Skip to content
This repository was archived by the owner on Aug 12, 2026. It is now read-only.
This repository was archived by the owner on Aug 12, 2026. It is now read-only.

Migrate production arithmetic to Boost.Multiprecision #149

Closed

Description

@acgetchell

Prerequisite

Do not begin this migration until microsoft/vcpkg#53269 has landed in a vcpkg
baseline that CDT++ can pin. The production configuration must not depend on a
repository-owned overlay of the cgal port.

Problem

#140 found that a Boost.Multiprecision-only arithmetic configuration preserves
the required exact predicates, 256-bit action distinctions, and extremely small
positive acceptance probabilities. Production still uses GMP/MPFR because the
current vcpkg cgal port unconditionally brings in those dependencies.

Goal

After upstream vcpkg support is available, migrate the production geometry,
action, and Metropolis arithmetic to supported Boost.Multiprecision backends and
remove CDT++'s production reliance on GMP and MPFR.

Scope

  • Pin a vcpkg baseline containing [cgal] Make GMP/MPFR an optional default feature microsoft/vcpkg#53269.
  • Request cgal without its GMP/MPFR default feature and add the direct
    Boost.Multiprecision dependency required by production.
  • Select CGAL's supported BOOST_BACKEND for exact predicates.
  • Replace production CGAL::Gmpfr, MPFR C API, and CGAL::Gmpzf use with the
    evaluated fixed 256-bit Boost type and backend-neutral value operations.
  • Preserve the MPFR-oracle fixtures during review, or replace them with
    versioned expected values generated by the existing harness before removing
    MPFR from production dependencies.
  • Remove the evaluation-only CGAL overlay and standalone package probe once the
    upstream production path covers their contract.
  • Document the fixed precision, rounding, exponent-range, and comparison
    contract in the public API.

Acceptance criteria

  • No repository-owned production vcpkg overlay is required.
  • GMP and MPFR are absent from the clean production dependency graph.
  • Near-degenerate orientation, co-spherical insertion, canonical reference
    fixtures, sub-double action distinctions, and sub-long double positive
    probabilities retain their documented behavior.
  • Clean-cache dependency and project builds pass on every supported platform,
    along with sequential, parallel, sanitizer, and CodeQL configurations.
  • The canonical reference comparison and representative Metropolis benchmark
    are rerun from a clean migration revision.
  • Any full-simulation regression materially larger than the approximately 2.6%
    model from Evaluate GMP/MPFR-free arithmetic without weakening exact predicates #140 is either resolved or explicitly accepted with rationale.
  • The final manifest, implementation, tests, and documentation no longer
    promise an MPFR-specific production type.

Related work

Activity

  1. acgetchell commented on Aug 12, 2026

    @acgetchell
    OwnerAuthor

    Closing as not planned for CDT++. The required upstream support tracked in microsoft/vcpkg#53269 has not materialized, and maintaining a repository-owned CGAL overlay is inappropriate for an archival release. The evaluated Boost.Multiprecision path also carries a substantial performance cost and less production evidence than the validated GMP/MPFR stack.

    Meanwhile, causal-triangulations already has its own arithmetic backend built on la-stack, so this migration would duplicate solved successor work without improving the handoff.

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

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions