Activity

1. github-actions

   added

   Office HoursIssue announcing office hours and soliciting associated agenda items.Issue announcing office hours and soliciting associated agenda items.

   on Jul 7, 2026
2. DivitJain26 commented on Jul 14, 2026

   

   DivitJain26

   on Jul 14, 2026

   Last edited by DivitJain26

   Member

   More actions

   • feat: add array/uint64 stdlib#12562 is ready for review.
   • I was looking at custom array methods, and I was wondering why not provide a default ordering and only use a custom compareFcn when the user supplies one?
3. prajjwalbajpai commented on Jul 14, 2026

   

   prajjwalbajpai

   on Jul 14, 2026

   Member

   More actions

   • I have a doubt regarding tests, we use ULP difference to see whether tests are passing or not. But, near-zero values like 2.4e-17 and -4.6e-19 are both essentially zero and they differ by ~9e18 ULPs due to opposite signs. This is causing my tests to fail. Is there any other way to write tests?
   • For the dorgqr routine, I have tried to implement the mathematical verification test. Does this look fine?

   tape( 'the function generates an `M-by-N` real matrix `Q` with orthonormal columns (multiple blocks, mathematical verification)', function test( t ) {
   	var WORK;
   	var info;
   	var TAU;
   	var LDA;
   	var QT;
   	var A;
   	var C;
   	var M;
   	var N;
   	var K;
   	var i;
   	var I;
   	var Q;
   
   	M = 162;
   	N = 162;
   	K = 161;
   	LDA = M;
   
   	A = new Float64Array( M * N );
   	TAU = new Float64Array( K );
   	WORK = new Float64Array( 2 * N );
   
   	for ( i = 0; i < N; i++ ) {
   		A[ i + ( i * LDA ) ] = 1.0;
   	}
   	for ( i = 0; i < K; i++ ) {
   		TAU[ i ] = 2.0;
   	}
   
   	info = dorgqr( 'column-major', M, N, K, A, LDA, TAU, WORK, 2 * N );
   
   	Q = ctor( 'float64', A, [ M, N ], [ 1, LDA ], 0, 'column-major' ); // Create ndarray view of Q
   	QT = transpose( Q ); // Compute Q^T
   	C = new Float64Array( N * N ); // Compute C = Q^T * Q
   	dgemm( 'no-transpose', 'no-transpose', N, N, M, 1.0, QT.data, QT.strides[ 0 ], QT.strides[ 1 ], QT.offset, Q.data, Q.strides[ 0 ], Q.strides[ 1 ], Q.offset, 0.0, C, 1, N, 0 ); // Compute C = Q^T * Q
   
   	I = new Float64Array( N * N ); // Create identity matrix
   	for ( i = 0; i < N; i++ ) {
   		I[ i + ( i * N ) ] = 1.0;
   	}
   
   	t.strictEqual( info, 0, 'returns expected value' );
   	t.strictEqual( isAlmostEqual( C, I, 1 ), true, 'Q^T * Q equals identity matrix' );
   	t.end();
   });
4. 0PrashantYadav0 commented on Jul 14, 2026

   

   0PrashantYadav0

   on Jul 14, 2026

   Last edited by 0PrashantYadav0

   Member

   More actions

   Discussion on this issues: stdlib-js/stdlib#13468
5. batpigandme commented on Jul 14, 2026

   

   batpigandme

   on Jul 14, 2026

   Last edited by batpigandme

   Member

   More actions

   AI-Generated Summary

   Caution

   These notes are likely imperfect, especially when it comes to spelling/function/package names

   Meeting Purpose

   Resolve technical questions and align on best practices for development.

   Key Takeaways

   • array-uint64 Sorting: Implement a default sort by comparing high then low words. Wrap user-provided comparators to ensure they receive uint64 instances, not raw buffer values.
   • Test Failures Near Zero: For values near zero with sign differences, use an absolute tolerance (abs(value) < eps) instead of ULP comparisons, which fail due to the large ULP distance across zero.
   • Test Fixture Optimization: To reduce large fixture sizes, Prajjwal will test a strategy of storing only column-major data and dynamically converting it to row-major for testing.
   • README Best Practices: Prioritize user clarity by hiding non-essential info (e.g., ESLint comments) in README examples. Use existing high-level APIs (e.g., random for real numbers) to simplify code.

   Topics

   array-uint64 Custom Sort

   • Problem: The array-uint64 sort implementation lacks a default comparator.
   • Rationale: Unlike complex numbers, uint64 values have a total order, making a default sort necessary for consistency with other array types (e.g., array-bool).
   • Solution:
     • Default Comparator: Define a function to sort by comparing high words, then low words.
     • User Comparator: Wrap user-provided functions to ensure they receive uint64 instances, not raw buffer values.
     • Scope: The default comparator can be defined in the parent scope; the user wrapper must be defined within the sort method.

   Test Failures Near Zero

   • Problem: ULP comparisons fail for values near zero with opposite signs (e.g., +1e-308 vs. -1e-308), which have a large ULP distance but are mathematically equivalent.
   • Solution:
     • Absolute Tolerance: For values expected to be near zero, use an absolute tolerance check: abs(value) < eps. This handles sign differences correctly.
     • ULP Comparison: Continue using ULP comparisons for all other values where signs are expected to match.

   Optimizing Mathematical Verification Tests

   • Context: Prajjwal is writing mathematical verification tests (e.g., Q * Q^T = I) to supplement Fortran reference tests.
   • Optimizations:
     • Transpose: Use DGETrans instead of manual buffer manipulation.
     • Identity Matrix: Use DLAset with alpha=0, beta=1 to create an identity matrix.
   • Fixture Optimization Idea:
     • Proposal: Store only column-major data in JSON fixtures.
     • Method: Dynamically convert column-major data to row-major for testing using DGETrans.
     • Goal: Reduce large fixture sizes and simplify maintenance.

   README & Example Best Practices

   • Problem: recent PRs used complex number random-generation logic for real-valued arrays and included ESLint comments in README examples.
   • Rationale for Corrections:
     • API Usage: Use existing high-level APIs (e.g., random for real numbers) to simplify code. The complex number logic is a temporary workaround for a missing API and should not be generalized.
     • README Clarity: Hide non-essential info (e.g., ESLint comments) from users. Use HTML comments (<!-- eslint-disable -->) to hide linter directives in markdown, ensuring examples are clean and focused.

   Full transcript

   Full office hours transcript
6. batpigandme

   closed this as completedon Jul 21, 2026