Skip to content

Structure "G" refuses views 'L'/'U' (cuDSS accepts them) #84

Closed

Description

@claude

Found while working on: T19 (general LU)

What: _entry_filter in src/symbolic/pattern.jl (T05) raises InvalidValueError for structure "G" with a view other than 'F'. CUDSS.jl's cudss_solver() test creates CudssSolver(A, "G", 'L') / 'U' and runs analysis and factorization, so cuDSS accepts (presumably ignores) the view of a general matrix. The T19 port of that test (test/ported/cudss_solver.jl) asserts the InvalidValueError instead. A caller migrating from CUDSS.jl that passes 'L'/'U' with "G" gets an error rather than a factorization.

Why it is out of scope: T19 implements the LU numeric phase; the view rules of the analysis are T05's and a PLAN §1.1 decision ("view 'F' for G" is not stated there). No later task owns it; it does not block anything.

Suggested fix or plan change: decide whether "G" ignores the view (read the full matrix, as cuDSS apparently does) or keeps the error; document it in PLAN §1.1.

Activity

  1. added
    found-by-agentOpened by an implementing agent for an out-of-scope finding
    on Oct 5, 2026
  2. michel2323 commented on Oct 5, 2026

    @michel2323
    Member

    Triage (owner): fix it. The cuDSS documentation of cudssMatrixViewType_t says "If the accompanying matrix type is CUDSS_MTYPE_GENERAL, the matrix view is ignored." Since the API contract is cuDSS's, "G" accepts 'L'/'U'/'F' and always reads the full matrix.

    Plan, after #85 (T19) merges, as one out-of-chain PR (host only):

    • _entry_filter (src/symbolic/pattern.jl) returns the all-entries filter for "G" whatever the view;
    • the "G" cases of test/ported/cudss_solver.jl run analysis, factorization and solve for the three views and assert equal results instead of the error; a pattern/matrix test covers the three views for "G";
    • PLAN §1.1, views row: "for G the view is ignored, as in cuDSS" (owner edit).

    Not fixed inside T19 (out of its scope). Closes with that PR.


    Generated by Claude Code

  3. added
    triagedOwner has looked at this found-by-agent issue; it no longer holds the task chain
    on Oct 5, 2026
  4. github-actions commented on Oct 7, 2026

    @github-actions
    Contributor

    Merged in #101.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    found-by-agentOpened by an implementing agent for an out-of-scope findingtriagedOwner has looked at this found-by-agent issue; it no longer holds the task chain

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions