ON FIRST 2 OUTPUT Lhs.3, Rhs.2 'node2', Rhs.1 'config', Rhs.0, Lhs.2 'node1', Lhs.4 'preservesValue', Lhs.5 't', Lhs.6 'cc'
                      24073     ~14%     {7} r3 = JOIN r2 WITH DataFlowImpl::localFlowStepNodeCand1#fff ON FIRST 3 OUTPUT Lhs.1 'node2', Lhs.3 'state', Lhs.4 'node1', Lhs.5 'preservesValue', Lhs.6 't', Lhs.7 'cc', Lhs.2 'config'
                      
                      16646     ~19%     {7} r4 = SCAN DataFlowImpl::LocalFlowBigStep::localFlowStepPlus#fffffff#prev_delta OUTPUT In.2, In.1, In.1, In.5, false, In.0 'node1', In.6 'cc'
                      5202      ~4%      {7} r5 = JOIN r4 WITH DataFlowImpl::LocalFlowBigStep::additionalLocalFlowStepNodeCand2#fffff_01342#join_rhs ON FIRST 4 OUTPUT Rhs.4 'node2', false, Lhs.5 'node1', Lhs.1 'state', Lhs.6 'cc', Lhs.3 'config', Rhs.4 'node2'
                      5116      ~0%      {7} r6 = JOIN r5 WITH DataFlowImpl::NodeEx::getDataFlowType0#ff ON FIRST 1 OUTPUT Lhs.6 'node2', Lhs.3 'state', Lhs.2 'node1', false, Rhs.1 't', Lhs.4 'cc', Lhs.5 'config'
                      
                      29189     ~15%     {7} r7 = r3 UNION r6
                      19928     ~6%      {7} r8 = r7 AND NOT DataFlowImpl::LocalFlowBigStep::localFlowStepPlus#fffffff#prev(Lhs.2 'node1', Lhs.1 'state', Lhs.0 'node2', Lhs.3 'preservesValue', Lhs.4 't', Lhs.6 'config', Lhs.5 'cc')
                      19928     ~3%      {7} r9 = SCAN r8 OUTPUT In.2 'node1', In.1 'state', In.0 'node2', In.3 'preservesValue', In.4 't', In.6 'config', In.5 'cc'
                                         return r9

@hvitved
Copy link
Copy Markdown
Contributor Author

hvitved commented Mar 17, 2022

Testing this out on the query I'm rewriting: localFlowStepPlus seems significantly slower now (compared to @MathiasVP's proposal).

Thanks; I'll fix.

@hvitved hvitved force-pushed the flow-state-changing-steps-should-be-in-path-explanation-alternative branch from 76e8527 to abe3045 Compare March 17, 2022 13:04
@hvitved hvitved marked this pull request as ready for review March 18, 2022 08:18
@hvitved hvitved requested review from a team as code owners March 18, 2022 08:18
@hvitved
Copy link
Copy Markdown
Contributor Author

hvitved commented Mar 18, 2022

@MathiasVP : Does the C++ DCA run look OK to you? There are some (unrelated?) result differences, and a bit of slow-down on two project, but I wonder if that is just noise.

@hvitved hvitved added the no-change-note-required This PR does not need a change note label Mar 18, 2022
@jketema
Copy link
Copy Markdown
Contributor

jketema commented Mar 18, 2022

@hvitved Which query set did you use for the C++ DCA experiment?

@MathiasVP
Copy link
Copy Markdown
Contributor

MathiasVP commented Mar 18, 2022

@hvitved Which query set did you use for the C++ DCA experiment?

Looking at https://github.com/github/codeql-dca-main/blob/data/PR-8474-0-cpp__2/outputs/workflow/main-inputs.json it looks like lgtm-full.qls was used (which is not a query suite I've ever used for DCA).

All of the result changes are on abseil-cpp-windows. The result changes on the queries involving IR can probably be explained by variables-without-types issue. However, the cpp/constant-comparison, cpp/long-switch, and loop-variable-changed query changes don't use IR and thus cannot be explained by that existing issue. But they don't use dataflow, so they must be unrelated to this PR.

@hvitved
Copy link
Copy Markdown
Contributor Author

hvitved commented Mar 18, 2022

Looking at https://github.com/github/codeql-dca-main/blob/data/PR-8474-0-cpp__2/outputs/workflow/main-inputs.json it looks like lgtm-full.qls was used (which is not a query suite I've ever used for DCA).

Yes, that is what I used.

@MathiasVP
Copy link
Copy Markdown
Contributor

MathiasVP commented Mar 18, 2022

@MathiasVP : Does the C++ DCA run look OK to you? There are some (unrelated?) result differences, and a bit of slow-down on two project, but I wonder if that is just noise.

I think the slow-downs look fine. There are a bunch of slowdowns in queries that aren't related to dataflow at all, so I expect that it's all noise. So if it's just C++ that's showing a slow-down, then I think we're good.

@MathiasVP
Copy link
Copy Markdown
Contributor

MathiasVP commented Mar 18, 2022

Since Python has already approved the test changes in #8381 I'll just go ahead and merge.

@MathiasVP MathiasVP merged commit 8bf1729 into github:main Mar 18, 2022
@hvitved hvitved deleted the flow-state-changing-steps-should-be-in-path-explanation-alternative branch March 18, 2022 09:09
@aschackmull aschackmull mentioned this pull request Apr 7, 2022
Merged
aschackmull
aschackmull reviewed Apr 7, 2022
View reviewed changes
Comment thread cpp/ql/lib/semmle/code/cpp/dataflow/internal/DataFlowImpl.qll
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@aschackmull aschackmull aschackmull left review comments

@MathiasVP MathiasVP MathiasVP approved these changes

Assignees

@MathiasVP MathiasVP

Labels

C# C++ DataFlow Library Java no-change-note-required This PR does not need a change note Python Ruby

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

5 participants

@hvitved @MathiasVP @jketema @aschackmull @owen-mc