Add basic LAGraph expression evaluation #2

georgiy-belyanin · 2025-08-01T13:00:50Z

This patch introduces basic RPQ evaluation logic. It basically
interfaces the LAGraph (experimental) library that offers a collection
of the sparse linear algebra-based algorithms. It uses GraphBLAS [^1] as
a backend library for evaluating sparse matrix algebra. Rust <-> C
interfacing is done via FFI. Currently, the project requires our fork
[^2] since the required algorithm is not upstreamed yet. If you are
reading this and notice we occasionally have removed this from the fork
we are asking for an excuse and hope that you will be able to find the
suitable algorithm by yourself from the interfaces described on the Rust
side.

[^1] https://github.com/DrTimothyAldenDavis/GraphBLAS
[^2] https://github.com/SparseLinearAlgebra/LAGraph

Turns out there was an extra brace when parsing the graph edges due to wrong bounds. Let's fix it.

This patch extends plain nvals information within the label in the execution plan with label name. It is needed to match the actual matrices and labels to generate complete plan.

This patch introduces basic RPQ evaluation logic. It basically interfaces the LAGraph (experimental) library that offers a collection of the sparse linear algebra-based algorithms. It uses GraphBLAS [^1] as a backend library for evaluating sparse matrix algebra. Rust <-> C interfacing is done via FFI. Currently, the project requires our fork [^2] since the required algorithm is not upstreamed yet. If you are reading this and notice we occasionally have removed this from the fork we are asking for an excuse and hope that you will be able to find the suitable algorithm by yourself from the interfaces described on the Rust side. [^1] https://github.com/DrTimothyAldenDavis/GraphBLAS [^2] https://github.com/SparseLinearAlgebra/LAGraph

Copilot

Pull Request Overview

This PR introduces basic Regular Path Query (RPQ) evaluation using the LAGraph library with GraphBLAS as the backend. The implementation adds FFI bindings to interface with C libraries and includes query execution with performance benchmarking capabilities.

Key changes:

Integration of LAGraph/GraphBLAS via FFI for sparse matrix operations
Implementation of RPQ evaluation logic with operator translation to matrix operations
Addition of randomized plan selection and performance measurement infrastructure

Reviewed Changes

Copilot reviewed 11 out of 13 changed files in this pull request and generated 7 comments.

Show a summary per file

File	Description
vendor/LAGraph	Added LAGraph as a git submodule dependency
src/query.rs	Added Clone trait derivations to enable query cloning
src/plan.rs	Introduced LabelMeta struct and replaced simple cost function with random selection
src/main.rs	Added query execution loop with performance benchmarking and statistics collection
src/grb.rs	Added GraphBLAS Matrix FFI wrapper
src/graph.rs	Extended graph loading to include matrix files and vertex mappings, added error handling
src/eval.rs	Implemented RPQ evaluation via LAGraph FFI bindings
build.rs	Added linker configuration for LAGraph libraries
_typos.toml	Configured typo checker to exclude vendor directory
Cargo.toml	Added rand dependency
.gitmodules	Configured LAGraph submodule
.github/workflows/ci.yml	Added build steps for GraphBLAS and LAGraph dependencies

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2025-11-19T12:25:43Z

src/plan.rs

-        let cost_func = SillyCostFn;
+        let cost_func = CostFn;
        let extractor = Extractor::new(&runner.egraph, cost_func);
        extractor.find_best(runner.roots[0]).1.to_string()


Test uses undefined CostFn type. This should be RandomCostFn to match the actual implementation that replaced SillyCostFn.

Suggested change

extractor.find_best(runner.roots[0]).1.to_string()

let cost_func = RandomCostFn;

Copilot · 2025-11-19T12:25:43Z

src/graph.rs

            let mut splits = line.split_whitespace();
            let edge = splits.next()?;
-            let edge = edge[1..edge.len()].to_string();
+            let edge = edge[1..edge.len() - 1].to_string();


String slicing edge[1..edge.len() - 1] will panic if edge.len() is 0 or 1. Add length validation before slicing or use safer string manipulation methods.

Copilot · 2025-11-19T12:25:43Z

src/graph.rs

+        .lines()
+        .filter_map(|line| {
+            let mut splits = line.split_whitespace();
+            let vert = splits.next()?;


String slicing vert[1..vert.len() - 1] will panic if vert.len() is 0 or 1. Add length validation before slicing or use safer string manipulation methods.

Suggested change

let vert = splits.next()?;

let vert = splits.next()?;

if vert.len() < 2 {

return None;

}

Copilot · 2025-11-19T12:25:44Z

src/eval.rs

+                let mut mat: grb::Matrix = grb::Matrix(std::ptr::null_mut());
+                let mat = graph
+                    .mats
+                    .get(&meta.name)
+                    .or({
+                        graph.verts.get(&meta.name).map(|vert_idx| {


Variable mat is shadowed immediately after declaration. The mutable mat on line 103 is never used before being shadowed on line 104, making its initialization pointless. Remove the declaration on line 103 or rename one of the variables.

