Use this guide when writing new declarations anywhere in the tree. It records the forms review asks for. It does not authorize rewriting code the task is not otherwise changing; rule 5 in the repository-root AGENTS.md governs untouched lines. Eigen predates most of these forms, so a tree-wide count is not the convention: new code uses the current form, and a file being edited heavily should come out uniform rather than half converted.
static constexpr members, not enum blocks; enum constants are being phased out. Give each the type it is used as: Flags is unsigned int by convention, predicates are bool. In C++14 the in-class declaration is not a definition: odr-using such a member of a class template — binding it to a const T& parameter such as numext::mini's, or taking its address — links at -O2 and fails at -O0 unless a namespace-scope template <...> constexpr T Cls<...>::kName; definition exists (arch/Default/Half.h has the form). Pass a prvalue (+kName, Index(kName)) or add the definition; the test suite builds optimized and will not catch the omission.using to typedef, nullptr to NULL, = default and default member initializers to empty constructor bodies that assign each member. using binds in every tree, test/ and contrib/ included: those were left out of the sweep that converted Eigen/src, so the aliases surrounding new code there are mostly still typedef and matching the neighbours reproduces the form the sweep removed. Do not rely on CI to catch it — the modernize-use-using gap recorded at scripts/check_style.py leaves function-local typedefs unreported.kCamelCase is an accepted spelling for static constexpr and static constants, alongside the older snake_case and SCREAMING_CASE forms. It is not a review finding.numext:: math functions rather than std:: in library code, and Eigen‘s metaprogramming aliases (bool_constant, void_t, remove_all_t; see Eigen/src/Core/util/Meta.h) rather than spelling out the standard forms. internal::is_arithmetic is not a spelling of std::is_arithmetic: it is deliberately specialized for packet and Eigen scalar types and differs on long double during GPU compilation, so use it only when Eigen’s extended arithmetic category is specifically intended.inline; a bare inline there is noise. Use EIGEN_STRONG_INLINE or EIGEN_ALWAYS_INLINE when inlining matters, and nothing otherwise.EIGEN_STATIC_ASSERT_* helper that names them (_VECTOR_ONLY, _SAME_MATRIX_SIZE, ...) or EIGEN_STATIC_ASSERT(cond, TOKEN). These honor EIGEN_NO_STATIC_ASSERT and a user-provided EIGEN_STATIC_ASSERT override. Write unconditional implementation invariants as static_assert(cond, "what must hold"); bare assertions bypass those configuration mechanisms.EIGEN_DEPRECATED or EIGEN_DEPRECATED_WITH_REASON("use ..."), keep it working by forwarding to the replacement, and name the replacement in its Doxygen block.scratch, not scr) and name traits for the property they assert.Supported headers compile as C++14, which rules out forms that review suggestions often reach for:
if constexpr is C++17: use EIGEN_IF_CONSTEXPR(...) wherever the condition is compile-time constant. Note the condition must still be valid C++14 either way — the macro lowers to a plain if there./*name=*/ comments. Fields derived from other fields of the same object must be computed from locals first; the braced temporary cannot read fields it is about to set.std::span, CTAD, fold expressions, constinit, and later library additions are unavailable outside guarded backends with a documented newer requirement (the SYCL configurations force C++17, for example).Code added to a hot inner loop grows the enclosing function and can displace it from the instruction cache even when an EIGEN_PREDICT_FALSE guard keeps it from executing. Watch for that when adding a check or a fallback to such a loop; where a benchmark shows the cost, moving the cold path into an EIGEN_DONT_INLINE helper is one way to recover it.
The comment rules in the repository-root AGENTS.md are enforced in review and are the most repeated style finding here. Before publishing a diff, reread each added comment and delete the ones that narrate code or restate an identifier. Keep the ones recording mathematics, invariants, compatibility constraints, provenance, or the reason a slower or unusual form is deliberate — stated at the construct, not in the merge request.
Prefer the most precise notation that fits. A recurrence, an error bound, an invariant written as an expression, or two lines of pseudo-code usually carry more than a paragraph and are read faster by this audience:
// Bad: the relative error in summing n elements this way is bounded by roughly twice the // machine epsilon multiplied by the quantity log base two of n over B, plus B, where B is // the number of elements summed sequentially in each leaf of the tree. // Good: tree summation, relative error <= ~2*eps*(log2(n/B) + B) for leaf size B.
A bound, invariant, or identity stated exactly beats prose that half-carries it (m kept in [1, 2) says more than “balanced form”), and a named theorem comes with its statement. Reuse the symbols of the surrounding file and the cited reference rather than inventing one for a sentence; notation that restates what the code shows is the same defect as prose that does. Prose remains the tool for a reason: why this form and not the obvious one. Comments are plain text, so write expressions the way the rest of the tree does, not in a markup language that does not render.
Every new source file needs accurate REUSE metadata. Original Eigen code normally uses MPL-2.0; prefer the collective form when an agent cannot truthfully attribute an individual author:
// SPDX-FileCopyrightText: The Eigen Authors // SPDX-License-Identifier: MPL-2.0
Use the language's comment syntax. Documentation or assets that should not carry inline tags must be covered precisely in REUSE.toml; do not add a broad annotation that hides unrelated files. Compatible adapted material may require a different license expression and attribution, which must be preserved rather than relabeled as MPL-2.0.