Migrating to mathlib-fp 2.0

This guide covers the published 2.0.0 release. It promotes the frozen 1.10.0 API; it is not a rewrite.

No-change paths

Supported 1.x source continues to compile unchanged. Existing units, compiled defaults, tolerances, numerical results, ownership rules, indexing, shape contracts, diagnostics, and the Lazarus package surface remain unchanged. No compatibility package or source edit is required for ordinary upgrades.

Preferred 2.0 paths

For new code, use the established typed/common paths documented during the 1.x runway: IDenseDoubleMatrix and TDenseDoubleMatrix for dense algebra, explicit typed sparse builders for sparse data, named options/results for optimisation, and TLocalRandom for reproducible local RNG state. These are documentation preferences, not new 2.0 declarations. They retain zero-based SizeInt indexing, explicit shape, owned/reference-counted results, and the existing diagnostic/status contracts.

Compatibility paths

IMatrix, TMatrixKit, and TMatrixKitSparse remain supported. Typed paths are preferred when their explicit copying, storage, scalar, and shape semantics fit the application; no implicit conversion is introduced. TPressureKit, EPressureError, TVelocityKit, and EVelocityError remain intentionally retained with their existing identity and import paths. The 21 reviewed plain aliases likewise remain supported; no warning, deprecation, removal, or package move begins in 2.0.

Actual breaking changes

There are none. The frozen decision decision-2.0.json records empty source, behaviour, warning, and packaging consequence lists. Therefore no source edit, ownership/copying change, indexing/shape change, diagnostic change, default/tolerance change, or numerical-result change is required to migrate supported 1.x code.

For detailed typed-dense guidance, see moving to typed dense matrices.