Suggested change

let mut mat: grb::Matrix = grb::Matrix(std::ptr::null_mut());

let mat = graph

.mats

.get(&meta.name)

.or({

graph.verts.get(&meta.name).map(|vert_idx| {

let mat = graph

.mats

.get(&meta.name)

.or({

graph.verts.get(&meta.name).map(|vert_idx| {

let mut mat: grb::Matrix = grb::Matrix(std::ptr::null_mut());

Copilot · 2025-11-19T12:25:44Z

src/main.rs

+                let runs: u32 = 1000;
+                let results: Vec<(RecExpr<Plan>, usize, Duration)> =
+                    run_random(&graph, runs, &expr).collect();
+                let first_n_runs = runs / 100;


Integer division runs / 100 will be 0 when runs < 100, causing no results to be printed in the 'First N runs' section. Consider using max(1, runs / 100) or a different calculation.

Suggested change

let first_n_runs = runs / 100;

let first_n_runs = std::cmp::max(1, runs / 100);

Copilot · 2025-11-19T12:25:44Z

src/eval.rs

+                    .or({
+                        graph.verts.get(&meta.name).map(|vert_idx| {
+                            unsafe {
+                                LAGraph_RPQMatrix_label(
+                                    &mut mat as *mut grb::Matrix,
+                                    *vert_idx,
+                                    graph.verts.len(),
+                                    graph.verts.len(),
+                                );
+                            }
+                            &mat
+                        })
+                    })


This closure references mat from the outer scope (line 103), but mat is shadowed on line 104. The reference &mat on line 117 would refer to the outer uninitialized matrix. This code would not compile as written, or has undefined behavior.

Copilot · 2025-11-19T12:25:45Z

src/eval.rs

+        LAGraph_RPQMatrix(&mut ans, plan as *mut RpqMatrixPlan, null_mut());
+        LAGraph_DestroyRpqMatrixPlan(plan);


The return value from LAGraph_RPQMatrix is ignored. Check the return code and propagate errors appropriately, similar to how LAGraph_MMRead is checked in graph.rs.

Suggested change

LAGraph_RPQMatrix(&mut ans, plan as *mut RpqMatrixPlan, null_mut());

LAGraph_DestroyRpqMatrixPlan(plan);

let status = LAGraph_RPQMatrix(&mut ans, plan as *mut RpqMatrixPlan, null_mut());

LAGraph_DestroyRpqMatrixPlan(plan);

if status != 0 {

return Err(format!("LAGraph_RPQMatrix failed with error code {}", status));

}

georgiy-belyanin added 2 commits August 1, 2025 15:59

Fix extra barce in prasing graph edges

792aefe

Turns out there was an extra brace when parsing the graph edges due to wrong bounds. Let's fix it.

Add meta information to labels in plan

0dca9e7

This patch extends plain nvals information within the label in the execution plan with label name. It is needed to match the actual matrices and labels to generate complete plan.

georgiy-belyanin force-pushed the basic-eval branch 9 times, most recently from 9c0dc19 to 2a17a79 Compare August 3, 2025 07:22

georgiy-belyanin force-pushed the basic-eval branch from 2a17a79 to b6644d3 Compare September 30, 2025 15:06

georgiy-belyanin force-pushed the basic-eval branch 2 times, most recently from a13a26a to 89dd251 Compare November 19, 2025 12:21

georgiy-belyanin marked this pull request as ready for review November 19, 2025 12:23

georgiy-belyanin requested a review from Copilot November 19, 2025 12:23

georgiy-belyanin force-pushed the basic-eval branch from 89dd251 to 4fb1fd0 Compare November 19, 2025 12:24

Copilot AI reviewed Nov 19, 2025

View reviewed changes

georgiy-belyanin merged commit 12f1f51 into main Nov 19, 2025
1 check passed

georgiy-belyanin deleted the basic-eval branch November 19, 2025 12:28

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Add basic LAGraph expression evaluation #2

Add basic LAGraph expression evaluation #2

Uh oh!

georgiy-belyanin commented Aug 1, 2025 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Nov 19, 2025

Uh oh!

Copilot AI Nov 19, 2025

Uh oh!

Copilot AI Nov 19, 2025

Uh oh!

Copilot AI Nov 19, 2025

Uh oh!

Copilot AI Nov 19, 2025

Uh oh!

Copilot AI Nov 19, 2025

Uh oh!

Copilot AI Nov 19, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

	extractor.find_best(runner.roots[0]).1.to_string()
	let cost_func = RandomCostFn;

	let first_n_runs = runs / 100;
	let first_n_runs = std::cmp::max(1, runs / 100);

		LAGraph_RPQMatrix(&mut ans, plan as *mut RpqMatrixPlan, null_mut());
		LAGraph_DestroyRpqMatrixPlan(plan);

Add basic LAGraph expression evaluation #2

Add basic LAGraph expression evaluation #2

Uh oh!

Conversation

georgiy-belyanin commented Aug 1, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull Request Overview

Reviewed Changes

Uh oh!

Copilot AI Nov 19, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Nov 19, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Nov 19, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Nov 19, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Nov 19, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Nov 19, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Nov 19, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

georgiy-belyanin commented Aug 1, 2025 •

edited

Loading